Commerce, Local and Performance Search Campaigns

Introduction

Search advertising has moved beyond the traditional model of buying individual keywords and writing manually controlled text advertisements. Commerce platforms, local inventory systems, mobile applications and automated cross-network campaigns now connect advertising decisions with product catalogues, business locations, audience signals, conversion data and machine-learning systems. This development creates substantial opportunities, but it also changes the work of the search marketer. The practitioner must increasingly design reliable inputs, select appropriate objectives, establish measurement governance and diagnose automated systems rather than merely adjust bids.

This chapter examines five related campaign domains: shopping feeds and product campaigns; local search advertising; app campaigns; cross-network automated campaigns; and performance diagnosis. The central principle is that automation does not remove managerial responsibility. It shifts responsibility towards data quality, goal definition, experimentation, budget control, creative assets, privacy compliance and interpretation of results.

Learning Objectives

By the end of this chapter, the learner should be able to:

1. Design and audit a commerce product feed for discoverability, policy compliance and accurate measurement.
2. Distinguish among shopping, local, app and cross-network campaign objectives and controls.
3. Select suitable conversion goals, bidding approaches, budgets and geographic settings.
4. Diagnose performance using profitability, incrementality, attribution and funnel metrics.
5. Evaluate the opportunities and risks of machine-learning-driven campaign automation.
6. Develop an integrated search campaign plan for a commerce or service organisation.

1. The Strategic Shift from Keywords to Commercial Signals

In conventional search advertising, the advertiser largely specified the query, bid, advertisement and landing page. Modern commerce and performance campaigns combine several signals:

• Intent signals: search queries, browsing behaviour and contextual relevance.
• Product signals: title, brand, category, price, availability, imagery and identifiers.
• Location signals: user location, business location, service radius, opening hours and local inventory.
• Audience signals: first-party lists, remarketing groups and predictive behavioural patterns.
• Outcome signals: purchases, qualified leads, store visits, app installs and in-app actions.
• Business constraints: margin, stock, delivery capability, regulatory restrictions and customer value.

The campaign therefore resembles a socio-technical system. Human decisions establish the objective and constraints; data feeds describe the offer; an advertising platform allocates impressions; and measurement systems report outcomes. A weak input can produce a technically efficient but commercially harmful result. For example, optimising towards inexpensive app installs may generate many low-quality users if the install event is easier to produce than a subscription or purchase.

A useful planning sequence is:

Business objective → measurable event → data and feed design → campaign type → controls → experiment → diagnosis → optimisation.

2. Commerce Search: Product Feeds and Shopping Campaigns

2.1 What a product feed does

A product feed is a structured catalogue supplied to a commerce advertising or merchant platform. It typically contains one record per offer or product variant. The platform uses these records to match products to relevant queries and to create product advertisements. Unlike a conventional text advertisement, a product advertisement depends heavily on the quality of catalogue attributes.

Common attributes include:

A feed should be treated as a governed data product, not as a one-time spreadsheet. It requires ownership, update schedules, validation rules, version control and an escalation process when website and feed information disagree.

2.2 Feed quality and eligibility

Feed quality has two dimensions. Eligibility concerns whether an item can be approved and served. Effectiveness concerns whether the item is attractive, discoverable and commercially competitive. An approved product can still perform poorly because its title is unclear, image is weak, price is uncompetitive or landing page is slow.

A professional audit checks:

1. Completeness: mandatory attributes and variants are present.
2. Consistency: price, currency, availability, delivery and returns agree across feed and site.
3. Identity: identifiers distinguish variants without duplication.
4. Taxonomy: product types and custom labels support analysis.
5. Freshness: stock and price updates occur frequently enough for the business model.
6. Policy: claims, products, destinations and data collection comply with platform and law.
7. Measurement: links contain reliable tracking parameters and purchase values are transmitted accurately.

Custom labels are especially useful for managerial control. They can classify products by margin band, season, clearance status, strategic importance or inventory risk. Such labels should describe business reality, not merely reproduce platform categories.

2.3 Campaign structures and bidding

Commerce campaigns may be organised by product category, margin, geography, brand status, season or promotional strategy. Excessive fragmentation can prevent the system from receiving enough conversion data, while excessive consolidation can hide important differences in profitability or stock availability. The appropriate structure is the smallest structure that supports a distinct decision.

Common objectives include:

• maximise transaction volume within a budget;
• achieve a target return on advertising spend;
• acquire new customers;
• clear inventory;
• protect branded demand;
• promote high-margin or strategically important products.

Return on advertising spend is commonly expressed as:

where R is attributed revenue and C is advertising cost, both measured in the same currency and period. ROAS is a revenue-to-cost ratio, not a profit measure. If product cost, fulfilment, payment fees and returns are material, a contribution-based measure is more useful:

where V is variable non-advertising cost associated with the orders. The measure is meaningful only when revenue, costs, attribution window and order status are defined consistently.

For a product with selling price £100, variable cost £60 and advertising cost £20, contribution is £40 and contribution ROAS is 40 ÷ 20 = 2.0. A revenue ROAS of 5.0 may appear strong, but a contribution ROAS of 2.0 must be compared with overheads and the organisation’s required return.

The break-even advertising cost ratio can be expressed as:

If contribution margin is £40 on £100 revenue, the break-even advertising cost ratio is 0.40, or 40%, before fixed costs and other adjustments. A campaign spending more than this proportion cannot cover the stated variable costs, although accounting treatment and customer lifetime value may change the decision.

3. Local Search Advertising

Local search connects digital intent with a physical location or service area. It is relevant to retailers, restaurants, healthcare providers, education institutions, dealerships, financial services and trades. Local campaigns may seek directions, calls, store visits, appointment requests, local purchases or awareness.

3.1 Locality as a conversion system

A local campaign requires more than a location keyword. It depends on accurate business profiles, address and phone data, opening hours, service areas, landing pages, reviews, local inventory and offline conversion records. Location extensions or equivalent assets should be synchronised with the organisation’s authoritative location database.

The marketing funnel often includes:

1. local discovery or map visibility;
2. website visit or profile interaction;
3. call, direction request, booking or store visit;
4. purchase or qualified service outcome;
5. repeat behaviour and customer value.

A click or direction request is an intermediate event. It should not automatically be treated as equivalent to a sale.

3.2 Geographic controls

Geographic settings should reflect the business model. A retailer may target people physically present in a trading area; a university may target prospective students in several regions; a home-services company may target addresses within a serviceable radius. Marketers must distinguish presence from interest in a location and select settings deliberately. Radius targeting is not a guarantee of equal reach or equal delivery density: population, competition, demand and eligibility vary within the radius.

Location exclusions matter as much as inclusions. They can prevent wasted spend in areas outside delivery capability, protect franchise territories and support local regulatory restrictions. Campaign reporting should compare impressions, calls, qualified leads, store outcomes and cost by location, while recognising that location attribution is probabilistic in many systems.

3.3 Offline measurement

For high-consideration services, the advertising platform may optimise towards a lead that is later rejected. A stronger system imports qualified lead, appointment, enrolment or revenue outcomes from a CRM. These events require stable identifiers, agreed definitions, privacy safeguards and sufficient volume. Offline conversion data should not be uploaded simply because it exists; it should be assessed for accuracy, latency, coverage and representativeness.

4. App Campaigns

App campaigns are designed to acquire users or generate valuable in-app actions. The main strategic distinction is between:

• install acquisition: increasing the number of installations;
• engagement: bringing existing users back to complete an action;
• event or value optimisation: acquiring users likely to register, subscribe, purchase or generate revenue.

Install volume is an input, not necessarily a business outcome. A gaming company may optimise for a tutorial completion or payer event; a banking application may value a verified account; a media application may value a retained subscription.

4.1 Measurement architecture

App measurement should define the event hierarchy before campaign launch:

Install → activation → core action → monetisation → retention.

Events should be deduplicated and passed consistently across operating systems and analytics tools. Privacy frameworks may limit user-level attribution, reduce signal availability or delay reporting. Consequently, marketers should avoid interpreting small changes in reported performance as precise behavioural facts.

Useful app metrics include install rate, activation rate, cost per activated user, purchase rate, revenue per user, retention and lifetime value. For a cohort, a simple observed revenue per acquired user is:

The numerator and denominator must refer to the same cohort and observation period. Early estimates of lifetime value are forecasts and should be labelled as such.

4.2 Creative and quality controls

Automated app campaigns distribute assets across multiple placements. Asset variety is therefore important: concise text, vertical and horizontal video, static images and platform-compliant brand elements. However, variety must not become inconsistency. Creative claims, app-store descriptions and privacy disclosures should communicate the same proposition.

Quality controls include app stability, onboarding friction, permission requests, deep links, store-page conversion and fraudulent-install detection. A low cost per install can conceal poor retention, incentivised traffic or users who cannot complete the onboarding process.

5. Cross-Network Automated Campaigns

Cross-network campaigns use automation to select placements, bids, audiences, creative combinations and sometimes product or local inventory across several surfaces. Their attraction is breadth and reduced manual operation. Their challenge is reduced transparency and the possibility that the platform optimises for a proxy rather than the organisation’s real objective.

5.1 Conditions for responsible automation

Automation is most defensible when five conditions are present:

1. A precise objective: for example, profitable purchases rather than generic traffic.
2. Reliable conversion data: events are correctly implemented, deduplicated and timely.
3. Adequate volume: the system can learn from representative outcomes.
4. Meaningful controls: budget, geography, brand safety, inventory and exclusions are configured.
5. A diagnostic plan: the team can distinguish learning volatility from structural failure.

Automation should be staged. Begin with a controlled scope, validate tracking, observe search terms and placement quality where available, then expand only after the campaign demonstrates acceptable economics and compliance.

5.2 Goal and constraint design

A platform objective is not identical to a business objective. “Maximise conversions” requires a definition of conversion; “target return” requires stable value reporting; “acquire new customers” requires a credible new-customer classification. Constraints may include maximum daily spend, product availability, minimum margin, target geography, prohibited content, frequency considerations and customer-service capacity.

A useful governance distinction is:

• Goal: what outcome should increase?
• Signal: what measured event represents the outcome?
• Constraint: what must not be violated?
• Diagnostic: what evidence indicates success or failure?

6. Performance Diagnostics

Performance analysis should move from delivery to economics and then to causality. A practical diagnostic tree is:

Delivery → interaction → conversion → value → incrementality.

6.1 Core metrics

Click-through rate is:

Conversion rate is:

Cost per acquisition is:

These ratios are descriptive. They do not, by themselves, prove that advertising caused the conversion or that the conversion was profitable.

6.2 Common diagnostic patterns

• Low impressions: eligibility, budget, bids, targeting, feed disapproval or insufficient demand may be responsible.
• Impressions but low clicks: weak price, creative, title, relevance, ranking or offer differentiation may be responsible.
• Clicks but few conversions: landing-page friction, mismatch, technical failure, poor trust or incorrect conversion tracking may be responsible.
• Conversions but weak profit: low margin, discounting, returns, fraud, poor customer quality or over-attribution may be responsible.
• Strong reported performance but no business growth: brand cannibalisation, attribution inflation, seasonality or demand capture rather than demand creation may be responsible.

Segmentation by device, geography, product, audience, new versus returning customer and time period can reveal concentration and risk. Yet segmentation must be interpreted cautiously: small cells are volatile, and repeated slicing can generate misleading patterns.

6.3 Attribution and incrementality

Attribution assigns credit; incrementality estimates additional outcomes caused by advertising. These are not interchangeable. A last-click report may credit a campaign for users who were already likely to purchase. Incrementality may be examined through geo experiments, holdout groups, platform experiments, time-based tests or matched-market designs. Each method has assumptions concerning contamination, randomisation, seasonality and interference.

A sound evaluation reports both platform-attributed outcomes and business-level evidence. Where experiments are not feasible, triangulate with trends, budget changes, organic demand, new-customer rates and independent analytics, while acknowledging uncertainty.

7. Worked Integrated Planning Example

Consider an illustrative retailer selling home equipment through a website and 18 stores. Its objectives are to increase profitable online sales, drive store visits for selected products and acquire new app users.

The retailer first standardises product IDs, prices, stock, store inventory and margin labels. It excludes products with insufficient stock and separates high-margin items from clearance products. Online purchase value is passed with transaction IDs; store outcomes are imported weekly; app campaigns optimise for activated users rather than installs.

Suppose a month produces 1,200,000 impressions, 24,000 clicks and 1,200 online purchases. Advertising cost is £36,000 and attributed revenue is £144,000.

If variable non-advertising cost is £100,800, contribution is £43,200 and contribution ROAS is 1.2. The campaign therefore looks strong on revenue ROAS but has limited contribution after variable costs. Management should investigate product mix, returns, new-customer value and store outcomes before increasing budget. The correct decision is not automatically to reduce spend: incremental store sales or future customer value might justify investment, but these must be measured rather than assumed.

8. Applied Case: Northstar Market (Illustrative)

Background

Northstar Market is a fictional omnichannel retailer with a national website, twelve urban stores and a loyalty application. It sells household goods with wide differences in margin and stock availability.

Problem

Its product campaign reported a revenue ROAS of 5.1, yet total contribution was declining. Local campaigns generated many direction requests but store managers reported little corresponding sales growth. The app team celebrated low install costs while three-month retention fell.

Analysis

An audit identified four problems. First, feed prices were updated daily but stock was updated only twice a week. Second, the campaign optimised all products to revenue, ignoring margin. Third, direction requests were counted as primary conversions even when stores were closed. Fourth, app optimisation used installs instead of activation and retained usage.

Approach and solution

Northstar introduced margin and inventory labels, increased stock-feed frequency and created rules for low-stock products. It separated clearance activity from normal acquisition and evaluated products using contribution ROAS. Local campaigns used accurate opening hours, serviceable locations and imported store transactions where available. The app event hierarchy changed from install to activation and then to a subscription-related event after sufficient data accumulated. A geo-based test compared selected locations with similar control locations rather than relying solely on platform-attributed store visits.

Outcome

The reported revenue ROAS declined because low-value attributed orders were removed from the primary goal and product mix changed. Nevertheless, contribution improved, unavailable-product complaints fell and the app generated fewer but more active users. The case demonstrates that a lower platform metric can represent better business performance when measurement becomes more economically meaningful.

Learning points

1. Feed freshness is a commercial issue, not merely a technical issue.
2. Revenue optimisation can conflict with margin optimisation.
3. Local interactions require offline validation.
4. App scale should be evaluated through user quality and retention.
5. Automation must be governed by business constraints and independent evidence.

9. Governance, Ethics and Regulation

Commerce, local and app campaigns process personal, commercial and location-related information. Organisations should apply data-minimisation, purpose limitation, consent and lawful-processing requirements applicable in their jurisdiction. They should provide understandable notices and respect opt-out choices.

Important governance issues include:

• misleading price, scarcity, health, financial or performance claims;
• discriminatory geographic or audience exclusion;
• targeting vulnerable users;
• inaccurate business hours or availability;
• fabricated reviews or manipulated local reputation;
• dark patterns in app permissions and subscriptions;
• opaque automated decisions and inadequate human review;
• unauthorised use of customer data for audience matching;
• unsafe or unsuitable placements;
• environmental and social costs of excessive data processing.

AI-generated product descriptions or creative assets require factual review, rights clearance, brand safeguards and disclosure where appropriate. Automation should never be used as an excuse to evade accountability. A named owner should be responsible for feed accuracy, conversion definitions, budget limits, policy compliance and incident response.

10. Contemporary Developments and Best Practice

Search platforms increasingly combine product data, broad intent, predictive modelling, generative creative tools and privacy-preserving measurement. This creates three strategic implications.

First, structured first-party data becomes more valuable. Clean product, customer, inventory and CRM data improve both human analysis and machine learning. Second, measurement uncertainty becomes unavoidable as identifiers decline and reporting systems differ. Decision-makers should use ranges, experiments and triangulation rather than false precision. Third, creative and merchandising quality become central performance levers because automation can distribute assets widely but cannot repair a poor proposition, inaccurate stock or an uncompetitive price.

A professional operating model includes:

1. documented business objectives and conversion hierarchy;
2. feed and tag validation before launch;
3. controlled campaign architecture;
4. budget and policy safeguards;
5. weekly operational diagnostics;
6. monthly economic and cohort analysis;
7. periodic incrementality testing;
8. change logs and learning documentation.

Chapter Summary

Commerce, local, app and cross-network campaigns integrate advertising with product catalogues, locations, applications, audience data and business outcomes. Product feeds determine eligibility and relevance; local campaigns connect digital intent to physical service; app campaigns should optimise for valuable user actions rather than installations alone; and automated cross-network campaigns require clear goals, reliable signals and explicit constraints.

Performance diagnosis should proceed from delivery to interaction, conversion, value and incrementality. CTR, conversion rate, CPA and ROAS are useful descriptive metrics, but none proves profitability or causality without appropriate definitions and controls. Contribution-based analysis, offline outcomes, cohort evaluation and experimentation provide stronger managerial evidence. Ethical and regulatory governance is essential because these systems influence commercial opportunity, privacy, location visibility and automated decision-making.

Key Terms

Product feed: Structured catalogue data used to represent products in commerce systems.

Merchant eligibility: Whether a product satisfies platform and policy requirements for serving.

Custom label: Advertiser-defined classification used for segmentation and control.

Shopping campaign: Campaign promoting products using catalogue information.

Local search advertising: Advertising designed to connect users with nearby businesses or services.

Location extension: Business-location information attached to an advertisement or profile.

Offline conversion: A conversion recorded outside the advertising platform and later imported.

App campaign: Campaign intended to acquire or re-engage application users.

Activation: A meaningful early user action indicating initial app value.

Cross-network campaign: Automated campaign distributing ads across multiple inventory types or surfaces.

Conversion signal: Measured event supplied to an optimisation system.

Constraint: Business, policy or operational condition that limits optimisation.

ROAS: Attributed revenue divided by advertising cost.

Contribution ROAS: Contribution after variable costs divided by advertising cost.

CPA: Advertising cost divided by conversions.

Incrementality: Additional outcome caused by advertising relative to a suitable counterfactual.

Attribution: Assignment of credit for an outcome to marketing touchpoints.

First-party data: Data collected directly by an organisation from its customers or users.

Cohort: Group of users or customers sharing a defined acquisition or behavioural characteristic.

Review Questions

Short Answer Questions

1. What is the difference between product-feed eligibility and product-feed effectiveness?
2. Why should a local campaign distinguish direction requests from completed sales?
3. Define contribution ROAS and explain how it differs from revenue ROAS.
4. Why can app-install volume be an inadequate primary objective?
5. What is the difference between attribution and incrementality?

Long Answer Questions

1. Explain how an organisation should design, govern and audit a commerce product feed.
2. Compare shopping, local, app and cross-network campaigns in terms of objectives, signals and controls.
3. Discuss the conditions under which automated bidding and placement decisions are strategically appropriate.
4. Develop a performance-diagnosis framework that moves from delivery metrics to business value.
5. Evaluate the ethical, privacy and regulatory responsibilities involved in automated search advertising.

Analytical / Application Questions

1. A retailer records £90,000 in attributed revenue, £25,000 in advertising cost and £54,000 in variable non-advertising cost. Calculate revenue ROAS and contribution ROAS, then explain which additional information management needs before changing budget.
2. A local service campaign has many calls but few qualified appointments. Design an investigation covering targeting, call quality, landing-page experience, CRM integration and conversion definitions.
3. An app campaign has a low cost per install but declining 30-day retention. Recommend a revised event hierarchy, measurement plan and optimisation strategy.

Multiple Choice Questions with Answers

Question 1

Which feed attribute most directly communicates whether a product can currently be purchased?

A. Availability

B. Brand

C. Product type

D. Image link

Correct Answer: A. Availability

Question 2

A campaign generates £80,000 in attributed revenue from £20,000 advertising cost. What is its revenue ROAS?

A. 0.25

B. 2.0

C. 4.0

D. 6.0

Correct Answer: C. 4.0

Question 3

Which measure most directly accounts for variable non-advertising costs?

A. CTR

B. Revenue ROAS

C. Contribution ROAS

D. Impression share

Correct Answer: C. Contribution ROAS

Question 4

What is the strongest reason to import qualified CRM outcomes into a local campaign?

A. To increase impressions automatically

B. To distinguish valuable leads from superficial interactions

C. To eliminate the need for geographic targeting

D. To guarantee store visits

Correct Answer: B. To distinguish valuable leads from superficial interactions

Question 5

Which app objective generally represents higher-quality acquisition than installs alone?

A. Ad impressions

B. App-store views

C. Activation or purchase event

D. Video starts

Correct Answer: C. Activation or purchase event

Question 6

What is a necessary condition for responsible campaign automation?

A. No conversion tracking

B. Unlimited budget flexibility

C. Reliable conversion signals and defined constraints

D. Manual adjustment of every auction

Correct Answer: C. Reliable conversion signals and defined constraints

Question 7

If a campaign receives 50,000 impressions and 1,000 clicks, what is its CTR?

A. 0.5%

B. 2%

C. 5%

D. 50%

Correct Answer: B. 2%

Question 8

Which pattern most strongly suggests a landing-page or conversion problem rather than an impression problem?

A. Few impressions and few clicks

B. Many impressions, many clicks and few conversions

C. No eligible products

D. Low geographic reach before launch

Correct Answer: B. Many impressions, many clicks and few conversions

Question 9

What does incrementality attempt to estimate?

A. The number of advertisements created

B. The additional outcomes caused by advertising

C. The total number of impressions purchased

D. The platform’s reported click count

Correct Answer: B. The additional outcomes caused by advertising

Question 10

Why should a campaign structure avoid unnecessary fragmentation?

A. Fragmentation always violates policy

B. It can prevent each segment from receiving sufficient learning data

C. It makes product titles longer

D. It eliminates the need for measurement

Correct Answer: B. It can prevent each segment from receiving sufficient learning data