Customer Journey, CRM and Lifetime Value

Introduction

Marketing analytics is increasingly concerned not only with attracting customers, but with understanding how relationships develop over time. A customer may first encounter a brand through a search advertisement, compare alternatives on a website, purchase through a mobile application, seek support through a call centre, and later become a repeat buyer or advocate. These interactions form a customer journey.

Customer relationship management (CRM) provides the organisational and technological processes used to collect, integrate and act on customer information. Lifetime value (LTV) translates relationship outcomes into an economic measure. Together, journey analysis, CRM and LTV help organisations decide which customers to acquire, how to personalise communications, when to intervene in a declining relationship and how much to invest in retention.

The central analytical challenge is that customers differ in probability, timing and value. A campaign may generate many conversions but still destroy value if acquisition costs are high or if customers quickly churn. Conversely, a small segment may be highly profitable because it purchases frequently, remains active for years and responds efficiently to service or cross-selling.

Learning Objectives

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

1. Explain acquisition, conversion, retention, churn, cohorts and customer journeys.
2. Select and calculate appropriate marketing and CRM metrics.
3. Estimate customer acquisition cost, retention value and customer lifetime value.
4. Use cohort analysis to distinguish customer quality from short-term volume.
5. Explain the purpose and limitations of propensity and churn models.
6. Design CRM, personalisation and next-best-action decisions.
7. Evaluate privacy, fairness, attribution and governance issues in customer analytics.

1. The Customer Journey as an Analytical Object

The customer journey is the sequence of customer experiences and interactions with an organisation before, during and after a purchase. It is not necessarily linear. Customers can enter through different channels, repeat stages, abandon a process, return later or interact with several departments simultaneously.

A useful distinction is between:

• Touchpoints: observable interactions, such as an advertisement impression, website visit, product review, purchase or support contact.
• Channels: environments through which interactions occur, such as email, search, retail stores, social media, telephone and mobile applications.
• Stages: analytical groupings of interactions, such as awareness, consideration, conversion, adoption, retention and advocacy.
• Customer experience: the customer’s perceived quality and meaning of interactions, which may not be fully captured by clickstream data.

A journey map links customer goals, organisational processes, touchpoints, measures and pain points. An analytical journey map should specify the unit of analysis. For example, the unit may be a person, account, household, subscription or order. Confusing these units creates serious errors: one household may contain several users, while one business account may contain many decision-makers.

A journey funnel

A funnel summarises movement through stages. If 10,000 visitors produce 500 purchases, the visitor-to-purchase conversion rate is:

The numerator and denominator must refer to the same period, population definition and event definition. If 500 of 10,000 visitors purchase, conversion is 5%. A funnel is useful for locating volume loss, but it can be misleading when customers move backwards, use several devices or convert after a long delay.

Journey analytics

Journey analytics commonly combines event data, transaction data, campaign exposure, service records and survey information. Important analytical tasks include:

• identifying common paths and path deviations;
• measuring time between stages;
• locating abandonment points;
• comparing journeys by segment or channel;
• linking experiences to conversion, retention or margin;
• detecting friction, such as repeated authentication or failed payment.

Correlation between a touchpoint and an outcome does not prove that the touchpoint caused the outcome. Customers who are already highly motivated may be more likely to open emails, use the app and purchase. Causal tests, such as controlled experiments or carefully designed quasi-experiments, are therefore preferable when deciding whether an intervention created incremental value.

2. Acquisition, Conversion, Retention and Churn

Acquisition

Acquisition concerns the creation of new customer relationships. Acquisition metrics include reach, qualified leads, conversion rate, acquisition cost and payback period. A lead should be defined consistently; a form submission, sales-qualified lead and paying customer are not interchangeable.

Customer acquisition cost (CAC) is commonly calculated as:

Costs may include media, agency fees, creative production, sales labour, incentives and technology, depending on the managerial purpose. If $60,000 of relevant acquisition spending creates 1,200 new customers, CAC is $50 per customer. CAC is not automatically comparable across channels unless cost scope, attribution rules and customer definitions are consistent.

Conversion

Conversion is the completion of a defined desired action. The conversion rate may be calculated for a session, visitor, lead, account or opportunity. A conversion window must be specified. A seven-day conversion rate and a ninety-day conversion rate answer different questions.

Retention

Retention measures the proportion of customers who remain active or continue a relationship during a specified period. For a cohort of customers active at the beginning of a period:

If 800 of 1,000 beginning-of-month customers remain active, retention is 80%. The organisation must define “active.” For a subscription service it may mean paid and not cancelled; for a retailer it may mean at least one purchase in a specified window.

Churn

Churn is the loss of customers or revenue. Customer churn and revenue churn are distinct. If 100 of 1,000 customers cancel, customer churn is 10%. Revenue churn may be higher or lower depending on the value of the customers who leave.

For a simple period in which customers either remain or leave:

Retention and churn sum to 100% only when they use the same population, period and mutually exclusive definitions. In practice, voluntary cancellations, involuntary payment failures, inactivity and product downgrades should be analysed separately.

3. Cohort Analysis

A cohort is a group sharing a meaningful starting characteristic, usually the month or quarter of first purchase or subscription. Cohort analysis compares customers with similar relationship age rather than mixing new and mature customers.

A cohort retention table may show the percentage of each acquisition-month cohort that remains active after one, two and three months. Cohorts reveal whether apparent growth is caused by sustained customer quality or by continually replacing churned customers with new acquisitions.

For example, suppose two acquisition campaigns produce the following three-month retention:

Campaign B appears stronger on retention, but a complete assessment also requires margin, acquisition cost, order frequency and customer quality. Cohort comparisons can be confounded by seasonality, pricing, product changes and differing customer mix.

4. Customer Lifetime Value

Customer lifetime value estimates the economic contribution expected from a customer over the relationship. It is an estimate, not a directly observed fact. Its usefulness depends on the time horizon, cash-flow definition, discounting, margin assumptions and prediction quality.

A simple non-contractual LTV approximation

For a relatively stable business with constant periodic values, a common approximation is:

where:

• m = contribution margin per customer per period;
• r = probability of retention from one period to the next;
• i = discount rate per period.

The formula assumes constant margin, a constant retention probability, an indefinite horizon, and retention decisions that are independent across periods. It is a perpetuity-style approximation and should not be used uncritically for rapidly changing businesses.

If contribution margin is $30 per month, monthly retention is 0.90 and the monthly discount rate is 0.01:

This is contribution value before acquisition cost. A simplified net value is approximately LTV − CAC, but this comparison is meaningful only when both quantities use compatible cost and time definitions.

Gross-margin and purchase-frequency approach

For non-subscription businesses, a common simplified estimate is:

If average order value is $80, annual purchase frequency is 3, gross margin is 40%, and expected duration is 2 years:

This estimate excludes discounting, service costs, returns, referrals and variation in customer behaviour. It is useful as a transparent planning approximation, not as a substitute for customer-level cash-flow modelling.

Discounted cash-flow LTV

A more general finite-horizon model is:

where p_t is the probability the customer is active in period t, m_t is expected contribution margin in that period, c_t is expected relationship cost in that period, i_t is the discount rate and T is the forecast horizon. All monetary terms must use the same currency and period. If acquisition cost is included, it should either be subtracted separately or included as a period-zero cost, but not counted twice.

LTV should normally be expressed as contribution or cash-flow value rather than revenue. Revenue ignores product cost, fulfilment, payment fees, returns and service costs. A high-revenue customer can therefore have low or negative value.

5. CRM and Customer Data Architecture

CRM is both a management philosophy and a set of processes and technologies for managing customer relationships. Operational CRM supports sales, service and marketing execution. Analytical CRM supports segmentation, measurement, prediction and decision-making. Collaborative CRM coordinates information across customer-facing functions and partners.

A practical CRM architecture may contain:

1. Source systems: web analytics, point-of-sale, e-commerce, email, advertising, call centre and product usage systems.
2. Identity resolution: rules or probabilistic methods linking records to a customer or account.
3. Customer data platform or warehouse: governed storage of events, profiles, transactions and permissions.
4. Analytical layer: metrics, cohorts, models and experimentation.
5. Activation layer: campaigns, service queues, recommendations and sales actions.
6. Feedback layer: responses and outcomes returned to improve measurement and models.

Data quality problems include duplicate identities, missing consent, inconsistent timestamps, bot traffic, delayed transactions and changing definitions. A CRM record should distinguish observed facts from inferred attributes. “Purchased product X on 12 March” is an observation; “price-sensitive customer” is an inference and should be treated as uncertain.

6. Propensity and Churn Models

A propensity model estimates the probability of a defined future event, such as purchase, renewal, response or upgrade. A churn model estimates the probability of becoming inactive or cancelling within a specified horizon.

A logistic regression is a common baseline for binary outcomes:

Here, Y is the outcome, x_j are predictor variables, β_j are estimated coefficients and β₀ is the intercept. The model estimates probability under the assumption that the log-odds are linear in the predictors. Coefficients describe associations conditional on included variables; they do not automatically establish causation.

Typical predictors include recency, frequency, monetary value, tenure, product usage, service contacts, payment failures, discount dependence and prior campaign response. Predictors must be available before the prediction date. Including a later outcome, such as a cancellation request already submitted, creates leakage and falsely inflates performance.

Evaluation

Accuracy alone is often unsuitable when churn is rare. Important measures include:

• Precision: among customers predicted positive, the proportion actually positive.
• Recall: among actual positive customers, the proportion identified.
• ROC-AUC: ranking performance across classification thresholds.
• Lift: response or event rate in a selected model-ranked group divided by the overall event rate.
• Calibration: agreement between predicted probabilities and observed frequencies.

A model should be evaluated using a time-based holdout where possible. Randomly splitting records can allow future behaviour to influence training and produce overly optimistic results.

A model score is not a decision. A retention action should be targeted when expected incremental benefit exceeds intervention cost:

The probability must represent incremental effect, not merely the probability that a customer will stay. A customer predicted to churn may be impossible to save, while a customer predicted to stay may not need an expensive incentive. Uplift modelling and randomized holdout groups are therefore valuable for next-best-action decisions.

7. Personalisation and Next-Best Action

Personalisation adapts content, timing, offer, product or service to a customer or segment. Next-best action (NBA) systems select an action from alternatives such as education, reminder, service escalation, cross-sell, renewal offer or no contact.

A responsible NBA process includes:

1. Define the customer objective and business constraint.
2. Establish eligibility, consent and exclusion rules.
3. Generate possible actions.
4. Estimate response and incremental value.
5. Apply frequency, budget and contact-pressure constraints.
6. Select and execute the action.
7. Measure outcome against a control or comparison group.
8. Record the result for learning and audit.

“No action” must be an available option. Excessive targeting can produce fatigue, perceived surveillance and channel conflict. Personalisation should be relevant, explainable and proportionate to the sensitivity of the data used.

8. Illustrative Case: StreamPlus Subscription Service

Background

StreamPlus is a fictional subscription video service. Management observes strong monthly sign-ups but increasing cancellations after the first three months. It wants to improve acquisition quality and prioritise retention activity.

Problem

During one quarter, StreamPlus spends $120,000 on attributable acquisition activities and obtains 4,000 new subscribers. Monthly contribution margin per active subscriber is $18. Historical monthly retention is 0.92, but a new low-price campaign appears to attract customers with lower engagement.

Analysis

CAC is:

Using the stable-retention approximation and a monthly discount rate of 0.01:

The approximate net value before other costs is $184 − $30 = $154. However, this result relies on historical retention and may overstate value if the new cohort behaves differently.

The analytics team creates monthly cohorts and finds that customers acquired through the low-price campaign have month-three retention of 55%, compared with 75% for customers acquired through content partnerships. It also develops a churn model using tenure, weekly viewing hours, failed payments and support contacts. The model is evaluated on a later cohort, calibrated, and reviewed for differential error rates across customer groups.

Approach and solution

StreamPlus takes four actions:

• reallocates some acquisition budget toward the higher-retention channel;
• sends onboarding content during the first four weeks rather than waiting for inactivity;
• offers service assistance to customers with payment failures, but does not automatically discount all high-risk customers;
• uses a randomized holdout group to estimate the incremental effect of each retention intervention.

The company calculates intervention value using expected contribution saved minus contact and incentive cost. It also reports cohort retention, contribution LTV, CAC, incremental retention and customer complaints rather than relying on sign-ups alone.

Outcome and learning points

The case illustrates that acquisition volume is not the same as acquisition quality. Cohort analysis exposed a difference hidden by aggregate retention. The churn model helped prioritise attention, but experimentation was required to determine who could actually be saved. The business should update LTV using observed cohort cash flows and should not assume that historical retention remains constant after a pricing or product change.

9. Limitations, Ethics and Governance

Customer analytics creates risks as well as value. Key issues include:

• Consent and purpose limitation: data should be collected and used for legitimate, communicated purposes.
• Data minimisation: organisations should avoid collecting sensitive data merely because it might become useful.
• Security: identity, payment and behavioural data require access control, encryption and retention policies.
• Fairness: models may disadvantage groups through biased data, proxy variables or unequal error rates.
• Transparency: customers should receive understandable information about meaningful automated decisions.
• Right to challenge: important decisions should allow review, correction and human escalation where appropriate.
• Attribution integrity: marketers should not claim credit for conversions that would have occurred without exposure.
• Dark patterns: retention should not depend on obstructing cancellation or exploiting behavioural vulnerabilities.

Model governance should document the target definition, observation window, prediction horizon, training data, exclusions, features, evaluation results, thresholds, intervention costs, monitoring plan and owner. Drift monitoring is essential because customer behaviour, competitors, prices and channels change.

10. Contemporary Developments and Best Practice

Modern CRM increasingly uses real-time event streams, unified profiles, recommendation systems and generative AI for content or service assistance. These technologies can improve responsiveness, but they do not remove the need for data quality, consent, human oversight and experimental evaluation. Generative systems may produce inaccurate or inappropriate messages and should not be given unrestricted access to sensitive customer information.

Best practice includes:

1. Define customers, events, periods and value consistently.
2. Separate revenue, gross margin, contribution margin and cash flow.
3. Analyse cohorts before interpreting aggregate growth.
4. Use time-aware validation and prevent leakage in predictive models.
5. Evaluate calibration and business value, not only discrimination metrics.
6. Measure incremental effects with holdouts or experiments.
7. Include “no action” and contact-frequency constraints in NBA systems.
8. Monitor fairness, privacy, model drift and customer experience.
9. Reconcile CRM metrics with finance and operational records.
10. Treat LTV as a forecast with uncertainty, not a permanent customer label.

Chapter Summary

Customer journey analytics examines how people interact with an organisation across acquisition, consideration, conversion, use, retention and advocacy. Funnels summarise stage movement, while journey analysis examines paths, timing and friction. Acquisition, conversion, retention and churn metrics are meaningful only when their populations, periods and event definitions are explicit.

Cohort analysis compares customers by relationship age and can reveal differences in acquisition quality that aggregate metrics conceal. CAC measures attributable acquisition spending per new customer. LTV estimates future contribution or cash-flow value and may be approximated with stable-retention formulas or modelled through discounted, time-varying cash flows. Units, assumptions, margins, discounting and acquisition costs must be handled consistently.

CRM integrates customer data and activates insights across marketing, sales and service. Propensity and churn models estimate future outcomes, but prediction is not causation and a high-risk customer is not necessarily persuadable. Next-best-action decisions should consider incremental value, costs, consent, fairness and the option of no contact. Strong practice combines modelling with experiments, governance and continuous monitoring.

Key Terms

Customer journey: The sequence of experiences and interactions between a customer and an organisation.

Touchpoint: A specific customer interaction with a brand or organisation.

Conversion: Completion of a defined desired action.

Customer acquisition cost (CAC): Attributable acquisition cost per new customer.

Retention: Continued customer activity or relationship over a specified period.

Churn: Customer loss, cancellation or inactivity during a specified period.

Cohort: A group sharing a relevant starting characteristic, commonly acquisition period.

Customer lifetime value (LTV): Expected economic contribution from a customer over a defined horizon.

Contribution margin: Revenue remaining after variable costs relevant to the decision.

Non-contractual relationship: A relationship in which inactivity does not clearly indicate cancellation.

CRM: Processes and technologies for managing customer relationships and information.

Propensity model: A model estimating the probability of a future customer event.

Churn model: A model estimating the probability of customer loss or inactivity.

Data leakage: Use of information unavailable at prediction time.

Calibration: Agreement between predicted probabilities and observed event frequencies.

Uplift: Incremental change caused by an intervention compared with a suitable control.

Next-best action: The most valuable eligible action selected for a customer under constraints.

Identity resolution: Linking records from different systems to the same customer or account.

Contact fatigue: Negative effects caused by excessive or poorly timed communications.

Review Questions

5 Short Answer Questions

1. Distinguish a customer journey touchpoint from a customer journey stage.
2. Why must the definition of an active customer be stated when calculating retention?
3. What is the difference between customer churn and revenue churn?
4. Why is contribution margin generally more useful than revenue in LTV calculations?
5. Explain data leakage in a churn model.

5 Long Answer Questions

1. Explain how cohort analysis can improve acquisition and retention decisions. Include a suitable example.
2. Compare the stable-retention approximation of LTV with a discounted finite-horizon LTV model.
3. Describe how a CRM architecture can connect data sources, analytics and customer-facing actions.
4. Explain why a predictive churn model should not be used alone to select retention interventions.
5. Discuss the major ethical and governance issues involved in personalisation and next-best-action systems.

3 Analytical / Application Questions

1. A company spends $75,000 on acquisition and obtains 2,500 customers. Each customer produces a contribution margin of $24 per month, monthly retention is 0.88 and the monthly discount rate is 0.01. Calculate CAC and approximate LTV using the stable-retention formula. Interpret the result and state two limitations.
2. A retailer’s new customers have strong first-month conversion but weak month-six retention. Design a cohort analysis and recommend three additional measures needed before reallocating acquisition budget.
3. A churn model identifies 20% of customers as high risk. Design an experiment to determine whether a service intervention creates incremental retention value, specifying treatment, control, outcome, time horizon and decision rule.

Multiple Choice Questions with Answers

Question 1

A company spends $40,000 on attributable acquisition and gains 1,000 new customers. What is CAC?

A. $4

B. $40

C. $400

D. $4,000

Correct Answer: B. $40

Question 2

Which analysis best compares customers according to the month in which they first subscribed?

A. Cohort analysis

B. Sentiment analysis

C. Inventory analysis

D. Variance analysis

Correct Answer: A. Cohort analysis

Question 3

If 900 of 1,000 beginning-period customers remain active, the retention rate is:

A. 9%

B. 10%

C. 90%

D. 900%

Correct Answer: C. 90%

Question 4

In the stable-retention LTV approximation, which variable represents contribution margin per period?

A. i

B. r

C. m

D. T

Correct Answer: C. m

Question 5

Why is a time-based holdout particularly useful for churn-model evaluation?

A. It guarantees perfect accuracy

B. It better reflects future deployment conditions

C. It removes the need for calibration

D. It makes churn definitions unnecessary

Correct Answer: B. It better reflects future deployment conditions

Question 6

Which situation is an example of data leakage?

A. Using tenure measured at the scoring date

B. Using prior-month purchase frequency

C. Using a cancellation request submitted after the prediction date

D. Using historical service contacts

Correct Answer: C. Using a cancellation request submitted after the prediction date

Question 7

A customer predicted to churn is not necessarily a good target for a discount because:

A. All churn predictions are invalid

B. The customer may be impossible to save or may stay without intervention

C. Discounts never affect behaviour

D. Retention and churn are identical concepts

Correct Answer: B. The customer may be impossible to save or may stay without intervention

Question 8

Which metric assesses whether predicted probabilities agree with observed frequencies?

A. Calibration

B. Reach

C. Frequency

D. CAC

Correct Answer: A. Calibration

Question 9

In a responsible next-best-action system, which option should normally be available?

A. Unlimited contact

B. Automatic discounting

C. No action

D. Ignoring consent

Correct Answer: C. No action

Question 10

If average order value is $50, annual purchase frequency is 4, gross margin is 30% and expected duration is 2 years, the simplified LTV is:

A. $60

B. $100

C. $120

D. $200

Correct Answer: C. $120