Customer Economics and Relationship Analytics

Introduction

Brand management is not only the management of awareness, associations, and identity. It is also the management of economically valuable customer relationships. A brand creates value when customers choose it, buy it repeatedly, remain active, recommend it, and contribute revenue and profit over time. Relationship analytics converts these behavioural patterns into evidence for customer relationship management (CRM), loyalty design, personalization, and resource allocation.

Customer economics asks a central question: Which customers, behaviours, and relationship interventions create profitable long-term value? This question requires more than counting sales. Revenue may be generated by customers who are expensive to acquire, frequently discounted, costly to serve, or likely to leave. Accordingly, managers need measures that connect customer behaviour to contribution margin, retention, and future cash flows.

This chapter develops the principal measures used in customer economics: customer acquisition cost, retention, churn, repeat purchase, share of wallet, and customer lifetime value. It then introduces RFM analysis, propensity modelling, next-best-action decisions, and the ethical and operational conditions required for responsible use.

Learning objectives

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

1. distinguish revenue, gross margin, contribution margin, and customer profitability;
2. calculate and interpret acquisition cost, retention, churn, repeat purchase, and share of wallet;
3. calculate customer lifetime value under simple and discounted models;
4. apply RFM and propensity concepts to segmentation and CRM decisions;
5. evaluate loyalty, personalization, and next-best-action strategies;
6. identify measurement, attribution, privacy, fairness, and governance risks.

1. The economic logic of customer relationships

A customer relationship has an economic flow. The firm incurs costs to attract and convert a prospect; the customer generates transactions; the firm incurs variable fulfilment and service costs; and the relationship may continue, expand, contract, or end. The relevant unit of analysis is often the customer or household, although business-to-business firms may analyse accounts, buying centres, or contracts.

Revenue and contribution

Customer revenue is the value of purchases during a period. Revenue alone is insufficient because two customers with identical revenue may have different margins, return rates, service requirements, payment costs, and promotional costs.

A useful contribution calculation is:

where:

• CM = contribution margin attributable to the customer;
• R = customer revenue;
• VC = variable product or service cost;
• SC = variable selling, fulfilment, and service costs;
• PC = customer-specific promotional or incentive cost.

All terms must be measured in the same currency and period. The formula is most useful for short-run customer profitability. Fixed corporate costs should not be deducted if the purpose is to compare incremental customer economics; they may be allocated separately for full-cost profitability analysis.

Example. A customer spends $240 in a quarter. Product cost is $120, delivery and payment costs are $18, and discounts total $24. Contribution is:

CM = 240 − 120 − 18 − 24 = $78.

The customer is economically attractive only if this contribution is sufficient to cover acquisition cost and any appropriate share of relationship-management costs.

Customer revenue cohorts

A cohort is a group of customers sharing a defining start condition, such as acquisition month, channel, campaign, geography, or product. Cohort analysis compares customers acquired under similar conditions over time. It can reveal that a low-cost social campaign produces many customers but weak retention, while a more expensive referral channel produces fewer but more valuable customers.

A cohort table should normally report, by period:

• customers acquired;
• active customers;
• revenue and contribution;
• repeat-purchase rate;
• retention and churn;
• refunds and service costs;
• cumulative acquisition and relationship costs.

Cohort analysis is preferable to a single blended average when channel mix, prices, customer composition, or economic conditions are changing.

2. Acquisition economics

Customer acquisition cost

Customer acquisition cost (CAC) estimates the average cost of obtaining a new customer during a defined period.

where:

• M = marketing media and campaign cost;
• S = sales compensation and selling cost;
• T = technology, agency, and marketing-operations cost included by policy;
• O = other attributable acquisition costs;
• N = number of new customers acquired in the same period.

The numerator and denominator must use the same time period and attribution rule. If the objective is incremental CAC, the numerator should include incremental spending and the denominator should be incremental customers caused by that spending. If all marketing costs are included, the result is an average blended CAC rather than a channel-specific CAC.

Example. A campaign costs $48,000 in media, $12,000 in sales support, and $5,000 in technology. It produces 1,300 new customers:

CAC = (48,000 + 12,000 + 5,000) ÷ 1,300 = $50.

CAC must be interpreted with quality metrics. A low-CAC customer who never repurchases may be less valuable than a high-CAC customer with strong retention.

Payback period

The payback period estimates how long it takes cumulative contribution to recover acquisition cost. If monthly contribution is constant and positive:

This simplified formula assumes stable contribution, no discounting, and no interim churn. For variable monthly contributions, managers should calculate cumulative contribution month by month and identify the first month in which it equals or exceeds CAC.

A useful decision rule is not that every customer must repay CAC immediately, but that expected discounted customer value should exceed acquisition and relationship costs, subject to liquidity and risk constraints.

3. Retention, churn, and repeat purchasing

Retention and churn

Retention measures the proportion of customers who remain active over a specified interval. For a beginning customer base:

where:

• B = customers active at the beginning of the period;
• A = new customers acquired during the period;
• E = customers active at the end of the period.

Subtracting new customers prevents acquisition from being mistaken for retention. This formula assumes that the beginning and ending populations use the same definition of “active.”

For a contractual subscription with no reactivation during the interval, churn is often:

If retention is 92%, churn is 8%. In non-contractual settings, churn requires an inactivity rule, such as “no purchase for 180 days.” That rule is a managerial convention and should be validated against normal purchase cycles.

Retention may be measured by customer count, revenue, units, or contribution. Revenue retention can be higher or lower than logo or customer retention because retained customers may expand or reduce spending.

Repeat-purchase rate and purchase frequency

Repeat-purchase rate is the proportion of customers who make at least a second purchase within a defined observation window:

The denominator must be eligible for observation. A customer acquired yesterday should not be treated as a failed repeat purchaser if the normal repurchase cycle is six months.

Purchase frequency can be expressed as:

Order frequency and repeat-purchase rate are related but not identical. A customer who purchases twice contributes to repeat-purchase rate, while frequency captures the number of orders across the active population.

Share of wallet

Share of wallet estimates the proportion of a customer’s category spending captured by a brand:

If a customer spends $180 on a brand and an estimated $600 across the category, share of wallet is 30%. The denominator is often difficult to observe. It may be estimated from customer surveys, panels, open banking data where permitted, category benchmarks, or inferred behaviour. Therefore, share-of-wallet figures should be reported with a data-quality assessment rather than treated as exact facts.

Share of wallet supports cross-selling and retention decisions, but a low share does not automatically justify more promotion. The brand must assess needs, profitability, competitive intensity, and the risk of encouraging unnecessary consumption.

4. Customer lifetime value

Customer lifetime value (CLV) is the expected present value of future customer contribution, net of relevant future costs. It is a forward-looking expectation, not a guaranteed amount.

Simple perpetual CLV model

Under strong simplifying assumptions—constant periodic contribution, constant retention probability, no discounting, and an indefinite relationship—the expected lifetime contribution is:

where m is contribution per period and r is the probability of retention from one period to the next. This is a special case of the general model and should not be used when contribution, retention, or discount rates vary materially.

Example. Monthly contribution is $20 and monthly retention is 90%:

CLV = 20 × 0.90 ÷ (1 − 0.90) = $180.

The result is expected future contribution, not revenue. If acquisition cost is $50, the simplified net value before other costs is $130.

Discounted finite-horizon CLV

A more general model discounts expected contribution over a finite horizon:

where:

• AC = acquisition cost, measured at time zero;
• T = number of future periods;
• m_t = contribution if active in period t;
• S_t = probability of surviving or remaining active through period t;
• d = per-period discount rate;
• t = period number.

The discount rate and contribution period must match. A monthly model requires a monthly discount rate; an annual model requires an annual rate. If S_t already includes predicted survival, it should not be multiplied by retention again.

For a constant retention probability r, S_t = r^t when survival through period t requires retention at each transition. In a contractual model, the analyst may estimate S_t from survival analysis rather than assume a constant rate.

Worked example. Suppose acquisition cost is $50, contribution is $18 per month, monthly survival probabilities are 0.90, 0.81, and 0.729 for months 1–3, and the monthly discount rate is 1%:

CLV₃ = −50 + (18×0.90/1.01) + (18×0.81/1.01²) + (18×0.729/1.01³).

The three discounted contributions are approximately $16.04, $14.27, and $12.85. Therefore:

CLV₃ ≈ −50 + 16.04 + 14.27 + 12.85 = −$6.84.

Under this three-month horizon, the relationship does not recover acquisition cost. Extending the horizon could produce positive value, but only if later contributions and survival assumptions are credible.

CLV limitations

CLV is sensitive to assumptions about retention, margin, discounting, returns, service costs, and future behaviour. Small errors in retention can create large errors in long-run value. CLV should therefore be presented with scenarios or confidence intervals where possible:

• conservative case: lower retention and margin;
• base case: most defensible estimates;
• optimistic case: stronger retention and expansion.

CLV should guide decisions, not justify treating customers merely as financial assets. Strategic value, referrals, learning, brand legitimacy, and community effects may be important but difficult to quantify.

5. RFM and behavioural segmentation

RFM analysis ranks customers on three behavioural dimensions:

• Recency: how recently the customer purchased;
• Frequency: how often the customer purchased during a defined period;
• Monetary value: how much the customer spent or contributed.

A common implementation assigns each dimension a score from 1 to 5 using quantiles. A customer with scores 5-5-5 is recent, frequent, and high value; a customer with 1-1-1 is inactive and low value. The scoring direction must be documented: recent purchases should receive higher recency scores, while long inactivity receives lower scores.

RFM is attractive because it is understandable, inexpensive, and useful when rich individual-level data are unavailable. It can support segments such as:

RFM is descriptive, not causal. A high score does not prove that a discount caused loyalty, and a low score does not prove that a customer will respond to a win-back message. RFM also ignores product category, customer need, channel preferences, and profitability unless these are added separately.

6. Propensity, response, and next-best action

A propensity score estimates the likelihood of an event, such as purchase, churn, response to an offer, or adoption of a service. Formally, a purchase propensity may be represented as:

where Y_i = 1 denotes the target event for customer i, and X_i is the customer’s available information, such as recency, product usage, channel engagement, tenure, and service history.

A score is useful only when it is calibrated and connected to a decision. A customer with a 0.70 predicted purchase probability should purchase approximately 70% of the time among comparable customers if the model is well calibrated. Ranking alone is insufficient for probability interpretation.

Uplift and incremental response

A response model predicts what customers may do; an uplift model estimates how an intervention changes behaviour. This distinction is essential. Customers likely to purchase without a message are “sure things,” while customers whose purchase is caused by the message are more valuable intervention targets. Some customers may be “sleeping dogs” who are annoyed or made less likely to purchase by contact.

The incremental effect can be estimated using a randomized treatment and control design:

Randomization supports causal interpretation under proper implementation. Observational comparisons can be confounded because targeted customers may already differ from non-targeted customers.

Next-best action

Next-best-action systems select an action—such as a product recommendation, service intervention, reminder, loyalty benefit, or no contact—based on predicted customer outcomes and constraints. A simplified expected incremental value is:

where p_T is the predicted probability under treatment, p_C is the probability under control, CM is incremental contribution from the outcome, and action cost includes incentive, contact, fulfilment, and potential service costs. This formula assumes the probabilities and contribution estimate refer to the same decision horizon.

The “no action” option is important. Contact fatigue, discount dependency, privacy risk, and brand damage can make silence the best action.

7. CRM, loyalty, and personalization decisions

Customer economics supports four connected management activities.

CRM prioritization

CRM teams can prioritize customers by expected value, risk, need, and strategic importance. High-value customers at risk may receive human service recovery. Low-value customers may receive automated education rather than expensive personal intervention. The objective is not to maximize contacts; it is to maximize mutually beneficial relationship value.

Loyalty design

A loyalty programme should create incremental behaviour rather than merely reward behaviour that would have occurred anyway. Evaluation should compare members with a valid control or matched non-member group and examine contribution after rewards, administration, breakage, redemption costs, and liability. Points can increase frequency while reducing margin if rewards are too generous.

Personalization

Personalization may use recommendations, content, timing, channel, or offer design. Effective personalization requires relevance, consent where required, accurate data, frequency controls, and an explanation of value to the customer. Personalization should not rely on sensitive attributes or hidden inferences in ways that create unfair treatment.

Retention intervention

A retention offer should be targeted at customers with both meaningful churn risk and a plausible reason to stay. Giving discounts to customers who would remain anyway wastes margin; giving discounts to customers who cannot be profitably retained also wastes resources. Service improvement may be more effective than a price incentive when churn is caused by delivery failures or poor product performance.

8. Data, measurement, and implementation

A reliable relationship analytics programme requires a customer data model linking customer identifiers, transactions, products, channels, contacts, service events, costs, permissions, and outcomes. Identity resolution must avoid merging different people or splitting one person across multiple accounts.

A practical implementation sequence is:

1. define the customer, active status, purchase window, and profit measure;
2. establish data quality rules and a documented metric dictionary;
3. calculate descriptive cohort, RFM, and profitability measures;
4. build and validate propensity or churn models;
5. design randomized tests for interventions;
6. connect scores to operational actions and suppression rules;
7. monitor outcomes, fairness, calibration, drift, and customer complaints.

Attribution requires particular care. Last-click attribution assigns all credit to the final touchpoint and may overvalue retargeting. Multi-touch models distribute credit but rely on assumptions. Incrementality testing is stronger for causal decisions, though it may be more expensive and operationally complex.

9. Ethics, privacy, and governance

Relationship analytics uses personal data and can affect prices, access, communication, and service. Governance should address:

• lawful basis, consent, purpose limitation, and data minimization;
• transparency about profiling and personalization;
• access, correction, deletion, and preference controls where applicable;
• security, retention periods, and third-party data contracts;
• fairness testing across relevant customer groups;
• human review for consequential decisions;
• audit trails for model versions, data, and actions.

A model can be statistically accurate yet ethically unacceptable. For example, a churn model may use a proxy for socioeconomic status and lead to systematically inferior service. Fairness assessment should examine both error rates and outcomes, while recognizing that fairness criteria can conflict and require contextual judgment.

AI can improve recommendation, forecasting, segmentation, and natural-language service, but it can also amplify biased data, generate inappropriate offers, expose personal information, or make decisions that employees cannot explain. Human accountability remains necessary. Models should be monitored for drift because customer behaviour, prices, channels, and economic conditions change.

10. Illustrative case: Aurora Home subscriptions

Background

Aurora Home is a fictional direct-to-consumer household-products brand. It acquires subscribers through paid search, referrals, and social media. The company reports strong first orders but is uncertain whether growth creates profit.

Problem

In one quarter, the company spent $84,000 on media, sales support, and marketing technology and acquired 2,000 new customers. Average first-order revenue was $60, but contribution after product, fulfilment, payment, and introductory discount was $18. The monthly retention estimate for the first year was 88%, and management used a monthly discount rate of 1%.

The blended CAC is:

CAC = 84,000 ÷ 2,000 = $42.

A first-order contribution of $18 does not recover CAC. Management therefore needs to understand later contribution rather than optimize only for acquisition volume.

Analysis

Cohort analysis shows that referral customers have CAC of $30 and monthly contribution of $22, while social-media customers have CAC of $55 and monthly contribution of $15. Social customers are not necessarily undesirable, but they require stronger onboarding, better creative targeting, or a different acquisition price.

RFM analysis identifies a valuable at-risk segment: customers with high historical monetary value, high frequency, and no purchase in the last 75 days. A blanket discount to all inactive customers would waste margin. Aurora instead tests two actions: a service-oriented reminder and a personalized replenishment recommendation, with a randomized no-contact control.

Suppose the test produces a purchase probability of 0.24 for the recommendation group and 0.18 for control. If incremental contribution per returning customer is $25 and the recommendation costs $1.20 per contacted customer, expected incremental value per contacted customer is:

(0.24 − 0.18) × 25 − 1.20 = 0.06 × 25 − 1.20 = $0.30.

The action is economically positive on average, although the result should be accompanied by confidence intervals and checked for variation across customer groups.

Approach and outcome

Aurora changes its dashboard from acquisition volume to cohort contribution, payback, repeat purchase, and retention. It introduces a second-purchase onboarding journey, reduces messages to customers with low predicted incremental response, and routes high-value service failures to human agents. After further testing, the company reallocates some budget from the high-CAC social segment to referrals and improves social targeting rather than eliminating social acquisition entirely.

Learning points

1. CAC must be interpreted with contribution and retention.
2. Blended averages conceal channel and cohort differences.
3. Propensity is not incrementality; control groups are required for causal claims.
4. The best next action may be service, education, an offer, or no contact.
5. Customer economics should improve customer experience, not reduce customers to scores.

Challenges and limitations

Customer economics is difficult when purchases are infrequent, customers share accounts, channels are poorly integrated, or costs are allocated inconsistently. Returns and cancellations can make early revenue appear larger than realized contribution. Privacy restrictions may reduce observability, while cookie loss and fragmented identities weaken attribution.

Models can overfit historical behaviour, fail when products or prices change, and encode inequities present in the data. CLV may systematically undervalue new segments because limited history is mistaken for low potential. RFM may classify seasonal customers as inactive. These limitations require domain knowledge, sensitivity analysis, experimentation, and continuous monitoring.

Best-practice checklist

• Define metrics, periods, units, and customer eligibility before calculation.
• Use contribution rather than revenue when evaluating economic value.
• Analyse cohorts and segments instead of relying only on blended averages.
• Discount future cash flows when the time value of money is material.
• Use appropriate observation windows for repeat and churn measures.
• Separate prediction from causal impact through treatment-control testing.
• Include no-contact and service actions in next-best-action systems.
• Test loyalty programmes for incremental value after reward costs.
• Protect privacy, explain profiling, and monitor disparate outcomes.
• Re-estimate assumptions and monitor model drift.

Chapter Summary

Customer economics links brand relationships to financial and behavioural outcomes. CAC measures the cost of acquiring customers, but it becomes meaningful only when compared with contribution, retention, and payback. Retention and churn require clearly defined active populations and observation windows. Repeat-purchase rate measures second-order behaviour, while share of wallet estimates the brand’s proportion of category spending.

CLV estimates expected future contribution and should use consistent periods, realistic survival probabilities, appropriate discounting, and scenario analysis. RFM offers a practical descriptive segmentation method, while propensity models estimate likely events. Uplift and randomized experiments are needed to estimate incremental response. Next-best-action systems should choose among offers, service, education, and no contact according to expected incremental value and ethical constraints.

Effective relationship analytics depends on clean data, valid attribution, experimentation, governance, privacy protection, fairness assessment, and human accountability. Its strategic purpose is not simply to extract more value from customers, but to create durable, mutually beneficial relationships that strengthen both customer experience and brand performance.

Key Terms

Customer acquisition cost (CAC): Average attributable cost of obtaining a new customer.

Contribution margin: Revenue remaining after specified variable and customer-attributable costs.

Cohort: Group of customers sharing a common starting condition or characteristic.

Retention rate: Proportion of the beginning customer base remaining active after a period.

Churn rate: Proportion of customers becoming inactive or cancelling during a defined period.

Repeat-purchase rate: Proportion of eligible customers making at least a second purchase.

Purchase frequency: Average number of orders per active customer in a period.

Share of wallet: Brand spending divided by the customer’s total category spending.

Customer lifetime value (CLV): Expected present value of future customer contribution net of relevant costs.

Discount rate: Rate used to convert future monetary amounts to present value.

RFM: Recency, frequency, and monetary-value behavioural segmentation.

Propensity score: Estimated probability that a customer will exhibit a defined event.

Uplift: Incremental difference in outcome probability caused by an intervention.

Incrementality: Outcome attributable to an action beyond what would have occurred without it.

Next-best action: Recommended customer-management action under specified objectives and constraints.

Cohort analysis: Comparison of customer groups over time from a common starting point.

Calibration: Agreement between predicted probabilities and observed event frequencies.

Customer profitability: Customer-level economic contribution after relevant costs.

Data drift: Change in data patterns that reduces model or measurement reliability.

Treatment-control design: Experimental comparison used to estimate an intervention’s incremental effect.

Review Questions

Short Answer Questions

1. Why is contribution margin generally more useful than revenue for evaluating customer value?
2. What is the difference between retention rate and repeat-purchase rate?
3. Define recency, frequency, and monetary value in RFM analysis.
4. Why must CAC numerator and denominator use compatible attribution rules?
5. What is the difference between propensity and uplift?

Long Answer Questions

1. Explain how a firm should calculate and interpret customer lifetime value, including the assumptions of the simplified perpetual model and the discounted finite-horizon model.
2. Discuss how cohort analysis can improve marketing budget allocation compared with a single blended CAC.
3. Evaluate the benefits and limitations of RFM segmentation for CRM decision-making.
4. Explain how loyalty programmes can create value while also reducing contribution margin.
5. Discuss the privacy, fairness, and governance requirements for AI-supported personalization.

Analytical / Application Questions

1. A company spends $72,000 to acquire 1,500 customers. Each customer generates $16 contribution in month one. Calculate CAC and explain why first-month contribution alone is insufficient to assess customer profitability.
2. A subscription brand begins a month with 10,000 customers, acquires 1,200 new customers, and ends with 10,100 active customers. Calculate the retention rate and churn rate under the stated definitions, then identify one limitation of the calculation.
3. An intervention produces a purchase probability of 0.31 compared with 0.25 in a control group. Incremental contribution per purchase is $30 and contact cost is $1.50. Calculate expected incremental value per contacted customer and recommend whether the action should be tested further.

Multiple Choice Questions with Answers

Question 1

A campaign costs $65,000 and acquires 1,300 new customers. What is the CAC?

A. $40

B. $50

C. $65

D. $84.50

Correct Answer: B. $50

Question 2

A firm begins with 8,000 customers, acquires 1,000 new customers, and ends with 7,600 customers. What is the retention rate?

A. 82.5%

B. 87.5%

C. 95.0%

D. 105.0%

Correct Answer: B. 87.5%

Question 3

Which measure is most directly concerned with the proportion of customers making at least a second purchase?

A. Churn rate

B. Share of wallet

C. Repeat-purchase rate

D. CAC

Correct Answer: C. Repeat-purchase rate

Question 4

A customer spends $300 with a brand and an estimated $1,000 in the category. What is the customer’s share of wallet?

A. 3%

B. 30%

C. 70%

D. 333%

Correct Answer: B. 30%

Question 5

Under the simplified perpetual CLV model, monthly contribution is $25 and monthly retention is 80%. What is CLV before acquisition cost?

A. $31.25

B. $100

C. $125

D. $500

Correct Answer: C. $125

Question 6

Which RFM dimension measures how recently a customer purchased?

A. Recency

B. Frequency

C. Monetary value

D. Retention

Correct Answer: A. Recency

Question 7

A propensity model primarily estimates:

A. The causal effect of an intervention

B. The probability of a specified customer event

C. The accounting value of fixed assets

D. The exact future revenue of every customer

Correct Answer: B. The probability of a specified customer event

Question 8

If treatment purchase probability is 0.20 and control probability is 0.14, what is uplift?

A. 0.06

B. 0.14

C. 0.20

D. 1.43

Correct Answer: A. 0.06

Question 9

Why is a randomized control group valuable in CRM testing?

A. It guarantees every customer will respond

B. It estimates incremental impact relative to what would otherwise occur

C. It eliminates the need to measure costs

D. It makes all predictions perfectly calibrated

Correct Answer: B. It estimates incremental impact relative to what would otherwise occur

Question 10

Which is an appropriate next-best-action principle?

A. Always contact the customers with the highest purchase propensity

B. Always offer the largest discount to customers at risk

C. Compare expected incremental value across actions, including no contact

D. Ignore fulfilment and incentive costs

Correct Answer: C. Compare expected incremental value across actions, including no contact