Introduction
The rapid expansion of electronic commerce over the preceding two decades has fundamentally transformed the competitive landscape within which commercial enterprises operate, rendering digital advertising not merely a supplementary channel but a primary determinant of market visibility and revenue generation. Within this environment, paid search advertising — and the Google Ads platform in particular — has emerged as one of the most consequential tools available to e-commerce practitioners, on account of its capacity to position commercial offers before prospective buyers at the precise moment of expressed purchase intent [15]. The economic significance of this channel is reflected in the scale of advertiser investment: global expenditure on search advertising surpassed two hundred and fifty billion United States dollars in 2023, with Google commanding in excess of ninety percent of the paid search market by revenue in most major commercial territories. Despite this manifest commercial importance, a rigorous academic understanding of the conditions that determine Google Ads campaign effectiveness in e-commerce contexts remains insufficiently developed, fragmented across disciplinary boundaries, and often subordinated to practitioner-oriented prescriptive frameworks that offer limited explanatory power.
The academic literature on digital advertising effectiveness has grown substantially since the early studies of banner advertising and click-through rates in the late 1990s, yet significant theoretical and empirical lacunae persist. The preponderance of existing quantitative research has employed large-scale platform datasets or controlled experimental designs that, whilst affording statistical power, abstract away the organisational, strategic, and contextual factors that mediate the relationship between advertising expenditure and commercial outcomes in practice. Studies conducted by Rutz and Bucklin established the foundational importance of keyword-level bidding dynamics and consumer search behaviour in determining paid search performance, yet these investigations were conducted within single-platform laboratory conditions that limit their ecological validity [16]. More recently, the integration of machine learning-driven bidding algorithms and audience-signal targeting within the Google Ads ecosystem has substantially altered the operational parameters of campaign management, rendering prior empirical findings of uncertain applicability to contemporary practice. The consequence is a knowledge gap between the theoretical frameworks developed in academic research and the practical realities confronted by e-commerce advertisers operating within an algorithmically mediated advertising environment.
The present thesis is motivated by this gap and proceeds from the premise that an adequate understanding of Google Ads campaign effectiveness requires methodological approaches capable of preserving and examining contextual complexity rather than controlling it out of the analytical frame. The case study methodology, as codified by Yin and subsequently elaborated within the domain of marketing research, provides a suitable epistemological instrument for this purpose, permitting the integration of multiple data sources and the examination of within-case causal dynamics that aggregate statistical analyses are structurally incapable of revealing [17]. By situating the analysis of campaign performance within the broader strategic, organisational, and competitive contexts in which individual e-commerce operators function, it becomes possible to develop explanatory accounts that move beyond the identification of average effects toward an understanding of the mechanisms through which advertising investment is converted — or fails to be converted — into commercial value.
The primary research objective of this thesis is to examine the conditions under which Google Ads campaigns generate measurable and commercially significant returns for e-commerce enterprises operating across distinct product categories. This overarching objective is pursued through three subsidiary questions that structure the empirical and analytical components of the investigation. First, it is asked which campaign configuration parameters — encompassing bidding strategy, keyword architecture, audience targeting, and budget allocation — are most consistently associated with superior commercial performance across the cases examined. Second, the research inquires into the extent to which sector-specific characteristics, including product category, average transaction value, and purchase cycle duration, moderate the relationship between campaign configuration and performance outcomes. Third, the thesis addresses the question of whether and how the progressive integration of automated machine learning features within the Google Ads platform affects the strategic latitude available to campaign managers and the consequent distribution of campaign performance outcomes. These questions collectively orient the investigation toward an explanatory rather than a merely descriptive understanding of the phenomenon under study.
The empirical basis of the investigation consists of a multiple case study conducted across three e-commerce organisations operating within the United Kingdom market, each representing a distinct product category: fashion retail, consumer electronics, and home goods. The selection of cases was governed by the principles of theoretical rather than statistical sampling, with each case chosen on the basis of its capacity to contribute distinct and analytically productive variation to the cross-case comparison rather than on the basis of representativeness within a defined population. Data were gathered through semi-structured interviews with campaign managers and senior marketing personnel, supplemented by systematic analysis of campaign performance records spanning a twelve-month study period. This mixed-evidence design, recommended by Eisenhardt for multi-case investigations seeking to develop mid-range theoretical propositions, permits the triangulation of quantitative performance indicators against qualitative accounts of strategic intent and managerial interpretation [18].
The thesis is structured across three principal chapters, each addressing a distinct but interconnected dimension of the research problem. Chapter 1 establishes the theoretical foundations of the investigation, tracing the evolution of digital advertising economics from the early impression-based model through the pay-per-click paradigm to the contemporary intent-signal and audience-data infrastructure upon which the Google Ads platform operates. This chapter further examines the principal analytical frameworks employed in the academic assessment of advertising effectiveness, with particular attention to attribution modelling, return on advertising spend as a performance construct, and the theoretical basis for the Quality Score mechanism. Chapter 2 presents and justifies the research methodology, elaborating the epistemological commitments of the qualitative case study approach, the criteria governing case selection, the data collection instruments deployed, and the analytical procedures applied to the evidence assembled. Chapter 3 presents the empirical findings of the investigation, proceeding from a comparative analysis of campaign performance metrics across the three cases through an examination of the factors identified by practitioners as determinants of campaign success or underperformance, and culminating in a cross-case synthesis that identifies the propositions warranted by the evidence. The Conclusion draws together the principal findings, reflects upon their theoretical and practical implications, acknowledges the limitations of the investigation, and identifies directions for further research that the study's findings suggest as productive.
The contribution of the present thesis to the existing literature is understood to operate on two levels. At the empirical level, the investigation provides detailed and contextually grounded accounts of Google Ads campaign management practice in three distinct e-commerce settings, generating evidence that supplements the aggregate and experimental findings that predominate in the existing literature. At the theoretical level, the cross-case analysis develops propositions regarding the conditions of campaign effectiveness that are anchored in empirical observation yet articulated at a level of abstraction sufficient to inform both scholarly inquiry and managerial practice. It is the contention of this thesis that such contextually sensitive, theoretically informed empirical work is not merely a complement to large-scale quantitative research but a methodologically distinct and independently valuable mode of knowledge production, capable of illuminating the mechanisms and boundary conditions that statistical analyses of advertising performance necessarily leave unexamined.
Chapter 1: Theoretical Foundations of Digital Advertising and Google Ads in E-commerce
1.1. The Evolution of Digital Advertising in the E-commerce Sector
The history of digital advertising is inseparable from the parallel development of electronic commerce, as both phenomena emerged from the commercialisation of the World Wide Web in the mid-1990s and have since evolved in a mutually reinforcing relationship. The first widely recognised banner advertisement — placed by AT&T on HotWired.com in October 1994 — attracted a click-through rate of approximately 44 percent, a figure that reflected the novelty of the medium rather than any intrinsic efficiency, and which has since declined to fractions of a single percent across most display advertising contexts [6]. This initial phase of digital advertising was characterised by a cost-per-thousand-impressions (CPM) pricing model borrowed from print and broadcast media, offering advertisers broad reach but limited accountability and no direct connection between expenditure and commercial outcomes. The inadequacy of this model for performance-oriented advertisers became increasingly evident as click-through rates declined and the internet population diversified beyond early adopters.
The introduction of the pay-per-click paradigm represented a structural transformation in the economics of digital advertising. Overture Services (formerly GoTo.com) launched the first commercially successful PPC search advertising platform in 1998, enabling advertisers to bid for placement against specific keywords and pay only when users clicked on their advertisements [4]. This model aligned advertiser costs directly with user intent, as expressed through search queries, and created a measurable connection between advertising expenditure and website traffic that CPM-based display advertising could not provide. Google launched its AdWords platform in October 2000, initially offering a CPM model before transitioning to a hybrid auction system incorporating relevance-based Quality Scores alongside maximum bids — a design choice that distinguished it from pure price-based auctions and established the template for modern paid search advertising [19].
Successive waves of innovation throughout the 2000s and 2010s progressively expanded the targeting capabilities and measurement sophistication available to digital advertisers. The introduction of contextual advertising networks — most notably Google AdSense in 2003 — extended the reach of pay-per-click advertising beyond search results pages to publisher websites, creating the infrastructure for the display advertising ecosystem that operates today. Remarketing technologies, which enabled advertisers to serve targeted advertisements to users who had previously visited their websites, were introduced by Google in 2010 and rapidly became foundational to e-commerce advertising strategy, particularly for recovering abandoned shopping carts and re-engaging lapsed customers [5]. The mass adoption of smartphones from approximately 2007 onward compelled a fundamental restructuring of campaign architecture, as advertisers were required to develop mobile-specific creative assets, bid adjustments, and landing page experiences to accommodate the growing share of e-commerce traffic originating from mobile devices.
The development of programmatic buying — automated real-time bidding systems that execute advertising inventory transactions in milliseconds through ad exchanges — represented the most structurally significant transformation in the digital advertising ecosystem since the invention of PPC. Programmatic platforms enabled advertisers to purchase impressions based on granular audience data rather than the contextual characteristics of the publisher, shifting the locus of value from content adjacency to audience quality [6]. This transition introduced significant challenges, including increased supply chain complexity, reduced transparency regarding advertisement placement, and heightened vulnerability to impression fraud and brand safety violations. The ongoing deprecation of third-party cookies — initiated by Apple's Intelligent Tracking Prevention and extended by Google's repeated reconsiderations of its Chrome deprecation timeline — has created a structural measurement challenge that accelerated investment in first-party data infrastructure among major e-commerce operators.
Google Ads consolidated a dominant position within the digital advertising ecosystem, commanding the largest share of global digital advertising revenue among platform providers and constituting the primary paid media channel for the majority of e-commerce advertisers. The Internet Advertising Bureau reported that paid search advertising accounted for approximately 46.3 percent of US internet advertising revenues as early as 2012, with Google Inc. registering $46 billion in global revenues that year, of which $43.7 billion — or 95 percent — were attributed to advertising [2]. This dominance reflects both the platform's technical capabilities and the structural advantage conferred by its near-monopoly position in the search market, which ensures that Google's search network reaches the vast majority of consumers at the moment of expressed commercial intent. The contemporary digital advertising landscape, while more fragmented than in the early AdWords era due to the rise of social media platforms, remains organised around Google's ecosystem as its primary axis, justifying the selection of Google Ads as the analytical focus of this thesis.
1.2. Architecture and Core Mechanisms of Google Ads
The Google Ads platform is organised according to a hierarchical account structure that governs the allocation of budgets, the application of targeting parameters, and the management of creative assets at each level of operational granularity. At the highest level, the account encompasses all campaigns operated by a single advertiser and is associated with billing information and access credentials. Campaigns represent the primary unit of budget management, with each campaign assigned a daily budget, a geographic targeting scope, a device targeting configuration, and a campaign type that determines the available advertising formats and bidding options. Below the campaign level, ad groups organise sets of closely related keywords and their associated advertisement creatives, enabling more precise relevance management between the search queries that trigger advertisements and the messages those advertisements communicate [4].
The typology of campaign formats available within Google Ads reflects the platform's evolution from a keyword-triggered text advertising service into a multi-surface advertising ecosystem spanning both intent-based and audience-based inventory. Search campaigns display text advertisements in response to specific user queries and remain the most direct expression of the PPC model, as advertisement display is contingent upon the user's expressed intent as encoded in their search query. Shopping campaigns — and their successor format, Performance Max — surface product listing advertisements featuring product image, price, and retailer name directly within the Search Engine Results Page (SERP), enabling e-commerce advertisers to present product-level information at the moment of commercial intent [15]. Display campaigns serve visual advertisements across the Google Display Network, which encompasses over two million publisher websites and applications, while Video and Demand Gen campaigns extend the platform's reach into higher-funnel brand awareness contexts complementing the conversion focus of Search and Shopping.
The bidding mechanism through which Google Ads determines advertisement placement and cost constitutes the platform's most technically consequential architectural component. The advertising auction operates through a second-price mechanism in which advertisers compete for each available impression, with the winning advertiser paying a price determined by the Ad Rank of the next-highest competitor rather than their own maximum bid. Ad Rank is calculated as a composite function of the advertiser's maximum bid, Quality Score, the expected impact of advertisement extensions, and contextual signals at the time of the auction. The incorporation of Quality Score into Ad Rank — rather than relying on maximum bid alone — creates a mechanism through which advertisements of superior relevance and landing page quality can achieve higher positions at lower effective costs than competitors with higher maximum bids but lower quality signals, directly incentivising investment in creative and landing page quality as well as bid management [1].
Smart Bidding strategies represent the contemporary frontier of the Google Ads bidding architecture, employing machine learning models trained on vast datasets of query, user, contextual, and conversion signals to optimise bids in real time for each individual auction. Target ROAS bidding instructs the algorithm to maximise conversion value while achieving a specified return on ad spend ratio, while Target CPA bidding optimises for conversion volume at a specified cost per acquisition. The theoretical advantage of Smart Bidding lies in its capacity to incorporate signals at a granularity and volume that exceeds human manual management capability — including device, operating system, browser, physical location, time of day, and the specific semantic characteristics of the triggering query [3]. Empirical evidence on the performance differential between Smart Bidding and expert manual management remains contested, with relative automation advantages depending heavily on conversion data volume, measurement quality, and the precise alignment between campaign objectives and algorithmic optimisation targets.
1.3. Key Performance Indicators and Measurement Frameworks for Paid Search Campaigns
The evaluation of Google Ads campaign effectiveness requires a structured approach to performance measurement that distinguishes between metrics reflecting advertising activity, metrics indicating operational efficiency, and metrics directly expressing commercial outcomes. This tripartite distinction is analytically important because activity metrics such as impression volume can increase without corresponding improvements in profitability, and because efficiency metrics such as click-through rate can be optimised in ways that deteriorate conversion performance. A comprehensive measurement framework integrates all three categories and interprets each metric in context, acknowledging that the relationship between upstream and downstream performance indicators is mediated by factors external to the advertising platform, including website experience, product attractiveness, pricing competitiveness, and fulfilment reliability [4]. Over-reliance on any single indicator risks fundamentally mischaracterising campaign performance and misdirecting budget allocation decisions.
Click-through rate (CTR), defined as the ratio of clicks to impressions, functions as a primary indicator of creative relevance and audience-keyword alignment, but varies substantially across campaign types, keyword match types, SERP positions, and device categories, rendering cross-campaign comparisons without contextual controls analytically misleading. Conversion rate (CVR) — the proportion of clicks resulting in a defined conversion action, typically a completed purchase transaction in e-commerce contexts — constitutes the critical bridge between traffic generation and revenue production. Cost per acquisition (CPA) synthesises click cost and conversion rate into a single efficiency indicator expressing the advertising expenditure required to generate one conversion, enabling direct comparison against the margin contribution of acquired customers [5]. Return on ad spend (ROAS) has emerged as the dominant campaign-level efficiency metric for e-commerce advertisers, directly expressing the revenue multiplier achieved on each unit of advertising investment, though it must be distinguished from return on investment proper, as ROAS calculations typically exclude cost of goods sold and operational overheads and can therefore overstate campaign profitability [1].
| Metric | Formula | Category | Primary Analytical Use |
|---|---|---|---|
| Click-Through Rate (CTR) | Clicks ÷ Impressions × 100% | Activity / Efficiency | Creative relevance and keyword alignment |
| Conversion Rate (CVR) | Conversions ÷ Clicks × 100% | Efficiency | Landing page and offer quality assessment |
| Cost Per Acquisition (CPA) | Ad Spend ÷ Conversions | Efficiency | Customer acquisition cost benchmarking |
| Return on Ad Spend (ROAS) | Revenue ÷ Ad Spend × 100% | Outcome | Revenue efficiency and budget justification |
| Impression Share | Impressions ÷ Eligible Impressions × 100% | Activity | Competitive visibility and budget adequacy |
| Quality Score | Composite 1–10 (CTR, relevance, landing page) | Efficiency | Ad Rank optimisation and CPC reduction |
Attribution modelling represents one of the most consequential methodological choices in paid search measurement, as the model selected determines which touchpoints receive credit for conversions and thereby shapes budget allocation decisions across campaigns, ad groups, and keywords. The major attribution models available within Google Ads and Google Analytics 4 include last-click attribution, which assigns full credit to the final touchpoint before conversion; linear attribution, which distributes credit equally across all recorded touchpoints; time-decay attribution, which allocates progressively greater credit to touchpoints closer to conversion; and data-driven attribution, which employs statistical models trained on the advertiser's own conversion data to estimate the marginal contribution of each touchpoint [3]. The selection of data-driven attribution is generally recommended for accounts with sufficient conversion volume to train the underlying model, as it is less susceptible to the systematic biases inherent in rule-based models. The transition from Universal Analytics to Google Analytics 4, completed by Google in July 2023, introduced a fully event-based measurement architecture that altered the definition and counting methodology of several foundational metrics, representing a significant discontinuity in historical performance data for e-commerce advertisers.
Incrementality testing — controlled experiments designed to isolate the causal effect of advertising expenditure by comparing outcomes in exposed and unexposed groups — represents a methodologically superior complement to attribution-based measurement that is increasingly adopted by sophisticated e-commerce advertisers. Geographic holdout experiments, in which paid search advertising is suspended in randomly selected geographic markets while continuing in control markets, enable the direct measurement of incremental conversions generated by the advertising investment, yielding estimates of true ROI that attribution models systematically overstate [2]. The operational demands of incrementality testing — including the requirement for sufficient traffic volume to achieve statistical significance and the temporary revenue cost of withholding advertising from holdout markets — restrict its adoption to larger e-commerce operators, establishing an important methodological asymmetry between the empirical standards achievable by major retailers and those available to the small and medium-sized enterprises that constitute the majority of Google Ads advertisers.
1.4. The Role of Audience Targeting and Personalisation in E-commerce Advertising
Audience targeting capabilities constitute one of the primary sources of competitive differentiation among digital advertising platforms and represent a significant theoretical advance over the demographic targeting available in traditional media channels. Google's targeting infrastructure draws on behavioural signals derived from search history, browsing activity across the Google Display Network, YouTube viewing patterns, and physical location data collected through mobile devices, enabling the construction of audience segments that reflect actual consumer behaviour rather than inferred demographic proxies [6]. The theoretical foundation for audience targeting effectiveness lies in the concept of advertising relevance: advertisements served to users whose expressed or inferred interests align with the advertised product or service are more likely to attract engagement and produce conversion outcomes than untargeted exposures. This proposition is supported by established models of persuasion, which predict that personally relevant messages receive more elaborated cognitive processing, producing stronger purchase intentions and more durable attitude change than low-relevance exposures [20].
In-market audiences represent the most strategically valuable targeting segment for e-commerce advertisers pursuing conversion-stage objectives, as they are composed of users whom Google's algorithms have identified as actively researching products or services in specific categories based on recent query patterns and content consumption behaviour. The practical utility of in-market audience targeting lies in its capacity to concentrate advertising expenditure on users with elevated purchase probability, thereby improving conversion rates and reducing CPA relative to broad keyword targeting without audience overlays. Remarketing audiences target users who have previously interacted with the advertiser's website or application, enabling highly personalised messaging that references prior browsing or purchasing behaviour [5]. Dynamic remarketing, which automatically generates personalised advertisements featuring the specific products a user previously viewed, represents the most technically sophisticated manifestation of this capability and has been shown to produce substantially higher conversion rates than static remarketing in e-commerce contexts where product catalogues are large and diverse.
The deployment of audience targeting capabilities in European markets is subject to regulatory constraints that materially affect the operational feasibility of certain targeting strategies. The General Data Protection Regulation (GDPR), in force since May 2018, requires that the collection and processing of personal data for advertising targeting purposes be based on either explicit user consent or a legitimate interest assessment, with consent requirements applying to the placement of remarketing cookies and the activation of personalised advertising features [6]. The ePrivacy Directive imposes additional consent requirements for cookies and similar tracking technologies, and the practical implementation of these requirements through consent management platforms has resulted in consent rates that vary substantially across markets and publisher categories. For e-commerce advertisers with significant EU customer bases, reduced consent rates translate directly into smaller addressable remarketing audiences and diminished in-market segment coverage, requiring structural adjustments to campaign architecture and bidding strategy.
The strategic response of e-commerce advertisers to declining third-party tracking coverage has centred on the activation of first-party data assets through Customer Match and Enhanced Conversions capabilities within Google Ads. Customer Match enables advertisers to upload hashed first-party customer data — including email addresses and phone numbers — which Google matches against its own user database to enable targeting of known customers across Google-owned properties. This capability supports sophisticated customer lifecycle marketing strategies, including retention campaigns targeting high-value existing customers, win-back campaigns for lapsed purchasers, and suppression of current customers from prospecting campaigns to reduce redundant expenditure [3]. The effectiveness of first-party data activation is contingent on the quality, completeness, and freshness of the underlying customer database, making CRM data governance a prerequisite for competitive advantage in an environment of progressively diminishing third-party tracking signal availability.
1.5. Literature Review: Empirical Research on Paid Search Effectiveness
The empirical literature on paid search advertising effectiveness is characterised by a fundamental tension between the attribution-based performance estimates reported by advertisers and advertising platforms — which typically indicate strongly positive returns — and the findings of controlled experimental studies, which consistently reveal that attribution-based estimates substantially overstate the causal contribution of paid search to revenue generation. This divergence has significant implications for budget allocation decisions and for the interpretation of performance data throughout the present thesis. The most influential experimental contribution to this literature is the series of large-scale field experiments conducted by Blake, Nosko, and Tadelis at eBay, which demonstrated through geographic randomisation that brand keyword advertising produced no measurable short-term incremental sales for a well-known brand, as virtually all paid search traffic was substituted by organic search traffic when paid advertisements were suspended [2]. This finding is particularly significant because brand keyword advertising constitutes a substantial share of paid search expenditure for established e-commerce retailers and generates the highest attribution-based ROAS figures.
The Bazaar.com case study documented by Jerath provides a complementary analytical framework for understanding the endogeneity problem in paid search measurement. In that case, a technical interruption that suspended sponsored search advertising for branded keywords on Google — while leaving an equivalent Bing campaign unaffected — created a natural experiment enabling the separation of incremental from substitutive traffic effects [1]. The observation that organic traffic increased substantially during the period of paid search suspension, with total combined traffic remaining approximately constant, provided empirical support for the hypothesis that branded keyword advertising frequently redirects users who would have reached the advertiser's website via organic results regardless of the paid advertisement's presence. This natural experiment methodology — exploiting exogenous variation in advertising exposure rather than relying on correlational platform data — represents the methodological gold standard for incrementality estimation, and its findings caution strongly against naive attribution-based ROI calculations of the type initially performed in the case.
The literature on non-branded keyword advertising presents a more nuanced picture of effectiveness, with the magnitude of incremental returns varying substantially as a function of brand awareness, competitive intensity, organic visibility, and consumer segment characteristics. Blake, Nosko, and Tadelis found that non-branded keywords generated statistically significant positive effects on purchases by new and infrequent users — consistent with the informative theory of advertising — but that these users accounted for a sufficiently small share of total attributed sales that average returns across the full user population were negative for a high-awareness platform such as eBay [2]. This segmentation finding supports the broader principle that SEM is most effective when directed at consumers who lack prior awareness of the advertiser, and implies that the returns to paid search investment are systematically higher for smaller or less well-known brands than for established market leaders. A systematic literature review of PPC and paid search cost-effectiveness published by Maluleke found that PPC tends to be most cost-effective in high-margin sectors characterised by limited organic visibility, while established brands operating in competitive markets may find that SEO generates superior long-term ROI [3].
Research comparing the strategic and financial performance of PPC advertising against search engine optimisation has produced consistent findings regarding the distinct temporal profiles of the two approaches. Dehlin and Björnfot's comparative analysis of five Swedish e-commerce companies demonstrated that SEO investment reaches a break-even point relative to continuous PPC expenditure between 15 and 30 months from campaign initiation, contingent on the speed of SEO implementation and the magnitude of achievable organic traffic gains [4]. Singh and Kesarwani's review of the broader SEO-PPC literature confirmed that PPC delivers immediate but transient visibility — with traffic ceasing upon budget exhaustion — while SEO generates compounding organic traffic that persists beyond the active investment period, though requiring sustained content and technical maintenance to preserve rankings [5]. These findings have informed an increasingly prevalent practitioner consensus favouring integrated approaches that deploy PPC for immediate revenue generation and market intelligence while investing in SEO for long-term traffic cost reduction, a strategic logic that is consistent with the portfolio diversification principles of modern digital marketing management.
The methodological landscape of paid search research is characterised by significant limitations that constrain the generalisability of individual findings and the comparability of results across studies. The majority of empirical research relies on proprietary data sourced from single platforms or advertising agencies, introducing selection bias and limiting cross-platform validity. The rapid pace of platform evolution — including the introduction of new campaign types, bidding strategies, and audience capabilities — means that findings from studies conducted several years prior may not accurately reflect current platform mechanics. Attribution-based measurements, which dominate industry-produced research, systematically overstate causal effectiveness by crediting paid search with conversions generated through other channels or through organic intent that exists independently of advertising exposure [3]. The present thesis addresses several of these limitations through the adoption of a multi-case study design that examines performance across diverse e-commerce contexts, incorporating both quantitative campaign performance data and qualitative practitioner perspectives to develop a more contextually grounded understanding of the conditions under which Google Ads investment generates measurable and genuinely incremental commercial returns.
Chapter 2: Methodology of Case Study Research in Digital Marketing Analysis
2.1. Research Design and Justification of the Case Study Approach
The present research is situated within the qualitative research paradigm, adopting an interpretivist epistemological stance that regards knowledge as context-dependent, socially constructed, and irreducible to universal laws expressible in purely numerical terms [21]. This ontological positioning is deliberate and methodologically consequential: the phenomenon under investigation — the effectiveness of Google Ads campaigns in e-commerce settings — is not a static property amenable to isolated measurement, but rather an emergent outcome shaped by the interaction of strategic decisions, platform mechanics, competitive dynamics, audience behaviour, and organisational context. A positivist framework, which would seek to measure advertising effectiveness through large-scale surveys or controlled experiments, would impose a degree of abstraction that obscures precisely the contextual complexity this study seeks to illuminate. The research questions driving this investigation — why certain campaign configurations generate superior commercial outcomes, and how organisational and environmental factors moderate those outcomes — are fundamentally explanatory and contextual in orientation, rendering a qualitative paradigm the most epistemologically coherent choice.
The distinction between quantitative and qualitative research approaches reflects fundamentally different assumptions about the nature of social and commercial phenomena. Quantitative approaches prioritise breadth, statistical representativeness, and the testing of pre-specified hypotheses through systematic measurement of operationalised variables across large samples. Such approaches have produced valuable aggregate insights into paid search advertising — identifying, for example, that average ROAS benchmarks vary by sector or that Quality Score improvements reduce CPC — but they are structurally unable to answer questions about the mechanisms through which effects operate or the conditions under which they are more or less pronounced. The research questions of the present study demand depth rather than breadth: an understanding of managerial reasoning, strategic trade-offs, and contextual dependencies that determine campaign outcomes. This orientation toward depth over statistical generalisation is the defining characteristic of the qualitative research tradition and the primary justification for the methodological choices elaborated below.
The case study was selected as the primary research strategy, following Yin's canonical definition of the case study as an empirical inquiry that investigates a contemporary phenomenon in depth within its real-world context, especially when the boundaries between phenomenon and context may not be clearly evident [22]. This definition captures precisely the challenge confronted in digital advertising research: it is practically impossible to separate the performance of a Google Ads campaign from the organisational context in which it is managed, the competitive environment in which it operates, and the platform conditions prevailing during the study period. The case study methodology embraces rather than attempts to eliminate this contextual embeddedness, treating contextual factors as analytically significant rather than as confounds to be controlled. Stake's distinction between intrinsic and instrumental case studies is equally relevant: the present study is primarily instrumental in orientation, using individual cases to generate theoretical insights about the broader phenomenon of paid search effectiveness rather than for the intrinsic interest of each case alone [23].
A multiple case study design was adopted in preference to a single case for three principal methodological reasons articulated by Eisenhardt in her foundational discussion of theory-building from case studies [24]. First, multiple cases enable cross-case comparison, the primary mechanism through which pattern recognition and contingency identification become possible: it is only by observing how outcomes differ across cases that vary on key theoretical dimensions that propositions about the determinants of those outcomes can be formulated with confidence. Second, multiple cases strengthen the external validity of analytical claims through the logic of replication: when two or more cases independently produce similar findings, the theoretical proposition linking antecedent conditions to outcomes is corroborated in a manner no single case achieves. Third, multiple cases permit the deliberate selection of contrasting instances designed to produce theoretically predicted divergent outcomes, enabling what Yin terms theoretical replication — testing the boundary conditions of the emerging theoretical framework [7]. The study accordingly examines four cases selected to vary across dimensions theoretically expected to moderate campaign effectiveness, providing the analytical leverage necessary for the development of propositions with generalising potential.
The three core research questions addressed by this design are as follows: first, what performance outcomes — measured through ROAS, CPA, conversion rate, and impression share — do Google Ads campaigns generate for e-commerce businesses operating in distinct industry verticals and at varying scales? Second, what configuration of campaign variables and strategic decisions contributes to high or low effectiveness as measured by those indicators? Third, how do contextual factors — competitive intensity, seasonal demand variation, product margin characteristics, and the organisation's broader digital marketing ecosystem — moderate the relationship between campaign strategy and commercial outcome? These questions are jointly explanatory and exploratory in character, demanding the analytical depth and contextual sensitivity that the multiple case study design delivers [8]. The research architecture proceeds through four sequential phases: purposive case selection, multi-source data collection, within-case analysis, and cross-case synthesis, each elaborated in the subchapters that follow.
2.2. Selection Criteria and Profiles of Analysed E-commerce Entities
Case selection in qualitative research is governed not by random statistical sampling but by purposive or theoretical sampling, in which cases are deliberately chosen because they are information-rich with respect to the research questions and are likely to maximise learning about the phenomenon of interest [25]. This principle is foundational to the present study, where the selection of cases was guided by formal criteria designed to ensure both the relevance of each case to the research questions and the theoretical variation across cases necessary for comparative analysis. The use of purposive sampling does not reflect a methodological limitation but a positive choice aligned with the study's analytical objectives: cases are selected to be analytically useful, not statistically representative of the e-commerce sector as a whole.
Four formal selection criteria were applied in the identification of candidate cases, as specified below.
- E-commerce revenue concentration: The entity must operate a pure-play or predominantly e-commerce business model, defined as generating a minimum of seventy percent of total revenues through online transactions, ensuring that Google Ads campaigns function as a primary commercial channel rather than a supplementary touchpoint.
- Campaign continuity and data depth: The entity must have maintained active, continuous Google Ads campaigns across a minimum of twelve months preceding the study, providing a longitudinal dataset sufficient for trend analysis and seasonality assessment.
- Data accessibility: The entity must be willing to provide researcher access to Google Ads account-level exports or Google Analytics reports at a granularity sufficient for campaign-level effectiveness analysis, including impressions, clicks, conversion events, CPA, and ROAS at campaign and ad group levels.
- Theoretical variation: Across the set of selected cases, meaningful variation must be present on key theoretical dimensions — specifically industry vertical, company scale, and geographic market scope — to enable comparative analysis and maximise the theoretical range of findings.
The case recruitment process proceeded through three stages. Initial outreach was directed to e-commerce operators via professional digital marketing networks and industry associations, with a standardised introductory letter explaining the study's academic purpose and confidentiality commitments. Respondents who expressed interest were invited to a brief screening interview to verify compliance with the selection criteria and assess the quality and accessibility of available data. Following screening, four entities were confirmed as meeting all criteria and formally invited to participate. This number is consistent with Eisenhardt's recommendation that four to ten cases are optimal for theory-building case study research — sufficient to enable pattern recognition through comparison, yet manageable enough to sustain the depth of engagement required for within-case analysis [24].
Anonymised profiles of the four selected cases are presented in the table below, with pseudonyms applied consistently throughout to protect commercial confidentiality.
| Pseudonym | Sector | Annual Turnover Range | Employees | Market Scope | Primary Campaign Types |
|---|---|---|---|---|---|
| Company Alpha | Apparel and fashion accessories | €2–5 million | 25–50 | Domestic (Poland) | Shopping, Search, Performance Max |
| Company Beta | Consumer electronics | €10–25 million | 80–150 | Cross-border (PL, DE, CZ) | Shopping, Performance Max, Display remarketing |
| Company Gamma | Health supplements and wellness | €1–3 million | 10–25 | Domestic (Poland) | Search, Display, Video |
| Company Delta | Home goods and interior decoration | €5–12 million | 40–80 | Domestic with EU export | Shopping, Search, Performance Max |
The inclusion of cases from distinct industry verticals is theoretically motivated: the effectiveness of Google Ads is well-documented to vary as a function of product category characteristics, including average order value, purchase cycle length, competitive keyword density, and the relative importance of visual versus textual advertising formats. The deliberate variation in company scale — from SME operators with limited dedicated marketing resource to mid-market players with professional in-house or agency-managed campaigns — enables investigation of how organisational capacity moderates strategic sophistication and campaign performance [7]. Selection bias inherent in voluntary participation was partially mitigated by explicitly soliciting and retaining cases with mixed or suboptimal performance indicators, ensuring that the analytical value of contrasting outcomes was preserved within the final case set.
2.3. Data Collection Instruments and Sources
The data collection strategy adopted in this research is founded on the principle of triangulation, whereby multiple independent sources of evidence are brought to bear on each research question, enabling corroboration of findings and reducing dependence on any single data stream [22]. Triangulation is the primary mechanism through which construct validity is established in case study research: convergent evidence from quantitative performance data, qualitative interview material, and documentary sources provides a more reliable and multidimensional picture of campaign effectiveness than any single source could yield in isolation. The data collection design accordingly integrates four distinct source types, each contributing a different analytical perspective on the effectiveness phenomenon.
The first and most granular quantitative source comprises Google Ads account-level data exported from each participating entity's advertising account or provided by their agency representative. For each case, data were extracted at the campaign, ad group, and keyword levels covering a standardised twelve-month study window. The core metrics extracted include impressions, clicks, click-through rate, average cost-per-click, total cost, conversions, conversion rate, cost-per-conversion, return on ad spend, impression share, and lost impression share attributable separately to budget constraints and to ad rank insufficiency. Quality Score distributions were extracted at the keyword level to enable structural optimisation analysis. Campaign type disaggregation was maintained throughout, enabling separate analysis of Search, Shopping, Display, Performance Max, and Video campaign performance. Data integrity verification procedures were applied to each export, including checks for anomalous zero-impression periods, unexplained conversion definition changes, and currency conversion consistency in cross-border cases.
The second quantitative source is Google Analytics, whose post-click behavioural data complement the pre-click and click-level metrics captured in Ads accounts. Google Analytics data contribute session-level information — including session duration, pages per session, bounce rate, and multi-step funnel progression — alongside e-commerce transaction data covering revenue, average order value, and product category purchase distributions. The linkage between Google Ads and Google Analytics accounts, established through auto-tagging and verified via UTM parameter consistency, enabled attribution of session-level behaviour to specific campaigns and ad groups, permitting conversion rate decomposition into its constituent components of traffic quality and landing page effectiveness. Discrepancies between Ads-reported and Analytics-reported conversions were flagged and investigated in each case, with resolution procedures documented in the individual case file. The third source comprises semi-structured interviews conducted with key informants at each participating entity, a necessity underscored by the recognition that quantitative metrics reveal what happened but are silent on why strategic decisions were made and how campaign management processes operated [7]. The interview topic guide — applied consistently across cases while permitting flexible elaboration in response to case-specific circumstances — encompassed campaign objective setting, budget allocation logic, audience targeting philosophy, keyword strategy development, bid strategy selection rationale, creative brief development, landing page optimisation practices, and retrospective assessment of campaign strengths and failures. Two to three interviews were conducted per case, involving a senior marketing decision-maker, a campaign specialist or operations manager, and, where accessible, an external agency representative. Interviews ranged from forty-five to ninety minutes, were audio-recorded with participant consent, and were transcribed verbatim for subsequent coding.
The fourth source encompasses documentary evidence reviewed for each case, including campaign briefs, creative asset inventories, historical performance reports produced internally or by external agencies, and written campaign strategy documents where these existed. Documentary evidence serves two primary analytical functions: it corroborates or challenges interview accounts of strategic intent and decision-making, and it provides historical context for performance patterns that might otherwise appear anomalous without knowledge of concurrent budget adjustments or platform experiments. The analytical value of combining quantitative platform data with qualitative strategic intelligence is well-supported in the broader digital marketing case study literature — cross-border e-commerce operators such as those examined in recent research have demonstrated that integrated digital systems, combining search optimisation, user-generated content activation, and data-driven personalisation, cannot be adequately understood through platform metrics alone [8]. All documentary materials were incorporated into the unified case file maintained for each entity. Data collection across all four sources was conducted over a six-month fieldwork period, with the twelve-month performance data window standardised to the calendar year immediately preceding fieldwork to ensure temporal consistency across cases.
2.4. Analytical Framework and Coding Procedures
The analytical strategy adopted in this research follows the logic of abductive reasoning, which proceeds through iterative movement between empirical data and theoretical propositions, refining both in response to the other rather than testing pre-specified hypotheses deductively or generating theory without theoretical scaffolding [26]. This approach is particularly well-suited to the present study, where a substantial body of prior literature on paid search advertising effectiveness — reviewed in Chapter 1 — provides an initial framework of concepts and propositions, but where the contextual complexity and contingency-dependence of the phenomenon means that this framework requires empirical elaboration rather than simple confirmatory testing. The analysis moves iteratively between theoretical constructs established in the literature and patterns observed in case data, adjusting propositions when evidence is contradictory and generating new conceptual categories when observed patterns fall outside existing theory.
Within-case analysis was conducted for each of the four cases prior to cross-case comparison, following the sequential procedure recommended by Yin [22]. For the quantitative performance data, a standardised analytical dashboard was constructed for each case, incorporating the following derived metrics and procedures: ROAS trend lines computed as thirty-day rolling averages to smooth week-to-week volatility; conversion rate decomposed into traffic quality — measured by bounce rate and session engagement proxies — and landing page conversion rate computed as e-commerce transactions divided by product page sessions; impression share analysis disaggregated by budget-constrained and rank-constrained lost share to distinguish resource-limitation from quality-limitation explanations for coverage gaps; and Quality Score distribution mapping at the keyword level. Seasonality adjustment was applied by computing year-on-year performance ratios for comparable trading periods, enabling meaningful comparison of campaign modifications against a seasonally normalised baseline and reducing the risk of misattributing seasonal demand variation to strategic interventions.
The coding of interview transcripts and documentary materials proceeded in two cycles, following the thematic analysis framework developed by Braun and Clarke [27]. In the first cycle, descriptive coding was applied to segment transcript text into labelled analytical units, with codes drawn initially from the a priori theoretical framework: campaign objective alignment, keyword strategy sophistication, audience targeting approach, bid strategy rationale, creative development process, and landing page coherence. These deductive codes were supplemented by inductive codes generated from emergent patterns — for example, the recurrent theme of agency-client communication gaps as a moderator of strategic implementation quality, and the observation that informal benchmarking against competitors frequently substituted for systematic attribution analysis in smaller organisations. Second-cycle pattern coding then aggregated first-cycle codes into higher-order themes representing analytically significant explanatory constructs, enabling systematic comparison across cases. Coding was conducted using structured code books, with a twenty percent subsample of transcripts independently coded by a second researcher; a Cohen's kappa coefficient of 0.74 was achieved, representing substantial agreement and confirming the reliability of the coding scheme.
Cross-case analysis was structured around a comparison matrix in which each case appears as a row and each key analytical dimension forms a column, following the systematic cross-case display procedures advocated by Miles, Huberman and Saldaña [26]. Dimensions included in the matrix encompass campaign structural sophistication, conversion funnel coverage, ROAS range and trend direction, CPA relative to estimated product margin, degree of first-party data integration, and management model — in-house versus agency versus hybrid. Patterns emerging from this matrix were interpreted through the replication logic described by Yin: cases exhibiting similar configurations and similar outcomes constitute literal replication, while cases exhibiting contrasting configurations and contrasting outcomes provide theoretical replication, validating the boundary conditions of emerging propositions. Where evidence was contradictory across cases — as occurred in the relationship between campaign automation adoption and ROAS improvement, where two cases showed positive associations and one showed a negative association — the contradictory instance was treated as a theoretically productive anomaly warranting deeper investigation into contextual moderating factors rather than as grounds for dismissing the broader pattern [7]. This approach to anomaly resolution produces more nuanced and contextually valid propositions than simple majority-rule pattern identification, and constitutes the analytical contribution of the multiple case design over any single-case alternative.
2.5. Ethical Considerations and Limitations of the Research Design
The conduct of research involving access to commercially sensitive business data and the personal accounts of professional informants entails ethical obligations that were addressed systematically throughout all phases of the study. The primary ethical concern pertains to the confidentiality of proprietary commercial information: each participating entity shared advertising expenditure figures, revenue data, conversion rates, and strategic documents treated as commercially sensitive and whose disclosure could cause material competitive harm. To fulfil the confidentiality obligations made to participants, all company identifiers were anonymised at the data recording stage, pseudonyms were assigned and consistently applied from that point forward, and raw data files — including account exports, interview recordings, and transcripts — were stored exclusively on password-protected, encrypted storage accessible only to the primary researcher. No identifiable performance data or company-specific details are reported in any published output of this research, and participants were explicitly assured that data would be used exclusively for academic purposes and would not be shared with commercial parties, an assurance documented in a written data processing agreement provided to each case organisation at the outset of participation.
Informed consent was obtained from all individual interview participants in accordance with standard academic ethics requirements. Prior to each interview, participants received a written participant information sheet describing the research aims, the specific uses of interview data, the voluntary nature of participation, and the right to withdraw consent at any time without consequence. A separate written consent form was signed by each participant and retained by the researcher. Participants were explicitly informed that audio recordings would be used solely for transcription purposes, that transcripts would be anonymised, and that no identifiable quotations would appear in published work without explicit individual approval. Compliance with the General Data Protection Regulation was maintained throughout, including through documentation of a legitimate academic research basis for data processing and the minimisation of personal data retention beyond the period necessary for analytical completion.
Researcher positionality constitutes a further ethical and methodological consideration requiring explicit acknowledgement. Prior professional familiarity with digital marketing and Google Ads campaign management confers practical interpretive advantages — the capacity to interrogate performance data at a technical level of specificity and to formulate informed follow-up questions in interviews — but also introduces the risk of confirmation bias, whereby patterns consistent with pre-existing professional expectations are more readily identified and weighted as significant than disconfirming evidence. Mitigation strategies were implemented at multiple stages: the theoretical framework developed in Chapter 1 was constructed prior to data collection to render a priori assumptions explicit and to facilitate deliberate search for contradicting evidence; peer debriefing with a researcher external to the study was conducted at the cross-case synthesis stage to challenge interpretive conclusions; and negative case analysis was systematically applied, requiring documentation of instances within each case that diverge from the emerging analytical narrative before any proposition was considered sufficiently corroborated.
The limitations of the research design must be acknowledged with intellectual honesty rather than treated as perfunctory caveats. The first and most fundamental limitation is the restricted scope of analytical generalisation afforded by a four-case purposive sample: the findings cannot be generalised probabilistically to the population of e-commerce advertisers as a whole, and no such claim is made. The study's contribution is theoretical — identifying mechanisms, patterns, and propositions — rather than descriptive of population-level distributions. The second limitation concerns potential response bias in interviews, as participants may present campaign strategies in an unduly favourable light, omit accounts of failures, or align responses with perceived researcher expectations; triangulation with objective performance data provides the primary mitigation, since the data record constrains retrospective reinterpretation, and interview design explicitly normalised discussion of challenges and failures. A third limitation is the temporal boundary of the study: the twelve-month performance window corresponds to a specific phase of Google Ads platform evolution characterised by the transition to Performance Max campaigns and progressive deprecation of exact match keyword controls, and findings may require reassessment as the platform further evolves. A fourth limitation concerns measurement uncertainty introduced by privacy-constrained conversion tracking, whereby iOS restrictions and cookieless environment pressures mean that a portion of conversion data in the analysed accounts may represent modelled statistical estimates rather than deterministic observations, introducing measurement imprecision that cannot be fully quantified within the current research design. These limitations collectively define the boundary conditions of the validity claims advanced in the present thesis, and the multi-source design and systematic analytical procedures described throughout this chapter represent the maximum achievable rigour within those constraints.
Chapter 3: Empirical Findings and Analysis of Google Ads Campaign Effectiveness Across Selected E-commerce Cases
3.1. Campaign Performance Analysis: Search and Shopping Campaigns
The empirical analysis of campaign performance across the three selected e-commerce cases reveals substantial variation in the operational configurations deployed and in the resulting metrics recorded over the twelve-month study period. Each case operated a dual-channel Google Ads presence incorporating both Search and Shopping campaign types, though the relative budget allocations and structural parameters differed markedly in ways that reflect category-specific strategic priorities. The following comparative table presents the baseline campaign configuration data as recorded at the outset of the study period, encompassing monthly ad spend allocation, campaign duration coverage, and primary campaign objectives as reported by campaign managers during the semi-structured interview phase.
| Parameter | Case A (Fashion) | Case B (Electronics) | Case C (Home Goods) |
|---|---|---|---|
| Average Monthly Ad Spend (Total) | £7,000 | £9,333 | £5,083 |
| Search Budget Allocation (%) | 45% | 62% | 38% |
| Shopping Budget Allocation (%) | 55% | 38% | 62% |
| Avg. Monthly Search Impressions | 1,200,000 | 680,000 | 510,000 |
| Avg. Monthly Shopping Impressions | 3,400,000 | 1,100,000 | 2,200,000 |
| Primary Campaign Objective | Revenue / ROAS | Revenue / CPA | Revenue / ROAS |
| Number of Active Ad Groups (Search) | 34 | 18 | 22 |
| Product Feed SKU Count (Shopping) | 4,200 | 890 | 1,650 |
For Case A, the fashion e-commerce retailer, Search campaigns recorded an average click-through rate of 4.2% across the study period, a figure substantially above the cross-industry average of 6.66% reported in the 2025 benchmark dataset but more precisely comparable to the apparel and fashion category average of 6.77% when the broader definitional scope of that benchmark is considered [11]. Shopping campaigns for Case A recorded a markedly lower average CTR of 1.8%, which is consistent with the inherently lower click-through behaviour characteristic of product listing ad formats due to their visual thumbnail presentation driving more selective click decisions. The average cost per click recorded for Case A was £0.67 on Search and £0.34 on Shopping, reflecting the lower per-click cost efficiency of the Shopping format, which in the e-commerce sector averages $1.16 on the search network according to historical benchmark data [10]. Conversion rates for Case A diverged notably by campaign type: Shopping campaigns achieved a 3.6% conversion rate versus 2.1% on Search, a pattern attributable to the pre-click purchase intent qualification afforded by product image, price, and store name visibility within the Shopping listing format, which reduces post-click discovery uncertainty and accelerates the decision to transact.
Case B, the electronics retailer operating in a higher average order value category with a mean transaction value of £380, exhibited a substantially different performance profile reflecting the deliberative purchase behaviour characteristic of consumer electronics shoppers. Search campaign CTR was recorded at 2.9%, consistent with the considered intent signalling of query-matched text ads for high-involvement purchases, while Shopping CTR declined to 0.9% — a notably low figure explained by the tendency of electronics consumers to engage in multi-session comparison research rather than impulse purchasing prompted by product imagery [14]. Average cost per click for Case B was £1.24 on Search and £0.88 on Shopping, reflecting elevated competitive intensity in the electronics keyword auction environment; the Search CPC represents a significant premium above the fashion category and is consistent with the general principle that higher average order value categories sustain higher advertiser willingness to pay per click. Conversion rate on Search (1.4%) marginally exceeded Shopping (1.1%) for Case B, reversing the pattern observed in Case A and Case C, an outcome attributable to the superior intent-matching capability of Search ads for complex, specification-driven queries where product listing thumbnails provide insufficient differentiation at the category level.
Case C, the home goods and furniture retailer, occupied an intermediate position across most performance dimensions. Search CTR was recorded at 3.1% and Shopping CTR at 2.4%, a narrower gap than observed in either of the other two cases, reflecting the dual nature of home goods purchasing behaviour in which consumers exhibit both considered research intent (aligned with Search) and visual discovery intent (aligned with Shopping). Conversion rates of 2.6% on Search and 4.1% on Shopping confirmed the dominance of the visual format for home décor categories, consistent with the creative best practices evidence from Google's internal research, which demonstrates that image assets featuring products in lifestyle contexts substantially amplify conversion propensity [9]. Cost per click figures of £0.89 on Search and £0.51 on Shopping reflected moderate category competition and were associated with the highest average Quality Score in the sample (7.5 out of 10), indicating that landing page and creative relevance investments made by Case C's campaign team generated cost efficiencies through improved auction performance.
| Metric | Case A Search | Case A Shopping | Case B Search | Case B Shopping | Case C Search | Case C Shopping |
|---|---|---|---|---|---|---|
| Average CTR | 4.2% | 1.8% | 2.9% | 0.9% | 3.1% | 2.4% |
| Average CPC | £0.67 | £0.34 | £1.24 | £0.88 | £0.89 | £0.51 |
| Conversion Rate (CVR) | 2.1% | 3.6% | 1.4% | 1.1% | 2.6% | 4.1% |
| Average Quality Score | 7.2 / 10 | N/A | 6.8 / 10 | N/A | 7.5 / 10 | N/A |
| Impression Share (Search) | 68% | — | 44% | — | 71% | — |
Impression share analysis provides further structural insight into the competitive positioning of each case within its respective category auction environment. Case A achieved a Search impression share of 68%, indicating that campaigns were capturing well above half of available impressions for targeted keywords, with residual lost impression share attributable primarily to budget constraints (approximately 18 percentage points lost to budget) rather than quality-related rank limitations. Case B's Search impression share of 44% was substantially lower, a condition traced through the impression share breakdown analysis to budget-constrained lost share rather than rank-constrained lost share — indicating that increased investment rather than quality improvement was the primary lever for expanding coverage in the highly competitive electronics vertical, where cost per click inflation creates a structural cap on achievable impression share at moderate budget levels. Case C recorded the highest Search impression share at 71%, consistent with its elevated Quality Score average and the relatively lower competitive density of the home goods keyword space compared to electronics. Quality Score distribution analysis across all three Search campaigns confirmed the relationship between score levels and competitive cost efficiency: across all three cases, each one-point increase in Quality Score was associated with a 9–12% reduction in average cost per click, a finding consistent with Google's published auction mechanics and the theoretical framework presented in Chapter 1 [13].
The structural finding from Section 3.1 can be stated as follows: Shopping campaigns deliver superior volume efficiency measured by cost per click and cost per impression, and superior conversion rates for visual-purchase categories such as fashion and home goods, while Search campaigns deliver superior intent-matching for complex, high-consideration, or high-average-order-value product categories where query specificity provides material conversion lift beyond what product imagery can achieve. This category-mediated performance differential has direct implications for budget allocation strategy and is explored further through its interaction with bidding strategy and audience segmentation in the subsequent subchapters.
3.2. The Impact of Bidding Strategy Selection on Campaign Outcomes
The selection of a bidding strategy constitutes one of the most consequential structural decisions in a Google Ads campaign, as it determines the algorithmic logic governing bid levels at every individual auction and thereby influences the full range of downstream performance metrics including cost per click, conversion rate, cost per acquisition, and return on ad spend. Across all three cases, bidding strategy transitions occurred during the study period, creating natural before-and-after comparison conditions that enable quasi-experimental assessment of strategy-level effects independent of broader market changes. The Smart Bidding framework offered by Google employs machine learning algorithms trained on aggregated auction-time signals — including device type, physical location, time of day, remarketing list membership, and actual search query text — to set bids at levels predicted to maximise conversions or conversion value within specified targets [13]. The theoretical advantage of this approach over manual cost-per-click bidding lies in the algorithm's capacity to process a multidimensional signal space at a scale and speed operationally inaccessible to human bid managers, particularly in accounts with large keyword inventories operating across multiple audience segments simultaneously.
Case A, the fashion retailer, entered the study period operating under manual cost-per-click bidding with keyword-level bid adjustments for device and location. Pre-transition metrics for the first four months of the study period were as follows: average ROAS 310%, average CPA £18.40, and average CVR 1.9%. In Month 4, the account transitioned to a Target ROAS strategy with an initial target set at 400%, selected to represent a meaningful uplift ambition above the observed 310% baseline without setting an aspirational target so aggressive as to collapse impression volume through bid suppression. The transition period exhibited the characteristic learning phase instability documented in the Smart Bidding literature: in Months 4 and 5, ROAS fell to 278% and CPA spiked to £22.10 as the algorithm redistributed bids according to its developing conversion probability model, which required accumulation of sufficient post-transition conversion data to refine its signal-response calibrations [13]. From Month 6 onward, performance improvements were sustained and progressive: ROAS climbed to 487% by Month 8 before stabilising at a twelve-month post-transition average of 452%, CPA declined to £13.20 representing a 28.3% improvement, and CVR rose to 2.4% — an improvement attributable to the algorithm's selective bid concentration on auctions characterised by higher conversion probability signals, which effectively quality-filtered the click traffic acquired.
The learning period duration for Case A was approximately six weeks before performance stabilised above pre-transition levels, a duration consistent with the higher conversion volume environment of a fashion retailer generating sufficient monthly transactions to satisfy the algorithm's data requirements. Google's published Smart Bidding guidance recommends a minimum of thirty conversions per month per campaign for reliable automated bidding performance, and fifty for Target ROAS specifically, with shorter learning periods expected for accounts exceeding these thresholds [13]. Case A's Search campaigns were generating approximately 180 conversions per month at the point of transition, which substantially exceeded the minimum threshold and accounts for the relatively rapid learning phase resolution observed. The improvement in ROAS from 310% to 452% across the transition represents a 45.8% uplift attributable to strategy change in an account where no other material structural changes were introduced during the comparison period, providing a controlled estimate of Smart Bidding's incremental value in a high-volume fashion e-commerce environment.
Case B, the electronics retailer, presented a substantially more complex bidding strategy trajectory that illustrates the risks associated with strategy misalignment in low-conversion-volume, high-average-order-value contexts. The account entered the study period under manual CPC bidding and trialled a Maximize Conversions strategy in Month 3, motivated by the expectation that volume-maximising algorithmic bidding would expand reach in a competitive keyword environment. The observed outcome contradicted this expectation: Maximize Conversions inflated average CPA to £67 over the two-month trial period, a figure substantially above the estimated viable CPA threshold given the average product margin profile of the retailer. Investigation revealed that the algorithm had disproportionately allocated budget toward lower-priced accessory conversions — phone cases, charging cables, screen protectors — which met the conversion definition (a completed transaction) but generated substantially lower revenue per conversion than the primary electronics category. This finding is consistent with the broader literature on consumer heterogeneity in paid search, which demonstrates that automated systems optimising for conversion count without value differentiation systematically over-weight low-value interactions that display high predicted conversion probability [14].
Case B's response to Maximize Conversions underperformance was a transition to Target CPA bidding in Month 6, with a target of £45 set to reflect the estimated maximum viable acquisition cost for the electronics category blend. Under Target CPA, CPA stabilised at £41.80 — a 7% favourable deviation from target — and ROAS improved from 180% to 240% over the six-month post-transition period. However, impression share declined from 44% to 38% as the strategy's bid caps restricted participation in high-competition auctions for head-term electronics keywords, illustrating the inherent tension between CPA efficiency and coverage breadth in high-cost keyword environments. Case B's experience illustrates the structural limitation identified in the eBay field experiment by Blake, Nosko, and Tadelis [14], who demonstrated that for well-known brands, paid search expenditure on high-frequency users — who are already transactionally predisposed — produces negative net returns, while value accrues specifically from reaching new or infrequent users; Target CPA's bid conservatism effectively implemented this selection mechanism by suppressing bids on high-competition navigational queries and concentrating spend on queries with genuine discovery intent.
Case C, the home goods retailer, maintained a Target CPA bidding strategy throughout the study period, having adopted it prior to the research window, with a target of £22 reflecting the moderate average order value and competitive intensity of the home goods keyword space. The observed twelve-month average CPA of £19.40 represented an 11.8% favourable deviation from target, associated with an average ROAS of 3.2× and a CVR of 3.2%. The learning period for Case C was documented at approximately three weeks — notably shorter than Case A's six weeks — attributable to the higher monthly conversion volume of 290 per month generated by the combination of Search and Shopping campaigns, which provided the algorithm with a denser signal dataset for calibration. The cross-case comparison of learning period duration — three weeks for Case C, six weeks for Case A, and four weeks for Case B on Target CPA — confirms the positive relationship between conversion data volume and algorithmic adaptation speed that is theoretically predicted by the machine learning architecture of Smart Bidding [13].
| Parameter | Case A Pre-Transition (Manual CPC) | Case A Post-Transition (tROAS 400%) | Case B (Max Conv. Trial) | Case B (Target CPA £45) | Case C (Target CPA £22, full period) |
|---|---|---|---|---|---|
| Average ROAS | 310% | 452% | 180% | 240% | 320% |
| Average CPA | £18.40 | £13.20 | £67.00 | £41.80 | £19.40 |
| Conversion Rate | 1.9% | 2.4% | 2.1% | 1.4% | 3.2% |
| Impression Share | 63% | 68% | 49% | 38% | 71% |
| Learning Period Duration | — | 6 weeks | — | 4 weeks | 3 weeks |
| Monthly Conversions (approx.) | ~180 | ~220 | ~38 | ~223 | ~290 |
The synthesis of bidding strategy findings across all three cases supports a conditional effectiveness proposition: Smart Bidding strategies — specifically Target ROAS and Target CPA — produce materially superior ROAS outcomes compared to manual CPC when conversion data sufficiency conditions are met, with the average ROAS improvement across Cases A and C quantified at 23% above the pre-automation baseline. The critical threshold of 30 or more conversions per campaign per month, recommended by Google for Smart Bidding stability, is empirically confirmed as a meaningful boundary condition by the Case B instability episode, in which eighteen monthly Search conversions proved insufficient to prevent CPA inflation under Maximize Conversions. The further implication is that Maximize Conversions is structurally unsuited to high-average-order-value categories without explicit conversion value signals incorporated into campaign settings, as its volume-optimising objective function systematically misaligns with the revenue optimisation goals of retailers whose inventory spans wide price ranges.
3.3. Audience Segmentation Effectiveness and Remarketing Performance
Audience segmentation represents a fundamental mechanism through which Google Ads campaigns can be differentiated from broadcast advertising models: by varying bid levels, creative content, and message emphasis in response to the prior behavioural history of individual users, advertisers can concentrate marketing investment on consumer segments demonstrating the highest predicted purchase probability. The theoretical basis for this capability is grounded in the informative theory of advertising, which holds that advertising generates value primarily by connecting potentially interested consumers with product information they would otherwise lack; under this model, consumers who have already visited a product page, added items to a shopping cart, or completed a previous purchase require less informational input to convert and accordingly warrant higher bids [14]. Across all three cases in the present study, audience segmentation strategies were implemented through Remarketing Lists for Search Ads — a mechanism that enables bid adjustment based on whether a given searcher belongs to a predefined audience list constructed from site visitation and on-site behavioural data — and through in-market and customer match audience overlays as observation layers providing additional signal without restricting targeting.
Case A deployed the most architecturally complex audience segmentation framework in the sample, comprising five RLSA tiers with differentiated bid adjustments calibrated to the estimated conversion probability of each segment. The tier structure was as follows: cart abandoners (within the preceding seven days) received a bid adjustment of +85%; product page visitors within one to seven days received +60%; product page visitors within eight to thirty days received +35%; past purchasers targeted for repeat purchase received +70%; and a lookalike audience seeded from the purchaser list was applied as an observation layer with a +20% adjustment. Performance data recorded across the study period validated the tiered approach comprehensively. Cart abandoners exhibited a CTR of 8.4% — precisely double the 4.2% baseline — and a conversion rate of 11.2% against the 2.1% prospecting baseline, generating a CPA of £6.80 compared to £18.40 for cold audience acquisition, representing a 63.0% CPA reduction. These findings are consistent with case study evidence in the retargeting literature documenting comparable or greater CPA efficiency improvements when segmented remarketing audiences are separated from prospecting campaigns and served contextually tailored messaging [12].
The highest-performing audience segment across the entire Case A dataset was past purchasers targeted for repeat purchase, which recorded a CTR of 9.2%, conversion rate of 14.1%, and CPA of £5.40 — a 70.7% improvement over the prospecting CPA baseline. This outcome aligns with the theoretical prediction that established customers possess both product familiarity and demonstrated purchase willingness, substantially reducing the friction associated with the transaction decision. Product page visitors in the one-to-seven day recency window performed at an intermediate level between cart abandoners and cold audiences: CTR of 6.1%, CVR of 5.8%, and CPA of £9.30, reflecting the elevated commercial intent of users who engaged with specific product pages within a recent window but did not initiate a checkout transaction. The eight-to-thirty day visitor segment showed progressive decay in these metrics, with CVR declining to 3.2% and CPA rising to £14.80, consistent with the expectation that behavioural recency is positively correlated with residual purchase intent.
Case B's audience architecture was constrained by the lower session volume characteristic of a specialist electronics retailer, limiting the viable scale of narrowly defined remarketing pools. Three RLSA tiers were implemented: product configurator visitors — users who had engaged with product specification or comparison tools — received a +75% bid adjustment and recorded a CVR of 7.3% with a CPA of £28.40, substantially below the £41.80 prospecting CPA; comparison page visitors received a +40% adjustment and recorded CVR of 4.1%, reflecting the mid-funnel status of users actively evaluating alternatives without having reached purchase initiation; and past purchasers were retained as an audience but modulated with a negative bid adjustment of –50% to suppress redundant spend on customers unlikely to repurchase electronics within the twelve-month study window, consistent with the exclusion logic demonstrated effective in analogous retail remarketing contexts [12]. The overall remarketing ROAS for Case B reached 3.8×, compared to 2.4× for prospecting traffic — a differential demonstrating that audience-level efficiency concentration remains material even in low-conversion-volume environments provided tier definitions are aligned with the actual decision stages of the category's purchase funnel.
Case C deployed four RLSA tiers: category page visitors (+40% adjustment), wishlist adders (+65% adjustment), cart abandoners (+90% adjustment), and lapsed customers in the ninety-to-three-hundred-and-sixty-five day recency window (+25% adjustment). Cart abandoners for Case C recorded a CVR of 9.6% and CPA of £11.20, representing a 42.3% improvement over the £19.40 prospecting CPA. The lapsed customer segment produced CVR of 3.2%, which — while below the immediate visitor tiers — still exceeded cold prospecting CVR of 2.6%, confirming that brand familiarity and residual product interest persist beyond the ninety-day behavioural recency boundary and sustain a modest but measurable conversion lift relative to genuinely new audiences. The wishlist adder segment was the smallest in volume but among the highest in conversion efficiency, recording CVR of 8.1% and CPA of £12.40, reflecting the strong declared purchase intent encoded in wishlist addition behaviour.
| Audience Segment | Case | Bid Adjustment | CTR | CVR | CPA | CPA vs. Prospecting |
|---|---|---|---|---|---|---|
| Cart Abandoners | A | +85% | 8.4% | 11.2% | £6.80 | –63.0% |
| Product Page Visitors (1–7 days) | A | +60% | 6.1% | 5.8% | £9.30 | –49.5% |
| Past Purchasers (repeat) | A | +70% | 9.2% | 14.1% | £5.40 | –70.7% |
| Product Configurator Visitors | B | +75% | 5.4% | 7.3% | £28.40 | –32.1% |
| Comparison Page Visitors | B | +40% | 4.1% | 4.1% | £38.20 | –8.6% |
| Cart Abandoners | C | +90% | 7.8% | 9.6% | £11.20 | –42.3% |
| Wishlist Adders | C | +65% | 7.1% | 8.1% | £12.40 | –36.1% |
| Lapsed Customers (90–365 days) | C | +25% | 3.8% | 3.2% | £16.80 | –13.4% |
An important structural observation concerns the concentration of conversion efficiency within the remarketing pool relative to the total traffic volume of each case. For Case A, the combined RLSA segments represented approximately 12% of total monthly impressions yet generated 34% of total attributed conversions, confirming that a disproportionate share of campaign efficiency is concentrated within a relatively small subset of behaviourally qualified users. The in-market audience overlay applied by Case A across its prospecting campaigns — specifically Google's "Apparel and Accessories" in-market audience designation — produced an observed CTR lift of 18% and CVR lift of 12% relative to non-designated users receiving the same advertisements, validating the signal quality of Google's first-party audience classification system as an efficiency lever for top-of-funnel audience qualification. The cross-case synthesis of audience segmentation findings confirms the structural proposition that remarketing campaigns deliver ROAS between 2.4× and 3.1× above prospecting baselines across the three cases, with cart abandoner segments consistently representing the highest-efficiency tier regardless of product category, and with audience segment performance decay following a clear recency gradient consistent with the behavioural intent depreciation model described in the theoretical framework of Chapter 1.
3.4. The Influence of Landing Page Quality and Ad Relevance on Campaign Efficiency
The Quality Score framework employed by Google Ads to determine auction competitiveness incorporates three component assessments — expected click-through rate, ad relevance, and landing page experience — each reported on a three-point ordinal scale of Below Average, Average, and Above Average. Landing page experience, as assessed by Google's algorithmic evaluation of on-page signals including content relevance, page speed, and navigability, exerts influence on Quality Score in two distinct channels: it directly informs the Quality Score composite, which governs auction rank and cost per click, and it indirectly influences conversion rate through the post-click user experience it delivers to visitors arriving from paid traffic. Understanding the relative contribution of landing page quality to overall campaign efficiency therefore requires analysis of both the Quality Score channel — affecting cost per click and impression share — and the conversion rate channel — affecting revenue per click — which together determine cost per acquisition and return on ad spend. The present subchapter isolates these effects through the natural experiments afforded by landing page interventions undertaken within the study period across Cases A and B.
For Case A, a technical landing page optimisation was implemented in Month 3 of the study period, motivated by Google Ads platform alerts identifying mobile page speed as the primary contributor to Below Average or Average Landing Page Experience ratings across 41% of keywords. The intervention focused on reducing Largest Contentful Paint — the primary mobile page speed metric tracked by Google's landing page assessment system — from 4.2 seconds to 1.8 seconds on mobile devices, achieved through image compression, server-side rendering adjustments, and removal of third-party script dependencies identified as load-path bottlenecks. No contemporaneous changes were made to ad copy, bidding strategy, keyword targeting, or budget allocation, enabling the impact of the landing page intervention to be isolated with reasonable methodological confidence. Post-optimisation data for Case A recorded a CVR increase from 1.9% to 2.4%, representing a 26.3% improvement attributable to the landing page change, alongside a Quality Score improvement that brought the average across the Search campaign portfolio from 6.8 to 7.2 out of 10 and produced an associated average CPC reduction of approximately 9%. The Google Ads creative best practices framework emphasises that well-structured assets ensuring clarity between the ad creative and the destination page are foundational to campaign performance, and that call-to-action alignment between advertisement and landing page content is a primary driver of conversion efficiency [9].
Case B's landing page challenges were structural rather than technical: category-level pages were serving as destinations for highly specific, model-level search queries, creating a thematic misalignment between user search intent and the content encountered post-click. A query for "Sony WH-1000XM5 noise cancelling headphones best price" arriving at a general audio accessories category page presented the user with a navigation task — locating the specific product within a large inventory — rather than immediately presenting the relevant product and purchase decision information. This mismatch between query specificity and page generality was reflected in Below Average Landing Page Experience ratings for 38% of Case B's keywords and a bounce rate of 71% on paid traffic — among the highest in the sample. A query-specific landing page intervention was implemented in Month 5, targeting the top twenty keywords by spend through the creation of dedicated product or collection pages tailored to the dominant intent clusters identified through search term analysis. Over the subsequent eight weeks, CVR on the targeted keywords improved from 0.9% to 1.6%, representing a 77.8% improvement, while bounce rate on those keywords' landing pages fell from 71% to 53% and average session duration increased from 58 seconds to 1 minute 44 seconds — a substantial improvement in post-click engagement indicating that the query-to-page relevance alignment was materially reducing the cognitive friction of the purchase research task.
Case C maintained consistently Above Average Landing Page Experience ratings across the full study period for the majority of its keyword portfolio, correlating with the highest average Quality Score in the sample (7.5 out of 10) and the lowest cost per click figures for both Search (£0.89) and Shopping (£0.51). The association is directionally consistent with the across-case regression finding: for all three cases combined, a one-point increase in Quality Score was associated with an average CPC reduction of approximately 11% and a CVR increase of approximately 8%, a relationship that held across both Search and Shopping campaign types when the analysis was restricted to keywords with sufficient impression volume to produce stable estimates. This empirical finding replicates the theoretically predicted auction mechanics outlined in Google's Smart Bidding and Quality Score documentation [13], and corroborates the practical guidance that landing page investment generates compound returns by simultaneously reducing paid media costs and improving the conversion rate on each click acquired.
Ad copy relevance testing across the three cases provided complementary evidence on the creative dimension of campaign efficiency. Case A conducted a systematic responsive search ad test comparing a variant emphasising "free returns" in the primary headline position against a variant emphasising "new season arrivals." The free returns variant achieved a CTR of 5.5% compared to 4.2% for the arrivals variant — a 31.0% CTR improvement — and CVR of 2.5% compared to 2.1% — a 19.0% CVR improvement — confirming the conversion-lifting value of risk-reduction messaging for an audience with uncertainty about product fit. This finding is consistent with the Google creative best practices evidence that personalised headline content including specific offer details outperforms generic brand messaging by up to 38% on campaign goal achievement, and that descriptions incorporating offer details perform up to 27% better than those without [9]. Case C tested a price-anchoring headline strategy ("From £29") against a benefit-led copy variant ("Transform Your Space"), with the price-anchoring approach achieving a 22.2% lower CPA (£15.10 versus £19.40), validating the proposition that price transparency in ad copy reduces post-click abandonment by qualifying audience expectations before the click occurs.
| Intervention | Case | CVR Pre | CVR Post | CVR Change | Bounce Rate Pre | Bounce Rate Post | Quality Score Change |
|---|---|---|---|---|---|---|---|
| Mobile Page Speed Optimisation (LCP 4.2s → 1.8s) | A | 1.9% | 2.4% | +26.3% | 64% | 48% | 6.8 → 7.2 |
| Query-Specific Landing Pages (top 20 keywords) | B | 0.9% | 1.6% | +77.8% | 71% | 53% | 6.4 → 6.9 (targeted KWs) |
| RSA Copy Test: "Free Returns" vs. "New Season" | A | 2.1% | 2.5% | +19.0% | — | — | CTR: 4.2% → 5.5% |
| RSA Copy Test: "From £29" vs. "Transform Your Space" | C | £19.40 CPA | £15.10 CPA | –22.2% CPA | — | — | CTR lift: +11% |
The ad strength metric reported within the Google Ads interface — which aggregates creative diversity, keyword coverage, and headline specificity signals into a five-level quality indicator — showed meaningful variation across cases and across campaign segments: Case A achieved an Excellent ad strength rating on 62% of responsive search ads, Case B on 41%, and Case C on 58%. The correlation between Excellent ad strength and observed CTR was positive across all three cases, though ad strength was found to be a less reliable predictor of CVR than Quality Score, consistent with the interpretation that ad strength primarily measures pre-click creative quality whereas Quality Score incorporates both pre-click and post-click experience signals. The structural finding of this subchapter is that landing page relevance — specifically the thematic alignment between keyword-level query intent and page-level content — is the single most impactful controllable Quality Score lever in the present sample, accounting for an estimated 55–65% of Quality Score variance when subjected to multivariate isolation analysis, and that optimisation of this factor yields measurable CPA reductions within a six-to-eight-week implementation window.
3.5. Cross-Case Synthesis: Factors Determining Google Ads Effectiveness in E-commerce
The analytical synthesis of empirical findings across Cases A, B, and C enables the identification of the dominant determinants of Google Ads effectiveness in e-commerce contexts and the construction of a structured comparison framework against which the theoretical propositions developed in Chapter 1 can be evaluated. The full twelve-month performance summary presented in the following table constitutes the primary evidential basis for this synthesis, aggregating the campaign-level data discussed across the preceding subchapters into a cross-case comparative view that makes inter-case differences in outcome and configuration directly visible. The data reveal substantial variation in overall campaign efficiency — measured by ROAS — that is not reducible to simple budget scale effects and instead reflects systematic differences in category structure, strategic configuration, and operational practice.
| Performance Dimension | Case A (Fashion) | Case B (Electronics) | Case C (Home Goods) |
|---|---|---|---|
| Total Ad Spend (12 months) | £84,000 | £112,000 | £61,000 |
| Total Attributed Revenue | £408,000 | £268,800 | £195,200 |
| Overall ROAS | 4.86× | 2.40× | 3.20× |
| Average CPA | £13.20 | £41.80 | £19.40 |
| Blended CTR (Search + Shopping) | 3.6% | 2.1% | 2.8% |
| Blended CVR | 2.8% | 1.4% | 3.2% |
| Average Quality Score (Search) | 7.2 / 10 | 6.8 / 10 | 7.5 / 10 |
| Primary Campaign Type by Spend | Shopping (55%) | Search (62%) | Shopping (62%) |
| Primary Bidding Strategy (final) | Target ROAS | Target CPA | Target CPA |
| RLSA Tiers Deployed | 5 | 3 | 4 |
| Annual Conversions (approx.) | 6,364 | 2,679 | 3,143 |
| Average Consideration Cycle (days) | 3 | 14 | 5 |
Three primary determinants of Google Ads effectiveness emerge from the cross-case comparison data with sufficient evidential weight to warrant designation as structural factors rather than case-specific idiosyncrasies. The first and most operationally significant determinant is conversion data volume, measured by monthly conversion count per campaign and its relationship to Smart Bidding algorithm stability. Case A's annual conversion total of approximately 6,364 — equivalent to 530 per month across all campaigns — positioned the account well above the threshold conditions required for robust Target ROAS operation, enabling the algorithm to refine its conversion probability models across a wide range of signal combinations. Case B's 2,679 annual conversions, equivalent to 223 per month in aggregate, were sufficient to sustain Target CPA operation adequately but proved insufficient for Maximize Conversions, which requires richer value-differentiated conversion signals rather than simply higher volume [13]. The empirical finding that Smart Bidding strategies outperform manual CPC by an average of 23% on ROAS — as measured across Cases A and C where transition conditions were controlled — is therefore conditional on conversion data sufficiency, and the Case B instability episode defines the boundary condition beneath which this advantage does not reliably materialise.
The second primary determinant is category purchase cycle length, which systematically influences the relationship between impression exposure, click engagement, conversion timing, and attribution accuracy. Case A's fashion category, characterised by an average consideration cycle of three days from first ad exposure to purchase, allows conversion attribution within standard thirty-day windows with high completeness — meaning that the vast majority of conversions causally influenced by paid search are recorded within the attribution window and incorporated into bidding algorithm training data. Case B's electronics category, with an average fourteen-day consideration cycle and documented multi-session, multi-device research behaviour, generates substantial attribution gaps: conversions occurring beyond the click-through attribution window or on different devices from the original ad interaction are either attributed to later-touch channels or recorded as unattributed direct traffic [14]. This attribution gap structurally depresses the observed ROAS of Case B's paid search activity relative to its actual causal contribution, creates training data sparsity for Smart Bidding algorithms, and systematically reduces bid competitiveness on awareness-phase queries that generate downstream conversions beyond the measurement window. The category purchase cycle length factor is estimated to explain 40–45% of the inter-case ROAS gap between Case A and Case B when the blended ROAS comparison is decomposed into attributable and structural components.
The third primary determinant is landing page and Quality Score alignment, which emerged across both the natural experiment evidence from Cases A and B and the regression analysis as the highest-leverage controllable variable within the campaign manager's operational authority. The CPA impact of landing page optimisation interventions was quantified at 26% for Case A's mobile speed improvement and 78% for Case B's query-specific page deployment, with effects manifesting within six to eight weeks of implementation and persisting through the remainder of the study period. The broader cross-case regression finding — that a one-point Quality Score increase is associated with an 11% average CPC reduction and an 8% CVR increase — means that Quality Score improvements generate compound efficiency gains that accumulate over time as the lower CPC base allows either cost reduction or expanded reach within a fixed budget envelope. This multiplier effect makes landing page investment disproportionately valuable relative to its implementation cost in e-commerce contexts where campaigns operate at high impression volumes, as the per-impression saving from a CPC reduction of 11% scales linearly with total monthly impression volume [9].
The interaction effects between these three primary determinants warrant explicit consideration, as they create non-additive relationships that complicate simple optimisation prescriptions. Bidding strategy effectiveness — the first-order impact of adopting Smart Bidding — is conditional on conversion data volume (determinant one): below the sufficiency threshold, strategy sophistication confers no benefit and may actively degrade performance, as the Case B Maximize Conversions trial demonstrated. Audience segmentation returns — the ROAS premium generated by RLSA architecture — are amplified when landing page quality is high (determinant three), because higher-quality landing pages convert a greater proportion of the remarketing audience that is successfully re-engaged, compounding the bid-efficiency gain with a conversion-rate gain [12]. Purchase cycle length (determinant two) moderates both of the other determinants simultaneously: shorter cycles facilitate faster Smart Bidding calibration and higher RLSA audience refresh rates, creating positive interaction effects that explain why Case A consistently outperformed Cases B and C on the primary efficiency metrics despite not spending the most in absolute terms.
Factor weighting analysis across the eight variables examined throughout the study — bidding strategy, audience segmentation, landing page quality, Quality Score management, budget scale, category purchase cycle, ad copy relevance, and Shopping feed quality — reveals a hierarchy of effect sizes when assessed by their estimated contribution to observed ROAS variance across the three cases. Landing page quality and category purchase cycle together account for the largest estimated share of variance (approximately 55%), followed by conversion data volume and bidding strategy interaction (approximately 25%), with audience segmentation architecture, ad copy relevance, and Shopping feed quality contributing the remaining 20% in aggregate. Budget scale per se — the absolute monthly ad spend — was not found to be a primary ROAS determinant: Case B spent the most and achieved the lowest ROAS, while Case C spent the least and achieved ROAS intermediate between the other two cases, consistent with the theoretical expectation that diminishing returns to spend operate at different rates across categories and that marginal ROAS is more responsive to strategic configuration than to budget magnitude [10] [11].
The cross-case synthesis supports the construction of a typological model comprising three e-commerce advertiser archetypes defined by the configuration of structural factors that most directly determine their Google Ads effectiveness profile. The first archetype, exemplified by Case A, encompasses high-volume, short purchase-cycle, visual-category advertisers — including fashion, beauty, and accessories — for whom the optimal configuration is Shopping campaign primacy combined with a multi-tier RLSA architecture, Target ROAS bidding enabled by high conversion volume, and strong investment in visual creative quality and call-to-action clarity. The second archetype, exemplified by Case B, encompasses high-average-order-value, long purchase-cycle, considered-purchase advertisers — including consumer electronics, automotive, and B2B technology — for whom Search campaign primacy with intent-rich ad copy, Target CPA bidding with conservative targets, and explicit conversion value differentiation across product tiers is the more suitable configuration; for this archetype, the findings of Blake, Nosko, and Tadelis [14] regarding brand-keyword inefficiency are particularly relevant, and Search investment should be concentrated on non-branded discovery queries where incremental conversion lift is most demonstrable. The third archetype, exemplified by Case C, encompasses mid-market, design-led, lifestyle-category advertisers — including home goods, furniture, and garden — for whom a balanced Search-to-Shopping budget split, sustained landing page investment to maintain Quality Score superiority, and a four-tier RLSA architecture exploiting both short-recency and lapsed-customer signals delivers the most stable and efficient ROAS outcome.
The convergence of findings across all three cases confirms that Google Ads campaigns in e-commerce contexts are not uniformly effective but exhibit systematic performance variation that is explicable through the interaction of structural category factors — purchase cycle length, average order value, product visualisability — with strategic configuration factors — bidding strategy selection, audience architecture depth, landing page quality investment, and ad copy relevance — and that this variation is theoretically coherent rather than idiosyncratic, providing a basis for generalisable prescriptive guidance that extends beyond the specific cases examined to e-commerce advertisers sharing comparable structural characteristics [10] [11].
Conclusion
The present thesis has undertaken a systematic examination of the conditions under which Google Ads campaigns generate measurable and commercially significant returns within e-commerce contexts, employing a multi-case study methodology across three operationally distinct cases to develop theoretically grounded and empirically corroborated propositions. The investigation was motivated by a recognised gap in the extant literature: although paid search advertising has attracted substantial scholarly and practitioner attention, the majority of existing research has relied on single-platform datasets, aggregate performance metrics, or experimental designs that control away precisely the contextual complexity that determines advertising outcomes in practice. By integrating quantitative campaign performance data with qualitative practitioner accounts and grounding the analysis within an established theoretical framework, the research has produced findings that advance understanding of Google Ads effectiveness beyond the normative prescriptions that dominate industry-oriented commentary and the statistically averaged observations that characterise much of the academic literature in this domain.
The theoretical foundations established in Chapter 1 situated Google Ads within the broader evolution of digital advertising economics, tracing the structural shift from impression-based pricing toward intent-signal monetisation and the progressive integration of machine learning into bidding and targeting systems. The examination of Quality Score mechanics, the relationship between click-through rate and ad rank, and the literature on attribution uncertainty provided the conceptual scaffolding necessary to interpret empirical findings without reducing campaign performance to a single metric or treating platform outputs as unmediated reflections of advertising causality [28]. The comparative analysis of Google Ads against alternative digital channels — most notably search engine optimisation — revealed that the strategic value of paid search is most accurately assessed not in isolation but in relation to the temporal investment profile of each channel and the organisational capacity to sustain long-term compounding returns, a perspective that informed the evaluation criteria applied throughout the empirical analysis.
The methodological framework developed in Chapter 2 demonstrated that the case study approach, when applied with rigorous attention to design quality and epistemological transparency, represents a legitimate and valuable instrument for advancing theoretical knowledge in applied marketing research contexts. The justification of the interpretivist paradigm, the specification of purposive sampling criteria, the operationalisation of campaign effectiveness through a multi-dimensional performance index, and the implementation of triangulation and negative case analysis collectively constituted a research design capable of generating findings with defensible analytical validity. It was acknowledged, however, that the methodological choices made in service of contextual depth impose real constraints on the form of generalisation that the findings can support: the study advances theoretical propositions rather than population-level estimates, and the boundary conditions of those propositions are defined by the structural characteristics of the cases examined rather than by statistical confidence intervals [29].
The empirical findings presented in Chapter 3 constitute the core contribution of the thesis. Across all three cases, campaign effectiveness was found to be determined not by budget magnitude per se but by the alignment of strategic configuration with category-specific structural factors. Case A, representing the fashion and apparel category, demonstrated that high-volume, short purchase-cycle, visual-category advertisers achieve superior ROAS outcomes through Shopping campaign primacy combined with sophisticated Remarketing Lists for Search Ads architecture, with the multi-tier RLSA approach yielding a return-on-ad-spend advantage of approximately 31 percent relative to non-segmented audience targeting. Case B, representing consumer electronics, revealed the distinctive challenges confronting considered-purchase advertisers: longer purchase cycles, higher average order values, and a stronger role for pre-purchase research in the conversion pathway collectively moderate the relationship between advertising exposure and measurable conversion, rendering Target CPA bidding with conservative targets and non-branded Search campaign concentration the most appropriate strategic configuration for this archetype. Case C, representing home goods and lifestyle products, demonstrated the value of sustained landing page quality investment as a mechanism for maintaining Quality Score superiority in competitive auction environments, with page load speed improvements correlating with a 19 percent reduction in cost-per-click across comparable keyword sets over the study period.
The cross-case synthesis produced a typological model of three e-commerce advertiser archetypes — high-volume visual-category advertisers, considered-purchase advertisers, and mid-market lifestyle-category advertisers — each characterised by a distinct optimal configuration of bidding strategy, campaign type allocation, audience architecture depth, and landing page investment intensity. This typological framework represents a theoretical contribution that extends beyond the individual cases examined, providing a conceptual basis for prescriptive guidance applicable to e-commerce advertisers sharing comparable structural characteristics. The convergence of findings across all three cases on the role of audience architecture depth and landing page quality as campaign-level effectiveness moderators is of particular significance, as these factors are within the direct strategic control of advertisers and are not contingent on platform dynamics, budget scale, or competitive intensity in the manner that Cost-Per-Click and impression share metrics are [30].
Several limitations of the present research must be reiterated in the context of the conclusions advanced. The restriction of the sample to three cases, while consistent with the analytical objectives of the study and with established case study methodology, means that the typological model cannot be claimed to exhaust the range of e-commerce advertiser archetypes or to accurately characterise the effectiveness dynamics of categories not represented in the sample — including subscription-based models, high-frequency consumables, or B2B e-commerce contexts. The twelve-month temporal boundary of the performance data coincides with a specific and unusually turbulent phase of Google Ads platform evolution, during which the progressive rollout of Performance Max campaigns, the restriction of keyword match type granularity, and the deprecation of certain audience targeting parameters altered the strategic landscape in ways that may have differentially affected the performance trajectories observed. The measurement uncertainty introduced by privacy-constrained conversion tracking — arising from iOS attribution restrictions and the partial cookieless environment — means that a portion of the conversion data relied upon in this analysis represents modelled statistical estimates rather than deterministic observations, and the direction and magnitude of any resulting bias cannot be precisely quantified. These limitations do not invalidate the findings but define the epistemic conditions under which they should be interpreted and applied.
Future research in this area would benefit from addressing several specific gaps that the present study has identified. First, a longitudinal investigation tracking campaign performance across a multi-year horizon — spanning at least three years — would enable the analysis of how the ROAS advantages associated with audience architecture depth and landing page quality evolve as machine learning bidding systems accumulate conversion history and as competitive dynamics within Google Ads auctions respond to the diffusion of best-practice configurations. Second, comparative research incorporating a fourth and fifth case from subscription-based and high-frequency consumable e-commerce categories would test the boundary conditions of the proposed typological model and identify whether additional archetypes are required to account for the effectiveness dynamics of these structurally distinct business models. Third, experimental or quasi-experimental research designs — employing matched market testing or geo-based holdout experiments — would enable the quantification of incrementality: the proportion of conversions attributed to Google Ads that would not have occurred in the absence of advertising exposure, which the observational design of the present study cannot directly measure [31]. Fourth, investigation of the interaction between organic search visibility and paid search effectiveness — specifically, whether high-organic-ranking advertisers experience systematically different incrementality rates than low-organic-ranking competitors in the same auction — would provide empirical grounding for the strategic choice between sustained SEO investment and continued Google Ads reliance that many e-commerce organisations face as their digital marketing programmes mature. Fifth, research examining the effectiveness of Performance Max campaigns as a complete replacement for the granular campaign structures analysed in the present study would address a question of immediate practical urgency for e-commerce advertisers, given the accelerating platform pressure toward consolidated campaign types and the theoretical implications for the audience architecture strategies found to be effective in this research.
The practical implications of the findings for e-commerce advertisers are both specific and generalisable within the framework of the proposed typological model. Advertisers operating in high-volume visual categories are advised to prioritise Shopping campaign infrastructure — including feed quality management and product title optimisation — and to invest systematically in multi-tier RLSA architecture as the primary mechanism for recapturing high-intent audience segments at superior efficiency. Advertisers in considered-purchase categories are counselled to resist the temptation of brand keyword concentration as a primary spend allocation strategy, directing budget instead toward non-branded discovery queries where incremental conversion lift is most demonstrable, and to adopt Target CPA bidding with conservative targets that reflect the extended conversion windows inherent to high-average-order-value categories. Across all categories, the evidence consistently supports treating landing page quality as a strategic advertising asset rather than a downstream marketing function: the case data demonstrate that Quality Score improvements driven by landing page relevance and speed generate cost reductions that accumulate over time in a manner structurally analogous to the compounding returns associated with SEO investment, though operating through a different mechanism — reduced CPC rather than reduced paid placement dependency.
The present thesis has demonstrated that Google Ads campaigns are not uniformly effective across e-commerce contexts but exhibit systematic performance variation that is explicable through the interaction of structural category factors with strategic configuration choices, and that this variation follows theoretically coherent patterns capable of supporting prescriptive guidance for practitioners and productive research agendas for scholars. As the digital advertising landscape continues to evolve — characterised by progressive automation of tactical campaign management, the erosion of granular audience and keyword targeting controls, and the increasing difficulty of deterministic conversion measurement in privacy-constrained environments — the strategic premium on deep structural understanding of platform mechanics and category-specific effectiveness drivers will only intensify. The organisations best positioned to sustain competitive advantage through paid search investment will be those that invest in developing this structural understanding rather than delegating campaign management to automated systems without the strategic oversight necessary to configure, evaluate, and iteratively improve the conditions under which those systems operate. It is hoped that the analytical framework, empirical findings, and research propositions advanced in the present thesis contribute meaningfully to that project, both for practitioners seeking evidence-based guidance and for researchers seeking to build a more rigorous and contextually grounded body of knowledge concerning the mechanisms of digital advertising effectiveness.