Digital Commerce and Omnichannel Analytics

Introduction

Digital commerce has changed retail from a sequence of store-based transactions into a connected system of customer journeys, data-generating interactions and fulfilment choices. Customers may discover a product through search, compare it on a mobile application, inspect it in a store, purchase it online, collect it from a locker and return it through a physical outlet. For the retailer, these are not separate channels but interconnected points in a commercial system.

Omnichannel analytics is the disciplined use of data, models and managerial judgement to understand and improve these connected journeys. It extends conventional retail analytics in four important ways. First, it analyses the complete digital funnel rather than only completed transactions. Second, it links customer behaviour across devices and channels. Third, it evaluates fulfilment, delivery and returns as part of the value proposition. Fourth, it measures profitability at the customer, order and journey level rather than relying on sales revenue alone.

The central managerial challenge is that the channel generating the sale may not be the channel that created demand, influenced consideration or incurred the greatest cost. A digital order may be fulfilled by a store, a store visit may be initiated by online research, and a return may create a later profitable purchase. Sound analysis therefore requires integrated data, explicit causal assumptions and metrics that distinguish activity from value.

Learning objectives

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

1. Explain digital commerce funnels and diagnose conversion performance.
2. Evaluate search, recommendation and personalisation systems.
3. Compare attribution approaches and identify their limitations.
4. Analyse fulfilment, last-mile delivery and returns economically.
5. Measure channel migration and omnichannel profitability.
6. Integrate physical and digital customer journeys using identity, event and transaction data.
7. Assess privacy, fairness, governance and strategic risks in omnichannel analytics.

1. The omnichannel retail system

A multichannel retailer operates several channels, such as stores, websites, marketplaces and mobile applications. An omnichannel retailer attempts to make these channels coordinated from the customer’s perspective and economically coherent from the firm’s perspective. Coordination can involve shared inventory, consistent pricing, unified loyalty accounts, cross-channel service and flexible fulfilment.

A useful analytical representation is a journey rather than a channel list:

Figure: An omnichannel journey in which post-purchase outcomes feed future discovery and retention decisions.

The journey contains both customer-facing and operational events. A customer impression, search, product-view, basket addition, payment attempt, order allocation, shipment, delivery, review, return and subsequent purchase may all be analytically relevant. The objective is not to collect every possible event indiscriminately, but to capture events that support valid decisions.

Important terminology

A touchpoint is an interaction between customer and retailer. A conversion is a defined desired action, such as a purchase, registration or store visit. Channel migration is movement of a customer, order or activity from one channel to another. Cross-channel fulfilment occurs when the purchase channel and fulfilment channel differ. Showrooming describes examining merchandise in a store and purchasing elsewhere, often online. Webrooming describes researching online and purchasing in a store.

2. Digital funnel analytics

A digital funnel describes progressive stages from exposure to value creation. A typical retail funnel includes impression, visit, search, product detail view, basket addition, checkout initiation, payment completion, delivery and repeat purchase. Funnel analysis identifies where volume is lost, but it does not by itself explain why loss occurs.

For stage i, a basic transition rate is:

Here, n_i is the number of eligible users or sessions entering stage i, n_i+1 is the number reaching the next stage, and r_i is the transition rate. The measure assumes a clearly defined population, consistent observation window and reliable event ordering. If users can enter the funnel at different stages, the analyst must state whether the denominator is all visitors, qualified visitors or only users who reached the preceding stage.

The overall funnel conversion rate from stage 0 to stage k is:

The product relationship applies when stages are sequential and the numerators and denominators are consistently defined. It should not be used when a customer can bypass stages or when stages are measured using incompatible units, such as sessions at one stage and unique people at another.

Worked example: diagnosing a funnel

Suppose 100,000 qualified visits produce 8,000 product views, 2,400 basket additions, 1,200 checkout starts and 720 completed orders. The transition rates are:

The largest relative loss occurs before product viewing, suggesting problems with landing-page relevance, navigation, page speed, inventory visibility or traffic quality. The checkout completion rate is comparatively strong, so directing all improvement effort toward payment would be poorly justified. Funnel data identify the location of friction; qualitative research, experiments and operational data are needed to identify its cause.

Important diagnostic distinctions include users versus sessions, new versus returning visitors, mobile versus desktop, traffic source, product category, geography and stock availability. A high conversion rate may simply reflect a small group of loyal customers, while high traffic may be generated by low-intent visitors. Analysts should therefore report both volume and rate, with confidence intervals or statistical tests when comparing campaigns or experiments.

3. Search, discovery and recommendation analytics

Retail search

Search is both a customer interface and a demand signal. Search analytics should examine query volume, zero-result rate, reformulation rate, click-through rate, add-to-basket rate, conversion rate, revenue per search and margin per search. These metrics should be segmented by query type: branded, generic, product-specific, problem-oriented and misspelled queries.

A zero-result query can indicate absent assortment, poor synonym handling, an inventory problem or an unmet market need. However, forcing a result for every query may reduce trust. Search quality should combine relevance, availability and commercial outcomes. Ranking solely by revenue can disadvantage new products and create a feedback loop in which already popular products receive more exposure and therefore become more popular.

Search-engine marketing requires separation of paid and organic traffic, careful treatment of brand searches and incrementality testing. A sale following a paid search click is not necessarily caused by that click; the customer may already have intended to buy.

Recommendations and personalisation

Recommendation systems select or rank products using rules, collaborative filtering, content similarity, sequence models or hybrid methods. Common business metrics include recommendation click-through rate, add-to-basket rate, attach rate, incremental revenue, gross margin and customer satisfaction. Offline ranking metrics such as precision, recall or normalized discounted cumulative gain can help evaluate relevance, but they do not establish commercial incrementality.

A recommendation may increase basket size while reducing margin if it shifts customers toward discounted products. It may also increase short-term conversion while narrowing product discovery and reinforcing popularity bias. A robust evaluation therefore considers revenue, contribution margin, assortment exposure, customer experience and long-term retention.

The preferred evidence for a recommendation intervention is a randomised controlled experiment, such as an A/B test, provided randomisation is valid and interference between users is limited. If treatment assignment is not random, propensity scores, difference-in-differences or other causal methods may be considered, but their assumptions must be made explicit.

4. Attribution and causal measurement

Attribution assigns credit for an outcome to marketing or other touchpoints. Common approaches include:

Attribution is not the same as causation. A touchpoint may be correlated with purchase because high-intent customers seek it out. The fundamental causal quantity is incremental outcome: what would have happened without the intervention? This unobserved alternative is the counterfactual.

For a treatment group and a comparable control group, an experimental incremental effect can be estimated as:

where Y_T and Y_C are total outcomes in treatment and control, and N_T and N_C are their respective populations. The outcome may be orders, profit or customer value. The estimate assumes valid randomisation or exchangeability, consistent measurement and no material spillovers between groups.

If a campaign produces 1,200 orders among 100,000 treated customers and 1,000 orders among 100,000 control customers, the incremental order rate is 0.2 percentage points, or 2 additional orders per 1,000 customers. The campaign should not be judged on attributed orders alone; incremental contribution after media, fulfilment and incentive costs is the relevant managerial measure.

Multi-touch attribution remains useful as a descriptive allocation tool, especially for budgeting and reporting, but it should not be presented as proof that each credited touch caused the purchase. Geo experiments, holdout groups, conversion lift tests and carefully designed quasi-experiments are stronger approaches to causal evaluation.

5. Fulfilment and last-mile analytics

Customer experience and profitability are strongly affected by what happens after checkout. Relevant fulfilment measures include order cycle time, on-time delivery rate, promise accuracy, split-order rate, pick accuracy, cancellation rate, delivery cost per order, failed-delivery rate and click-and-collect readiness time.

A delivery promise is a probabilistic operational commitment, not merely a marketing statement. It depends on inventory accuracy, warehouse capacity, carrier reliability, distance, cut-off times and demand peaks. Promising speed without capacity can increase refunds, contacts and churn.

The last mile is often the most expensive and variable portion of delivery. Analysts should distinguish fixed and variable costs, first-attempt and reattempt costs, parcel density, route length, service level and customer willingness to pay. Delivery mode comparisons should include store collection, home delivery, lockers, crowd-shipping and carrier networks. A low monetary delivery cost may be offset by slower service or higher failure rates.

A basic contribution calculation for an order is:

P is collected selling price; D is discount and promotional cost; C_goods is product cost; C_fulfilment is picking and packing cost; C_delivery is delivery cost; C_payment is payment processing cost; C_service is customer-service cost; and C_returns is the expected return-related cost. All terms must use the same currency and order boundary. The formula is a contribution measure, not a full accounting profit measure, because overhead allocation, tax, depreciation and financing may be excluded.

For an order with selling price £80, discount £8, product cost £36, fulfilment £4, delivery £6, payment £2, service £1 and expected returns cost £5:

C_order = £80 − £8 − £36 − £4 − £6 − £2 − £1 − £5 = £18

If a store-collection option reduces delivery cost by £4 but adds £1 of store handling cost, its contribution becomes £21, assuming customer behaviour and other costs remain unchanged. This is an assumption to be tested, not an automatic conclusion: collection may also stimulate store purchases or impose congestion costs.

6. Returns analytics

Returns are simultaneously a cost, a service feature, a source of information and a possible route to future revenue. Return rate should be measured at item, order and customer levels because these denominators answer different questions:

A retailer should also measure return reason, time to return, resale status, markdown loss, transport cost, processing cost, refund time and subsequent purchase. Apparel may have high fit-related returns, while electronics may have defect or expectation-related returns. Aggregation can conceal actionable differences.

The expected return cost per order can be expressed as:

Here, p_r is the probability of a return; the C terms represent relevant costs; and V_recovery is recovered value from resale or refurbishment. The calculation assumes that costs are measured on a comparable expected-value basis. It should be expanded where multiple return outcomes have materially different probabilities.

Reducing returns through better size guidance, product imagery, reviews, quality control and accurate descriptions is generally preferable to simply restricting returns. Restrictive policies may reduce direct cost but damage conversion, trust, accessibility and customer lifetime value. Fraud analytics should be risk-based and subject to human review, appeal and fairness monitoring.

7. Channel migration and customer identity

Channel migration can be desirable or undesirable depending on economics and strategy. Moving a customer from store purchase to app purchase may lower transaction cost, but it may also eliminate valuable advice or impulse purchases. Moving an online customer to collection may reduce delivery cost and increase store traffic. The correct question is not “Which channel won?” but “What customer and firm value resulted from the journey?”

Analysis requires a persistent but privacy-respecting identity layer. Useful identifiers include loyalty account, authenticated login, consented email or phone token, order identifier and store transaction token. Device identifiers and cookies are less stable and may be restricted by browser, platform or regulation. Identity resolution should record confidence, provenance and consent status rather than silently treating uncertain matches as fact.

A unified customer data model should distinguish:

• Customer or household entity: a person or permitted household relationship.
• Event: an interaction with timestamp, channel, device context and consent status.
• Transaction: an order, line item, payment, fulfilment and return record.
• Location: store, warehouse, delivery address or collection point.
• Product: stable product, variant, category and inventory identifiers.
• Outcome: revenue, margin, satisfaction, retention or operational result.

Time alignment is critical. A store visit may precede an online order by days, while a return may occur weeks later. Analysts should define attribution windows and avoid assigning a later event to an earlier campaign merely because both appear in the same customer record.

8. Omnichannel profitability and customer value

Revenue is often misleading in omnichannel comparisons. A channel may have higher average order value but also higher discounts, delivery costs and returns. Profitability should be reported at several levels:

1. Order level: contribution after variable transaction costs.
2. Customer level: contribution over a defined period.
3. Journey level: value created across all touchpoints and fulfilment events.
4. Channel level: aggregate contribution after channel-specific costs and shared-cost treatment.

A simplified customer contribution over period t is:

The summation includes orders during the specified period. The formula requires a stated treatment of shared costs and does not automatically represent lifetime value. A forward-looking customer lifetime value estimate may discount expected future contribution:

E[C_t] is expected contribution in period t, d is the discount rate per period, T is the forecast horizon and CAC is customer acquisition cost. The model assumes that expected contributions, horizon and discount rate are defensible. It is highly sensitive to retention and margin assumptions; therefore scenario analysis is preferable to a single precise-looking number.

Omnichannel profitability also requires treatment of store-originated online orders, buy-online-pick-up-in-store transactions, ship-from-store orders and returns to store. If sales credit is assigned entirely to the website while store labour and inventory bear the cost, channel reports become internally inconsistent. Organisations should establish contribution rules, cost-sharing conventions and incentive systems that encourage total enterprise value rather than channel competition.

9. Integrating physical and digital journeys

Integration is organisational as well as technical. A retailer may possess data but still fail to create an omnichannel experience because merchandising, marketing, stores, supply chain and technology teams optimise different objectives. Effective integration normally includes:

• a shared product and location master;
• near-real-time inventory availability;
• common customer and consent controls;
• event and order data with consistent timestamps;
• cross-channel measurement definitions;
• experimentation capability;
• governance for model and metric changes;
• incentives based on enterprise contribution.

Physical behaviour can be measured through loyalty-linked transactions, appointment records, QR interactions, Wi-Fi or Bluetooth systems where lawful and consented, digital receipts and collection events. These data are informative but imperfect. A customer may share an account, disable location services or visit without interacting digitally. Absence of a digital trace is not evidence of absence of a store visit.

A practical integration process is to begin with a small number of high-value journeys: product discovery to store visit, online order to collection, store purchase to digital loyalty and return to repurchase. For each journey, define the customer objective, operational dependencies, data events, success measures and failure modes. This produces more reliable progress than attempting to integrate every journey simultaneously.

10. Artificial intelligence, experimentation and decision automation

Artificial intelligence is increasingly used for demand forecasting, search ranking, recommendations, offer selection, fraud detection, service automation and delivery routing. Its value depends on the decision context and feedback design. A model trained on historical purchases may reproduce existing assortment bias. A fraud model may over-flag particular customer groups. A recommender optimised for clicks may promote low-margin or unsuitable products.

Good practice includes:

• defining the business decision before choosing a model;
• separating training, validation and genuinely future test data;
• monitoring drift in demand, identity and channel behaviour;
• measuring business and customer outcomes, not merely model accuracy;
• retaining human escalation for high-impact decisions;
• documenting features, thresholds, owners and permitted uses;
• testing disparate error rates and accessibility impacts.

Experimentation should account for novelty effects, seasonality, interference, repeated exposure and long-term outcomes. A recommendation change that raises immediate clicks but increases returns may not be beneficial. Similarly, a delivery-fee experiment may affect both order conversion and basket composition. The experiment’s outcome set should therefore include contribution, service quality and customer experience.

11. Governance, ethics and regulation

Omnichannel analytics involves personal data, behavioural inference and potentially sensitive location information. Governance must address lawful basis, transparency, purpose limitation, data minimisation, retention, security, access rights and vendor accountability. Exact legal requirements differ by jurisdiction, so organisations need legal review rather than assuming that a general analytics practice is permissible.

Important risks include:

• Identity overreach: linking records beyond the customer’s reasonable expectations.
• Manipulative personalisation: exploiting vulnerability or obscuring prices.
• Discrimination: offering different service or prices through biased models.
• Surveillance: monitoring store movement without meaningful notice or consent.
• Dark patterns: making cancellation, privacy choices or returns unnecessarily difficult.
• Security exposure: creating concentrated datasets attractive to attackers.
• Metric gaming: rewarding attributed sales rather than incremental value.

Governance should include data dictionaries, access controls, model registers, consent records, audit trails, retention schedules and procedures for correction and appeal. Privacy-preserving aggregation, pseudonymisation and differential privacy may reduce exposure, but pseudonymisation is not equivalent to anonymisation and technical measures do not remove legal obligations.

12. Illustrative case: Northstar Home

Background

Northstar Home is a fictional mid-sized home-furnishings retailer with 45 stores, a website and a mobile application. Online sales grew rapidly, but online contribution margin declined. The company reported a strong last-click return on advertising spend, while stores argued that digital customers depended on store advice and inventory.

Problem

Management observed four symptoms: a rising checkout abandonment rate, high split shipments, expensive home delivery and increasing returns for selected furniture categories. Channel reports credited online sales to the website even when stores fulfilled orders or provided product consultations.

Analysis

The analytics team created a consented customer key, linked store transactions and online orders, and defined an order-level contribution measure. It found that last-click reporting overstated paid-search incrementality because many customers searched the brand immediately before purchase. A holdout test indicated that paid search generated fewer incremental orders than attributed reports suggested.

Funnel analysis showed that mobile product pages had strong basket-addition rates but unusually poor checkout completion. Qualitative testing identified address-entry difficulty and unclear delivery dates. Operational analysis showed that split shipments increased delivery cost and customer contacts. Store collection had lower variable cost and higher attachment of complementary items, but only when inventory availability was accurate.

Approach and solution

Northstar introduced clearer delivery promises, saved addresses for consented users, and a stock-confidence indicator. It tested store collection in selected regions rather than assuming universal benefit. Search ranking was modified to include availability and margin guardrails, while new products received controlled exploration exposure. Marketing budget decisions used holdout evidence alongside descriptive attribution.

Returns analysis led to improved room-dimension diagrams and delivery expectation content. The company did not simply shorten the return window because management recognised the effect on trust and conversion. Store managers received credit for assisted digital orders and collection outcomes, but the incentive was based on contribution rather than gross sales alone.

Outcome

In the illustrative case, checkout completion improved, split shipments declined and collection contribution exceeded home delivery contribution in the tested regions. The company also discovered that some store-originated online orders were highly profitable because they combined advice, accurate inventory and complementary purchases. The result was not a universal shift from store to digital; it was a more selective allocation of journeys to the channel configuration that created the greatest total value.

Learning points

1. Attribution can overstate the value of high-intent closing channels.
2. Fulfilment and returns must be included in digital profitability.
3. Store and digital channels may be complements rather than substitutes.
4. Tests should measure incrementality and contribution, not only conversion.
5. Governance and incentive design are prerequisites for sustainable integration.

Challenges and limitations

Omnichannel measurement faces persistent limitations. Data may be incomplete, delayed, duplicated or structurally biased toward identifiable customers. Cookies and device identifiers are unstable, while loyalty data may overrepresent frequent shoppers. Cross-device identity can create false certainty. Store visits without transactions are difficult to observe, and online-to-offline effects may occur outside the chosen attribution window.

Causal inference is difficult because customers self-select channels, promotions and fulfilment methods. Randomised experiments may be operationally expensive or contaminated by customer sharing. Long-term effects, such as trust or store habit, may not appear during a short test. Profit calculations also depend on cost allocation, inventory valuation and the treatment of shared infrastructure.

Managers should respond through measurement discipline: define denominators, document assumptions, report uncertainty, triangulate quantitative and qualitative evidence, test important interventions and avoid false precision. A dashboard is not a substitute for a causal design or a coherent decision process.

Contemporary developments and best practices

Current developments include retail media networks, connected commerce through marketplaces and social platforms, composable commerce architectures, real-time personalisation, electronic shelf labels, computer vision, autonomous fulfilment and increasingly sophisticated privacy controls. These developments create opportunities but also increase data fragmentation, platform dependence and governance complexity.

Best practice is to:

• begin with customer journeys and decisions, not available data;
• use consistent definitions for users, orders, returns, margin and channels;
• distinguish descriptive attribution from causal incrementality;
• evaluate search and recommendations on margin, satisfaction and long-term outcomes;
• include fulfilment, returns and service costs in contribution analysis;
• use controlled experiments where feasible;
• maintain transparent identity and consent practices;
• monitor model fairness, drift and operational failure;
• align incentives with total customer and enterprise value;
• communicate uncertainty to executives clearly.

Chapter Summary

Digital commerce and omnichannel analytics analyse connected customer journeys across discovery, evaluation, purchase, fulfilment, return and retention. Funnel metrics identify transition losses, but explanation requires segmentation, experimentation and operational evidence. Search and recommendation systems should be evaluated through relevance, availability, margin, customer experience and incremental outcomes rather than clicks alone.

Attribution is a descriptive allocation method unless supported by a credible causal design. Holdouts and controlled experiments provide stronger evidence of incremental effect. Fulfilment, last-mile delivery and returns are central components of profitability, and contribution must include their variable costs. Channel migration should be assessed as a change in total journey value, not as a contest between store and digital teams.

Integrated identity, event, product and transaction data enable analysis of physical and digital journeys, but identity uncertainty, privacy, bias and data quality must be governed explicitly. Artificial intelligence can improve search, recommendations, forecasting and operations when its objectives, feedback loops and fairness risks are carefully managed. The most effective omnichannel strategy combines customer-centred journey design, rigorous measurement, operational coordination and responsible governance.

Key Terms

Attribution: Allocation of observed outcomes to marketing or customer touchpoints.

Channel migration: Movement of customer activity, orders or fulfilment between channels.

Click-and-collect: Online ordering with collection from a store or designated point.

Contribution margin: Revenue remaining after specified variable costs.

Counterfactual: The outcome that would have occurred under an alternative condition.

Digital funnel: Sequential stages from digital exposure or visit to conversion and retention.

Incrementality: The additional outcome caused by an intervention relative to a valid comparison.

Identity resolution: Linking records believed to represent the same customer or entity.

Last mile: The final movement of an order to the customer or collection location.

Multichannel retailing: Operating more than one customer-facing sales or service channel.

Omnichannel retailing: Coordinating channels so that the customer experiences an integrated journey.

Personalisation: Adapting content, ranking, offers or service to customer or context data.

Recommendation system: Technology that selects or ranks products likely to be relevant to a customer.

Return rate: Returned units or orders divided by a defined sold-unit or order denominator.

Search zero-result rate: Proportion of recorded searches producing no available result.

Showrooming: Examining products in a physical store and purchasing elsewhere.

Touchpoint: An interaction between a customer and retailer.

Webrooming: Researching products online and purchasing in a physical store.

Review Questions

Short Answer Questions

1. Define omnichannel retailing and distinguish it from multichannel retailing.
2. Why should analysts distinguish users from sessions in funnel measurement?
3. State two limitations of last-touch attribution.
4. Identify four operational metrics relevant to last-mile delivery.
5. Why is identity resolution important in physical-digital journey analysis?

Long Answer Questions

1. Explain how a retailer should analyse a digital funnel from visit to repeat purchase.
2. Compare descriptive attribution with causal incrementality measurement.
3. Discuss how fulfilment and returns affect omnichannel profitability.
4. Evaluate the opportunities and risks of AI-based recommendations in retail.
5. Design a governance framework for privacy-respecting omnichannel analytics.

Analytical / Application Questions

1. A retailer has 200,000 qualified visits, 12,000 product views, 3,600 basket additions, 1,800 checkout starts and 1,080 orders. Calculate each transition rate and identify the most important diagnostic stage. State one limitation of drawing conclusions from these rates alone.
2. Design an experiment to determine whether paid brand search creates incremental orders. Specify the treatment, control, primary outcome, relevant costs and at least two threats to validity.
3. Compare home delivery and store collection for a product category using order contribution, delivery cost, handling cost, return behaviour and complementary store purchases. Recommend the data required before changing the default fulfilment option.

Multiple Choice Questions with Answers

Question 1

A funnel contains 100,000 visits and 720 completed orders. What is the visit-to-order conversion rate?

A. 0.072%

B. 0.72%

C. 7.2%

D. 72%

Correct Answer: B. 0.72%

Question 2

Which measure provides the strongest evidence that a marketing intervention caused additional orders?

A. Last-touch attributed orders

B. Total clicks

C. A randomised holdout comparison

D. Impressions divided by reach

Correct Answer: C. A randomised holdout comparison

Question 3

Which cost is included in the illustrative order contribution formula?

A. Only product cost

B. Only advertising cost

C. Expected return-related cost

D. Only warehouse rent

Correct Answer: C. Expected return-related cost

Question 4

A search query producing no available result is most directly measured by which metric?

A. Attach rate

B. Zero-result rate

C. On-time delivery rate

D. Customer acquisition cost

Correct Answer: B. Zero-result rate

Question 5

What is showrooming?

A. Researching online and buying in a store

B. Buying online and collecting from a locker

C. Examining products in a store and purchasing elsewhere

D. Returning an online order to a store

Correct Answer: C. Examining products in a store and purchasing elsewhere

Question 6

In the formula for incremental outcome, Δ = (Y_T ÷ N_T) − (Y_C ÷ N_C), what does N_C represent?

A. Total treatment outcome

B. Control-group population

C. Campaign cost

D. Number of channels

Correct Answer: B. Control-group population

Question 7

Why can last-click attribution overstate paid-search effectiveness?

A. It ignores all checkout events

B. It assigns credit to a channel that may be used by already high-intent customers

C. It measures only store transactions

D. It always excludes branded searches

Correct Answer: B. It assigns credit to a channel that may be used by already high-intent customers

Question 8

Which practice best supports responsible AI use in omnichannel retail?

A. Optimising only for clicks

B. Removing human review from high-impact decisions

C. Monitoring model drift and disparate error rates

D. Using every available personal attribute

Correct Answer: C. Monitoring model drift and disparate error rates

Question 9

If an order sells for £80 and total specified variable costs are £62, what is its contribution?

A. £18

B. £22

C. £62

D. £142

Correct Answer: A. £18

Question 10

Which statement about channel migration is most appropriate?

A. Every move from store to digital is beneficial

B. The channel making the sale always created the demand

C. Migration should be evaluated using total journey value and cost

D. Channel migration makes fulfilment irrelevant

Correct Answer: C. Migration should be evaluated using total journey value and cost