Digital Marketing and CRM Analytics

Introduction

Digital marketing and customer relationship management (CRM) generate large volumes of data: impressions, clicks, visits, leads, purchases, service interactions, repeat orders and cancellations. Analytics converts these observations into evidence for managerial decisions. It helps marketers determine which audiences to target, which channels to fund, where prospects leave the buying journey, and how customer relationships develop over time.

Analytics is not simply the production of attractive charts. It involves defining a business question, selecting valid measures, collecting reliable data, applying an appropriate analytical method, interpreting uncertainty and taking action. A high traffic volume may be commercially weak if visitors do not convert. Conversely, a small customer segment may be highly valuable if it produces strong retention and lifetime value.

This chapter examines key performance indicators (KPIs), marketing funnels, attribution, cohort analysis, dashboards, experimentation and customer economics. It also considers data quality, privacy, ethical use and the limitations of digital measurement.

Learning Objectives

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

1. Distinguish objectives, KPIs, metrics and dimensions.
2. Calculate and interpret funnel conversion, CAC, ROAS, retention and customer lifetime value.
3. Compare attribution approaches and recognise their assumptions.
4. Use cohort analysis to investigate retention and customer behaviour.
5. Design a useful management dashboard.
6. Explain the logic of A/B testing and identify common sources of error.
7. Communicate analytical findings as evidence-based recommendations.
8. Evaluate privacy, governance and ethical risks in CRM analytics.

1. From Business Objectives to KPIs

A measurement system should begin with a business objective rather than with the data that happens to be available. For example, “increase profitable repeat purchases among existing customers” is a strategic objective. Suitable KPIs may include repeat-purchase rate, retention rate, contribution margin per customer and customer lifetime value. Page views alone do not directly measure this objective.

A metric is a quantitative measure, such as orders or revenue. A dimension is an attribute used to classify data, such as channel, device, region or customer segment. A KPI is a metric judged to be especially important for a defined objective. A target is the desired level of a KPI within a specified period.

Good KPIs are relevant, clearly defined, consistently measured and actionable. They should also be interpreted with supporting measures. For example, conversion rate should be examined alongside traffic quality, average order value, refunds and profit. Optimising one isolated metric can produce harmful behaviour, such as buying cheap but unqualified traffic.

Leading and lagging indicators

Leading indicators provide early evidence of future performance. Examples include email engagement, product-page visits, trial activation and abandoned-cart recovery. Lagging indicators record outcomes, such as revenue, margin, retention and profit. A balanced measurement system uses both: leading indicators support timely intervention, while lagging indicators test whether activity produced commercial value.

Metric definitions and data contracts

An organisation should document each KPI’s numerator, denominator, population, time period, source system, exclusions and owner. For example, “monthly active customer” might mean a unique identified customer who completed a purchase during the calendar month. It should not silently change to include website visitors. Such documentation is often called a metric dictionary or data contract.

2. The Digital Marketing Funnel

The funnel represents the movement from exposure to relationship. A common sequence is:

Reach → engagement → consideration → conversion → retention → advocacy

A funnel is a useful analytical model, not a claim that every customer follows a strictly linear path. Customers may compare products, return through several devices, speak to staff or purchase without interacting with an advertisement.

For a stage, the stage conversion rate is:

The denominator must be defined carefully. If 10,000 landing-page sessions produce 400 leads, landing-page lead conversion is 400 ÷ 10,000 × 100% = 4%. If the organisation counts unique people rather than sessions, the result may differ.

If stages are sequential and the same population is used, overall conversion can be expressed as:

When rates are written as decimals, for example 0.40 and 0.05, overall conversion is 0.40 × 0.05 = 0.02, or 2%. This multiplication is appropriate only when the stages are defined consistently and progression is sequential. In real-world reporting, overlapping audiences and repeated sessions often make a simple product inappropriate.

Funnel analysis should identify both the largest volume loss and the most economically important loss. A checkout stage may have fewer visitors than an awareness stage but still deserve priority because improvements there affect high-intent users.

3. Core Marketing and CRM Metrics

Traffic and engagement

Common traffic measures include users, sessions, source or medium, landing-page views and cost per visit. Engagement measures may include video completion, content downloads, product comparisons, email clicks and time spent. These measures are diagnostic rather than automatically valuable. For instance, a long time on a page may indicate interest or confusion.

Click-through and conversion rate

CTR describes the proportion of recorded impressions that generated clicks. It does not measure post-click quality.

“Eligible opportunities” may be sessions, users, leads or ad clicks. Reports must state which definition is used. A conversion should also have a clear event definition, such as completed payment rather than an add-to-cart event.

Customer acquisition cost

CAC must specify the period, channels, costs included and customer definition. If £12,000 of eligible marketing and sales expenditure acquires 300 new customers, CAC is £40 per customer. Including retention spending in acquisition cost without explanation reduces comparability.

Return on advertising spend

ROAS is a ratio, not a percentage unless multiplied by 100. Advertising spend of £8,000 producing £24,000 attributed revenue gives ROAS = 3.0, or £3 revenue per £1 of advertising. ROAS does not subtract product cost, fulfilment, salaries, discounts or returns; therefore, a high ROAS can still be unprofitable.

Profit and contribution

A more decision-relevant measure is contribution after variable costs:

Fixed costs and accounting treatment should be specified. A campaign may be acceptable at a lower ROAS if it attracts customers with strong repeat purchases, while a high first-order ROAS may be unattractive when refunds are common.

Retention and churn

For a defined customer cohort and period:

If 1,000 customers begin a month and 760 remain active according to the stated activity rule, retention is 76%.

Retention and churn are complements only when the population is stable and “lost” is precisely defined. In subscription services, churn may mean cancellation. In retail, a customer who has not purchased for 30 days may not truly have churned.

Customer lifetime value

Customer lifetime value (CLV or LTV) estimates the economic value of a customer relationship. A simple historical estimate is:

For a simplified subscription model with constant monthly contribution margin, constant monthly retention probability and no discounting:

The formula assumes a stable churn rate, constant contribution, an indefinite relationship and no discounting. If monthly contribution is £18 and monthly churn is 0.06, estimated CLV is £18 ÷ 0.06 = £300. This is an expectation, not a guarantee for every customer.

A discounted model is more realistic when timing matters:

Here, t is the period, T is the forecast horizon, and d is the discount rate per period. The contribution should be net of variable costs, and the model should state whether acquisition cost is included. Forecast CLV is highly sensitive to retention assumptions, so scenario analysis is preferable to a single precise-looking number.

A useful relationship is the CLV-to-CAC comparison. If expected CLV is £300 and CAC is £40, the ratio is 7.5. However, a high ratio may reflect underinvestment in growth or optimistic forecasting. Managers should consider payback period, cash flow, margin and strategic value as well.

4. Attribution: Connecting Activity to Outcomes

Attribution assigns credit for a conversion to marketing touchpoints. It is needed because customers may see a social advertisement, click a search result, read an email and later purchase directly.

Common models include:

Attribution is not the same as causation. A person who was likely to purchase may click a branded advertisement shortly before buying. Giving that advertisement credit does not prove that it caused the purchase. Incrementality experiments, holdout groups and geo-experiments can provide stronger evidence.

Tracking is complicated by cookie restrictions, consent choices, ad blockers, cross-device behaviour, offline sales and platform-specific definitions. Reported revenue can therefore differ between an advertising platform, web analytics system and finance ledger. Organisations should establish a source-of-truth hierarchy and reconcile differences rather than treating any platform’s number as unquestionably correct.

5. Cohort Analysis

A cohort is a group sharing a defined starting characteristic, commonly the month of first purchase, subscription or activation. Cohort analysis compares similar customers over time and avoids confusing growth in customer acquisition with improvement in customer quality.

A retention table may show the percentage of each acquisition cohort active in months 0, 1, 2 and so forth. For example, if the January cohort contains 500 customers and 300 purchase again in month 1, month-1 retention is 300 ÷ 500 × 100% = 60%. A February cohort should be compared at the same customer age, not merely by calendar month.

Cohorts can be defined by channel, campaign, product, geography, device or customer need. Useful questions include:

• Do customers acquired through discount advertising retain less well?
• Did a product or onboarding change improve month-2 retention?
• Which channels produce the greatest contribution after 90 days?
• Are high-value customers concentrated in a particular segment?

Cohort analysis requires a stable event definition, correct time zones, deduplicated customer IDs and an explicit treatment of refunds, returns and reactivations. Small cohorts can produce volatile percentages, so counts should be shown alongside rates.

6. Dashboards for Decision-Making

A dashboard is a visual interface for monitoring defined measures. It should support a specific audience and decision rhythm. A chief marketing officer may need channel investment, CAC, contribution and forecast CLV. A campaign manager may need spend pacing, reach, CTR, conversion and creative performance. A CRM manager may need activation, retention, churn, repeat purchase and service contacts.

An effective dashboard normally includes:

1. Headline KPIs: current result, target and comparison period.
2. Trend views: daily, weekly or monthly movement.
3. Segment views: channel, audience, product, region or device.
4. Funnel view: volume and conversion at each stage.
5. Diagnostic details: anomalies, data freshness and definitions.
6. Action area: owner, recommended action and review date.

Use rates and volumes together. A 20% conversion rate from 50 visitors may be less important than a 5% rate from 20,000 qualified visitors. Avoid excessive colour, three-dimensional charts and unexplained benchmarks. A dashboard should distinguish actual, estimated and forecast data and should display the period and denominator.

A practical dashboard governance process assigns an owner to every KPI, records data refresh time, monitors missing values and reconciles revenue with finance. Access controls should prevent unauthorised exposure of personal data.

7. A/B Testing and Experimentation

An A/B test randomly assigns eligible units—such as users, accounts or geographic areas—to a control version and a treatment version. Only the planned element should differ materially. The purpose is to estimate whether the treatment changes a defined outcome.

A basic process is:

1. State a hypothesis and the decision it will inform.
2. Define the eligible population, primary outcome and guardrail metrics.
3. Randomise units and determine the allocation ratio.
4. Run the test for a justified period covering normal behavioural cycles.
5. Check implementation, exposure and sample integrity.
6. Analyse the pre-specified outcome and report uncertainty.
7. Decide whether to implement, iterate or reject the treatment.

A conversion-rate difference can be described as:

If control conversion is 4% and treatment conversion is 5%, absolute lift is 1 percentage point and relative lift is 25%. These are different statements and should not be confused.

Randomisation reduces systematic differences on average, but it does not guarantee perfect balance in a small sample. Tests can be invalidated by running multiple variants, stopping when a favourable result appears, changing the outcome after viewing results, interference between users, or allowing the same person to enter both groups. Statistical significance alone does not establish commercial importance; effect size, confidence interval, cost and customer experience also matter.

Guardrail metrics can prevent local optimisation. For example, a discount banner may increase conversion but reduce margin, increase returns or damage long-term retention. Where randomised testing is impossible, quasi-experimental methods may be useful, but their assumptions must be stated.

8. Communicating Insights

An analytical communication should move from evidence to meaning to action. A concise structure is:

• Finding: What changed, for whom and over what period?
• Evidence: Which KPI, comparison and sample support the finding?
• Interpretation: What are the plausible explanations and limitations?
• Recommendation: What action should be taken, by whom and when?
• Measurement plan: How will the action be evaluated?

For example: “The March email cohort generated 12% higher 60-day contribution than the February cohort, while open rates were similar. The difference is concentrated among customers receiving product-use guidance. Because the cohort is observational, causation is not established. Test the guidance sequence against the existing sequence for four weeks, with contribution and unsubscribe rate as primary and guardrail outcomes.”

Analysts should avoid causal language when the design supports only association. They should report denominators, time windows, uncertainty and commercially relevant effects. A useful recommendation is specific, feasible and linked to an owner.

9. Applied Case: Illustrative Online Retailer

Background

GreenBasket, an illustrative online grocery retailer, spends £30,000 on paid social and search during one month. It records 20,000 landing-page sessions, 1,600 add-to-cart sessions, 800 completed first orders and 600 identifiable new customers after deduplication. First-order revenue is £72,000, and variable product, delivery and payment costs are £50,400.

Problem

Management wants to know whether the campaign should be expanded. The platform reports strong ROAS, but finance is concerned about margin and the quality of newly acquired customers.

Analysis

The landing-page-to-cart rate is 1,600 ÷ 20,000 × 100% = 8%. The cart-to-order rate is 800 ÷ 1,600 × 100% = 50%. Using the 600 deduplicated new customers, CAC is £30,000 ÷ 600 = £50. Attributed first-order ROAS is £72,000 ÷ £30,000 = 2.4.

Contribution before acquisition expenditure is £72,000 − £50,400 = £21,600. After campaign expenditure, first-order contribution is £21,600 − £30,000 = −£8,400. Thus, the campaign loses £14 per acquired customer on the first order (£8,400 ÷ 600), even though ROAS is 2.4.

A cohort analysis shows that customers receiving a replenishment reminder have 45% 60-day retention, compared with 32% for similar customers who do not receive it. This is an association, not proof that the reminder caused the difference. The company should test the reminder using random assignment, while monitoring unsubscribes and contribution.

Approach and solution

GreenBasket should not scale solely on first-order ROAS. It should segment results by channel and cohort, calculate contribution after variable costs, reconcile platform conversions with deduplicated customers, and estimate CLV using conservative retention scenarios. It should also test checkout improvements because the cart-to-order stage is a substantial loss point.

Outcome and learning points

The appropriate management decision is conditional expansion: continue channels whose forecast contribution exceeds the company’s payback requirement, pause low-quality segments, and run controlled retention and checkout experiments. The case demonstrates that revenue attribution can overstate value, deduplication changes CAC, and retention analysis may justify investment that first-order metrics reject.

10. Limitations, Ethics and Governance

Digital analytics can reproduce bias when historical data reflects unequal access, discriminatory targeting or unequal service. Automated customer scoring may disadvantage individuals using shared devices, privacy tools or incomplete profiles. Sensitive attributes should not be inferred casually or used without a lawful and ethical purpose.

Organisations should collect only necessary data, communicate purposes clearly, obtain consent where required, honour opt-outs, protect data and establish retention periods. Applicable privacy and electronic-marketing rules depend on jurisdiction, but responsible practice generally includes lawful processing, data minimisation, access control, security, correction rights and careful vendor management.

Identity resolution creates particular risks. Linking website behaviour, purchases, email activity and offline records can reveal detailed personal profiles. Pseudonymisation reduces exposure but does not automatically make data anonymous. AI-based propensity and recommendation systems should be tested for bias, monitored for drift and subject to human oversight where decisions materially affect customers.

Other limitations include missing conversions, duplicate records, changing platform algorithms, seasonality, survivorship bias, attribution overlap, small samples and metric manipulation. Analytics should therefore be treated as structured evidence rather than an infallible description of reality.

Contemporary Developments and Best Practice

Privacy-enhancing measurement is becoming more important as third-party identifiers decline. First-party data, consented customer databases, clean rooms, server-side measurement and aggregated conversion reporting can support analysis while reducing unnecessary exposure. These approaches do not remove the need for governance or guarantee causal accuracy.

Best practice is to:

• Begin with a decision and business objective.
• Define events, populations, windows and exclusions before reporting.
• Reconcile analytics data with finance and CRM records.
• Use cohort and contribution analysis alongside acquisition metrics.
• Treat attribution as a model, not as proof of causality.
• Prefer experiments or credible quasi-experiments for causal claims.
• Show counts, rates, targets, trends and uncertainty together.
• Protect personal data and document access and retention rules.
• Communicate one clear recommendation with an owner and evaluation plan.

Chapter Summary

Digital marketing and CRM analytics connect customer behaviour with commercial decisions. KPIs must be derived from objectives and defined through consistent numerators, denominators, populations and time periods. Funnel analysis locates leakage, while CAC, ROAS and contribution evaluate acquisition economics. Retention and CLV extend analysis beyond the first transaction, although CLV forecasts depend strongly on assumptions.

Attribution helps organise touchpoint data but does not establish causality. Cohort analysis compares customers at equivalent relationship ages and reveals differences in retention and value. Dashboards should be audience-specific, governed and designed to support action. A/B testing provides stronger evidence when randomisation, pre-specified outcomes and guardrails are used. Finally, privacy, fairness, data quality and transparent communication are essential to responsible analytics.

Key Terms

KPI: A measure judged critical to a defined business objective.

Metric: A quantitative measure of an activity or outcome.

Dimension: An attribute used to classify or segment data.

Funnel: A model showing progression through marketing and customer stages.

Conversion rate: The proportion of eligible opportunities producing a defined conversion.

CAC: Acquisition expenditure divided by new customers acquired.

ROAS: Attributed revenue divided by advertising expenditure.

Contribution: Revenue remaining after specified variable costs.

Retention: The proportion of a starting customer population remaining active.

Churn: The proportion of customers lost during a defined period.

CLV: Estimated economic value of a customer relationship.

Attribution: Assignment of conversion credit to marketing touchpoints.

Cohort: A group sharing a defined starting characteristic.

Dashboard: A visual interface for monitoring measures and supporting decisions.

A/B test: A randomised comparison of control and treatment conditions.

Guardrail metric: A measure monitored to prevent harmful side effects.

Incrementality: The additional outcome caused by an intervention.

Data minimisation: Collecting and retaining only data necessary for a legitimate purpose.

Metric dictionary: Documentation of definitions, sources, owners and calculation rules.

Review Questions

Short Answer Questions

1. Distinguish a metric from a KPI.
2. Why can a high ROAS campaign still be unprofitable?
3. What is the purpose of cohort analysis?
4. State two limitations of last-touch attribution.
5. What is a guardrail metric in an A/B test?

Long Answer Questions

1. Explain how a marketing manager could design a KPI framework from a business objective.
2. Compare first-touch, last-touch, linear and data-driven attribution models.
3. Explain the assumptions behind the simplified CLV formula.
4. Describe the design features of an effective management dashboard.
5. Discuss the principal ethical and governance risks of CRM analytics.

Analytical / Application Questions

1. A campaign spends £12,000, acquires 240 new customers and generates £36,000 attributed revenue. Calculate CAC and ROAS, then explain why these figures are insufficient for a profitability decision.
2. A control email converts 400 of 10,000 recipients and a treatment email converts 460 of 10,000 recipients. Calculate each conversion rate, the absolute lift and the relative lift. Identify one additional metric that should be monitored.
3. A retailer observes higher 90-day retention among customers receiving personalised recommendations. Design an analysis or experiment that would better determine whether the recommendations caused the difference.

Multiple Choice Questions with Answers

Question 1

Which measure is calculated as attributed revenue divided by advertising spend?

A. CAC

B. ROAS

C. Retention rate

D. CTR

Correct Answer: B. ROAS

Question 2

A campaign spends £10,000 and acquires 200 new customers. What is CAC?

A. £20

B. £50

C. £200

D. £2,000

Correct Answer: B. £50

Question 3

Which method compares customers at the same age since acquisition?

A. Last-touch attribution

B. Funnel analysis

C. Cohort analysis

D. Brand tracking

Correct Answer: C. Cohort analysis

Question 4

A control conversion rate is 4% and a treatment rate is 5%. What is the relative lift?

A. 1%

B. 5%

C. 20%

D. 25%

Correct Answer: D. 25%

Question 5

What is a principal limitation of last-touch attribution?

A. It gives equal credit to all touches

B. It ignores the final conversion

C. It may overcredit the closing channel

D. It requires randomisation

Correct Answer: C. It may overcredit the closing channel

Question 6

Which statement best describes a KPI?

A. Any raw data point

B. A critical measure linked to an objective

C. A customer identity record

D. A visual chart type

Correct Answer: B. A critical measure linked to an objective

Question 7

If monthly contribution is £18 and monthly churn is 0.06, what is simplified expected CLV under the stated assumptions?

A. £30

B. £108

C. £300

D. £1,080

Correct Answer: C. £300

Question 8

What is the strongest basic feature of a valid A/B test?

A. Random assignment to conditions

B. Changing several variables simultaneously

C. Stopping whenever results look favourable

D. Using only the largest observed segment

Correct Answer: A. Random assignment to conditions

Question 9

Why should a dashboard show counts alongside conversion rates?

A. Rates are never useful

B. Counts provide context for the size of the evidence

C. Counts automatically prove causation

D. Rates cannot be calculated from counts

Correct Answer: B. Counts provide context for the size of the evidence

Question 10

Which statement about attribution is most accurate?

A. Attribution always proves causation

B. Attribution is unnecessary when customers use multiple channels

C. Attribution allocates credit according to a selected model

D. Attribution can only be performed for offline sales

Correct Answer: C. Attribution allocates credit according to a selected model