CRM, Email, Mobile, and Lifecycle Analytics

Introduction and Context

Customer relationship management (CRM) and lifecycle analytics examine how customers enter, develop, deepen, weaken, and sometimes end relationships with an organisation. Whereas acquisition analytics asks how to attract prospects, lifecycle analytics asks a broader managerial question: what action should be taken with each customer, through which channel, at what time, and at what expected economic value?

The analytical unit is not always an individual customer. It may be an account, household, device, subscription, loyalty member, or business buying centre. This distinction matters because identity resolution, privacy, and revenue attribution differ across units. A robust lifecycle programme integrates transactional data, behavioural events, campaign exposure, service interactions, consent status, and profitability measures.

Lifecycle analytics supports email campaigns, mobile messaging, loyalty programmes, marketing automation, customer service, and omnichannel decision-making. Its purpose is not simply to increase message volume. It is to improve relevant engagement and customer value while controlling contact costs, fatigue, risk, and regulatory exposure.

Learning Objectives

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

1. Explain the relationship between CRM, segmentation, lifecycle management, and customer value.
2. Construct and interpret RFM, cohort, retention, churn, and customer lifetime value measures.
3. Design email, mobile, loyalty, and automation experiments using appropriate control groups.
4. Evaluate personalization and next-best-action systems critically.
5. Integrate behavioural, financial, and ethical considerations in omnichannel decisions.
6. Recommend an analytics operating model for a lifecycle marketing programme.

1. The Lifecycle Analytics Perspective

A customer lifecycle can be represented as a sequence of states: prospect, activated customer, repeat customer, loyal or high-value customer, at-risk customer, dormant customer, and churned customer. Customers do not necessarily move through these states in a fixed order. A customer may reactivate after dormancy, change product category, or maintain a subscription while reducing usage.

Lifecycle management combines three elements:

• State estimation: What is the customer’s current condition?
• Value estimation: What future contribution is reasonably expected?
• Action selection: Which intervention has the greatest incremental benefit?

A CRM platform stores customer records and orchestrates interactions; analytics supplies measurement, prediction, experimentation, and decision rules. A campaign response is not equivalent to business value. For example, a discount may generate an order from a customer who would have purchased anyway. The relevant quantity is often incremental outcome rather than observed outcome.

Core data architecture

A useful analytical model commonly contains:

Data must distinguish an event from an outcome. An email send is an exposure; an open is a measured interaction; a purchase is an outcome. Tracking systems also contain missingness and measurement bias. Apple Mail Privacy Protection, blocked cookies, shared devices, and cross-device use can make opens or identities unreliable. Clicks, authenticated events, transactions, and holdout outcomes are generally stronger evidence than opens alone.

2. Segmentation and RFM Analysis

Segmentation divides customers into groups that are sufficiently similar for a decision. Useful segmentation criteria include need, behaviour, value, lifecycle state, channel preference, risk, and predicted response. Demographic segmentation can be useful, but it should not be treated as a substitute for observed behaviour or customer need.

A practical first model is RFM:

• Recency (R): how recently the customer purchased.
• Frequency (F): how often the customer purchased during a defined period.
• Monetary value (M): how much value the customer generated during that period.

For customer i, a basic monetary measure is:

Here, T is the observation period and revenue should be defined consistently. For profitability analysis, contribution margin is preferable to gross revenue because discounts, returns, fulfilment, and service costs may differ substantially across customers.

RFM scores are often assigned using quantiles, such as scores from 1 to 5. The method is easy to explain but has limitations: scores are relative to the selected population, are sensitive to the observation window, and treat unequal differences as equal categories. A customer at the 80th percentile is not necessarily twice as valuable as one at the 40th percentile.

Worked example: RFM interpretation

Suppose a retailer observes four customers over six months:

Customer A is recently active, frequent, and valuable. Customer B has the highest frequency and margin but may be becoming inactive. Customer C is recent but not yet frequent, making activation or cross-sell appropriate. Customer D is a reactivation candidate. The analytical response should therefore differ: loyalty recognition for A, retention or replenishment for B, onboarding for C, and carefully tested win-back for D.

A segmentation should be actionable. If two segments receive the same offer, channel, timing, and treatment, the distinction may have no operational value. Segment definitions should also be monitored for stability and fairness.

3. Cohorts, Retention, and Churn

A cohort is a group sharing a meaningful starting event, such as first purchase month, app installation month, or subscription start month. Cohort analysis compares customers with similar tenure rather than mixing new and mature customers.

For cohort c and period t, a simple retention rate is:

The denominator is the cohort size at its starting point. “Active” must be defined, for example as purchasing, logging in, using a core feature, or remaining subscribed. Retention is not meaningful without a specified activity definition and time interval.

For a subscription service with no reactivation, period churn can be expressed as:

Under those definitions, retention and churn are complements. However, they may not sum to 100% when the organisation defines activity, churn, pauses, and reactivation differently.

A useful retention table places tenure periods in columns and cohorts in rows. It can reveal whether newer customers retain better, whether a product change altered the onboarding curve, and whether apparent growth is merely acquisition masking poor retention.

Churn prediction requires a labelled outcome and a prediction horizon. Features may include declining usage, failed payments, service complaints, reduced purchase frequency, and unredeemed benefits. Predictive accuracy is not enough: a model should be calibrated, operationally actionable, and evaluated for differential error across groups. A high-risk score is not proof that a customer will churn; it is an estimate used to prioritise intervention.

4. Customer Lifetime Value

Customer lifetime value (CLV) estimates the present or expected future contribution associated with a customer relationship. It is a decision measure, not a directly observed fact. Definitions must specify revenue versus margin, time horizon, discounting, retention assumptions, acquisition costs, and treatment of service costs.

A simple finite-horizon discounted CLV is:

Where p_t is the probability of being active in period t, m_t is expected contribution margin if active, c_t is period-specific relationship or servicing cost, d is the discount rate per period, T is the forecast horizon, and CAC is customer acquisition cost. All monetary terms must use the same currency and period; d must correspond to the period length.

For a stationary subscription with constant periodic contribution margin m, constant retention probability r, and discount rate d, an infinite-horizon approximation is:

This formula assumes that the customer contributes m at the end of each period, remains active from one period to the next with probability r, and that retention and margin are constant. It should not be used when churn changes materially with tenure, margins vary by customer, or reactivation is important. In such cases, a cohort or probabilistic model is more appropriate.

Worked example

A subscription generates £12 contribution margin per month. Monthly retention is 0.95, the monthly discount rate is 0.01, and acquisition cost is £80. The stationary approximation gives:

The estimated net CLV is £110. This result is highly sensitive to retention: improving retention may create more value than increasing short-term conversion through discounts. The calculation is an estimate under stated assumptions, not a guarantee of future profit.

5. Email, Mobile, and Automation Analytics

Email and mobile programmes should be managed as controlled communication systems rather than isolated campaigns. Important measures include delivery rate, click rate, conversion rate, unsubscribe rate, complaint rate, revenue or margin per recipient, and incremental outcome relative to a control group.

Open rate is increasingly noisy because of privacy protections and machine-generated opens. It may still help diagnose delivery or creative problems, but it should not be the sole success metric. Mobile analytics must distinguish push permission, delivery, notification interaction, in-app event, and downstream conversion. SMS has strong immediacy but higher perceived intrusiveness and cost; push notifications depend on permission and app installation.

Automation commonly includes:

1. Trigger: an event or time condition occurs.
2. Eligibility: consent, suppression, geography, inventory, and frequency rules are checked.
3. Decision: segment, propensity, or next-best-action logic selects treatment.
4. Delivery: message and channel are executed.
5. Measurement: exposure, response, incremental outcome, and downstream value are recorded.
6. Learning: results update rules, models, and contact policies.

Frequency management is a constrained optimisation problem. More messages may increase opportunities for conversion but also increase fatigue, complaints, unsubscribes, and brand damage. A suppression rule can be economically rational even when a customer is predicted to respond, if expected incremental margin is below contact cost or reputational risk.

Incrementality and experimentation

A campaign’s observed conversion rate is:

This describes performance among recipients but does not establish causality. Incremental lift is better estimated through randomised treatment and control groups:

If 8% of 10,000 treated customers purchase and 6% of 10,000 control customers purchase, the estimated incremental purchases are 200, and lift is 2 percentage points. Statistical uncertainty should be reported, especially for small samples or many simultaneous tests. Randomisation should occur before treatment, and contamination, noncompliance, seasonality, and unequal exposure must be considered.

6. Personalization and Next-Best Action

Personalization ranges from simple rule-based selection to machine-learning recommendation. Examples include replenishment reminders, product recommendations, content variation, loyalty benefits, and service recovery. Effective personalization requires a meaningful decision context; changing a customer’s name is not necessarily valuable personalization.

A next-best-action (NBA) system evaluates feasible actions such as send email, send push, offer assistance, recommend a product, delay contact, or suppress communication. A simplified expected-value rule is:

Here, A is the set of eligible actions, P(Y_a=1) is the predicted probability of the desired outcome under action a, V_a is the contribution value, Cost_a is execution cost, and Risk_a represents estimated negative consequences. In practice, the system should use uplift or treatment-effect estimates where possible, because a high probability of purchase may reflect natural demand rather than persuasion.

NBA systems require guardrails: consent, contact frequency, inventory availability, affordability rules, fairness checks, explainability appropriate to the decision, and human review for sensitive actions. A model that optimises clicks may learn to over-message customers or promote low-margin products. The objective function must therefore reflect durable customer and business value.

7. Loyalty and Omnichannel Measurement

Loyalty analytics evaluates repeat behaviour, reward liability, tier progression, redemption, breakage, share of wallet, and incremental retention. A loyalty member’s higher spend may reflect self-selection rather than programme impact. Randomised enrolment invitations, matched comparisons, or phased rollouts can help estimate incrementality.

Omnichannel analysis must avoid double counting. A customer may receive an email, view a mobile notification, visit a store, speak to an agent, and purchase through a website. Last-touch attribution assigns the outcome to the final interaction but can undervalue earlier influences. Multi-touch models distribute credit but depend on assumptions and can create false precision. Holdout experiments and customer-level incrementality are generally stronger for causal decisions.

Identity graphs should use stable, governed identifiers and record confidence. Probabilistic matching can create false merges, while deterministic matching can leave customers fragmented. Both errors affect frequency control, CLV, and personalisation.

8. Illustrative Case: Northstar Outfitters

Background. Northstar Outfitters is an online and store-based apparel retailer. It has 1.2 million marketable email profiles, a mobile app, and a points-based loyalty programme. Management reports rising email revenue but declining repeat purchase rate and increasing unsubscribe volume.

Problem. The existing programme sends weekly promotions to almost every opted-in customer. Reporting uses opens and last-touch revenue. New customers receive no structured onboarding, and high-value customers receive the same discount as dormant customers.

Analysis. The analytics team constructs six-month RFM features, contribution margin, product category, consent status, service contacts, and email exposure. Cohorts are formed by first purchase month. The analysis finds that customers receiving a second purchase within 45 days have materially higher six-month retention, while repeated discount exposure increases margin erosion. Open rates are unreliable, but clicks and authenticated purchases are usable. A churn model identifies declining category engagement and delivery problems as important signals.

Approach. Northstar designs four interventions:

• a 30-day onboarding journey for first-time buyers;
• replenishment and complementary-product messages based on purchase category;
• a high-value loyalty treatment using early access rather than universal discounts;
• a reactivation test for dormant customers, with a no-contact control group.

Eligibility rules suppress recent purchasers, customers without appropriate consent, and customers who have reached the contact cap. Each journey includes treatment and control groups. The primary metric is incremental contribution margin per eligible customer over 60 days, with unsubscribe and complaint rates as guardrails.

Outcome. In this illustrative scenario, onboarding increases second-purchase rate, but the largest economic improvement comes from reducing unnecessary discounts among high-value customers. The reactivation programme produces positive incremental margin only for a subset with recent historical category activity; the broad dormant audience is not profitable to contact. Management therefore replaces one universal win-back journey with a selective policy.

Learning points. First, lifecycle states should determine treatment. Second, revenue must be replaced or supplemented by incremental margin. Third, a control group can reveal that apparently successful messages merely harvest existing demand. Fourth, contact policy and suppression are strategic decisions, not technical afterthoughts.

9. Governance, Ethics, and Regulation

Lifecycle analytics relies on personal data and therefore requires purpose limitation, data minimisation, access control, retention schedules, and transparent notices. Organisations must respect applicable privacy and electronic marketing law, including consent requirements, opt-out mechanisms, and restrictions on sensitive profiling. Legal compliance varies by jurisdiction and should be verified with qualified counsel.

Ethical practice extends beyond formal compliance. Customers should not be manipulated through exploitative urgency, discriminatory targeting, opaque pricing, or excessive contact. Sensitive attributes and their proxies require special care. A churn model may systematically over-contact financially vulnerable customers or deny benefits to groups with incomplete data. Fairness assessment should examine selection rates, error rates, treatment effects, and business justification across relevant groups, while recognising that no single fairness metric resolves every ethical conflict.

Governance should document data lineage, model purpose, training window, performance, drift, exclusions, human override, and escalation procedures. Customers should have practical ways to change preferences and stop communications. Security controls must protect both customer data and campaign credentials.

10. Contemporary Developments and Best Practice

Generative AI can assist with content variation, summarisation, and analyst productivity, but it should not independently determine sensitive targeting or make unreviewed claims. Retrieval from approved product and policy sources, brand controls, testing, and human approval are essential. Real-time decisioning is increasingly common, yet real time is not always optimal: a considered delay may improve relevance and reduce fatigue.

Best practice includes:

1. Define the customer, event, outcome, value measure, and time horizon precisely.
2. Treat consent and eligibility as model constraints, not merely campaign filters.
3. Prefer contribution margin and incremental outcomes to gross attributed revenue.
4. Use cohorts to distinguish tenure effects from campaign effects.
5. Combine descriptive segmentation with predictive and causal analysis.
6. Maintain treatment and control groups for important interventions.
7. Cap contact frequency and measure fatigue as an economic and customer outcome.
8. Validate identity resolution and document uncertainty.
9. Monitor model drift, calibration, fairness, and operational feasibility.
10. Build closed-loop learning in which results update rules without removing managerial oversight.

Chapter Summary

CRM and lifecycle analytics connect customer data to decisions across email, mobile, loyalty, automation, and omnichannel programmes. Segmentation becomes valuable when it changes action. RFM provides an interpretable behavioural starting point, while cohort analysis clarifies retention by tenure. Churn models estimate risk but require actionable interventions and fairness controls. CLV converts expected future contribution into an economic basis for acquisition, retention, and contact decisions, provided its assumptions are explicit.

Campaign metrics describe observed activity; randomised controls estimate incremental impact. Personalization and next-best-action systems should optimise expected incremental value subject to consent, cost, inventory, frequency, privacy, and fairness constraints. Effective lifecycle management therefore combines data architecture, statistical reasoning, experimentation, managerial judgement, and responsible governance.

Key Terms

CRM: Systems and practices used to manage customer relationships and interactions.

Lifecycle analytics: Analysis of customer states, transitions, value, and interventions over time.

Segmentation: Division of a customer population into actionable groups.

RFM: Recency, frequency, and monetary value analysis.

Cohort: Group sharing a defined starting event or period.

Retention: Continued activity or subscription according to a stated definition.

Churn: Loss of an active customer or failure to remain active.

Customer lifetime value: Expected present or future contribution from a customer relationship.

Contribution margin: Revenue remaining after variable costs attributable to the transaction or relationship.

CAC: Customer acquisition cost.

Incrementality: Outcome caused by an intervention beyond what would have occurred without it.

Holdout group: Eligible customers deliberately withheld from treatment for comparison.

Uplift: Difference in treatment effect or outcome relative to a comparison condition.

Personalization: Adaptation of content, offer, timing, or channel to customer context.

Next-best action: The eligible action estimated to produce the greatest net value.

Contact fatigue: Negative response caused by excessive or poorly timed communication.

Identity resolution: Linking records or events to the correct customer entity.

Attribution: Assignment of credit for an outcome to one or more interactions.

Model drift: Change in data or relationships that reduces model performance over time.

Review Questions

Short Answer Questions

1. Why is contribution margin often preferable to revenue in CLV and campaign evaluation?
2. Define recency, frequency, and monetary value in RFM analysis.
3. What is the purpose of a holdout group?
4. Why can email open rate be an unreliable primary outcome measure?
5. Identify two governance controls required for lifecycle analytics.

Long Answer Questions

1. Explain how cohort analysis can distinguish acquisition growth from genuine retention improvement.
2. Compare rule-based segmentation, predictive churn modelling, and next-best-action decisioning.
3. Discuss the assumptions and limitations of the stationary CLV formula.
4. Design an email and mobile measurement framework for a subscription business.
5. Evaluate the ethical risks of personalisation based on sensitive or proxy variables.

Analytical / Application Questions

1. A retailer’s treatment conversion rate is 9% and its control conversion rate is 7% among 20,000 customers in each group. Calculate incremental purchases and explain what additional evidence is needed before scaling the campaign.
2. Design an RFM-based lifecycle programme for customers who are recent but infrequent, frequent but becoming inactive, and dormant. Specify channel, objective, and guardrail for each group.
3. A next-best-action model recommends a discount to a high-propensity customer. Explain how uplift modelling, margin, contact cost, consent, and fatigue constraints could change the decision.

Multiple Choice Questions with Answers

Question 1

Which measure is most directly intended to estimate the causal effect of a campaign?

A. Open rate

B. Last-touch revenue

C. Treatment-minus-control conversion rate

D. Number of messages sent

Correct Answer: C. Treatment-minus-control conversion rate

Question 2

In RFM analysis, monetary value most commonly represents what?

A. Time since the last purchase

B. Number of purchases

C. Value generated during a defined observation period

D. Probability of churn

Correct Answer: C. Value generated during a defined observation period

Question 3

A cohort is best defined as a group of customers who share what characteristic?

A. The same predicted CLV

B. A common starting event or period

C. The same email provider

D. An identical purchase basket

Correct Answer: B. A common starting event or period

Question 4

Under the stationary CLV assumptions, which variable represents the probability of remaining active from one period to the next?

A. m

B. CAC

C. d

D. r

Correct Answer: D. r

Question 5

Why is a control group important in lifecycle campaign evaluation?

A. It guarantees every customer receives a message

B. It estimates what would have happened without treatment

C. It increases the open rate

D. It removes the need to define an outcome

Correct Answer: B. It estimates what would have happened without treatment

Question 6

Which is a key limitation of last-touch attribution?

A. It cannot record any transaction

B. It always overvalues early interactions

C. It may undervalue interactions that occurred before the final touch

D. It requires randomisation

Correct Answer: C. It may undervalue interactions that occurred before the final touch

Question 7

Which action is most appropriate for a recently purchased but infrequent customer?

A. Immediate permanent suppression

B. Onboarding and activation support

C. A generic win-back message after one day

D. A high-risk fraud investigation without evidence

Correct Answer: B. Onboarding and activation support

Question 8

In a next-best-action framework, why is predicted purchase probability alone insufficient?

A. Probability cannot be estimated from data

B. A customer may purchase anyway, without incremental influence

C. Probability is always equal to margin

D. It eliminates consent requirements

Correct Answer: B. A customer may purchase anyway, without incremental influence

Question 9

Which metric is a useful guardrail for detecting communication fatigue?

A. Complaint or unsubscribe rate

B. Database row count

C. Number of product categories

D. Average screen size

Correct Answer: A. Complaint or unsubscribe rate

Question 10

If treatment conversion is 9% and control conversion is 7% for equal groups, the estimated incremental lift is:

A. 2 percentage points

B. 7 percentage points

C. 16 percentage points

D. 63 percentage points

Correct Answer: A. 2 percentage points