Marketing and Customer Analytics

Overview

A customer journey map is a purpose-bound representation of an experience for a specified segment, context, and time. Customer-journey research emphasizes experience over time and multiple touchpoints across channels; this chapter adds observed actions, evidence, friction, ownership, and measures to generate improvement hypotheses. The map itself does not establish emotion, causality, or business impact. ^[1]

How to Apply

1. Define Scope and Persona: Choose one specific customer persona and a specific journey to map (e.g., "the first 90 days for a new mid-market customer"). Don't try to map everyone at once.
2. Identify Journey Stages: Outline the major phases from the customer's perspective (e.g., Awareness, Consideration, Purchase, Onboarding, Engagement, Advocacy).
3. Map the Details for Each Stage:
   • Customer Actions: What is the customer doing? (e.g., "Googling solutions," "requesting a demo").
   • Touchpoints: Where are they interacting with you? (e.g., Website, Ads, Sales Rep, Support Ticket).
   • Emotions: How are they feeling? (e.g., Confused, Anxious, Hopeful, Relieved). Use qualitative data from interviews.
   • Pain Points: Where is the friction? What takes too long? What is confusing?
4. Identify Opportunities: For each pain point, brainstorm a corresponding improvement opportunity. This becomes your action plan. Prioritize opportunities based on their impact on the customer and the business versus the effort required.

Figure 5.1. Evidence-to-improvement customer-journey loop. This original synthesis shows stages, stage-level evidence, friction, ownership, and measurement rather than treating a journey as a one-way funnel.

flowchart LR
    A[Awareness] --> B[Consideration]
    B --> C[Purchase]
    C --> D[Onboarding]
    D --> E[Engagement]
    E --> F[Advocacy]
    F --> E
    A --> G[Evidence: Behavior, Interviews, Service Data]
    B --> G
    C --> G
    D --> G
    E --> G
    F --> G
    G --> H[Friction, Needs, and Accessibility]
    H --> I[Owner, Hypothesis, Metric, and Guardrail]
    I --> J[Test or Service Improvement]
    J --> G

    style A fill:#4ecdc4
    style G fill:#ff6b6b
    style J fill:#95e1d3

Text equivalent: For a defined segment and journey, collect behavioral, interview, and service evidence at awareness, consideration, purchase, onboarding, engagement, and advocacy. Translate friction, needs, and accessibility issues into an owned hypothesis with a metric and guardrail; test the change and feed the results back into the map.

Source note: Author-created teaching diagram informed by Lemon and Verhoef's customer-journey review; no arrow represents a proven causal effect, and the example stages are illustrative rather than a reproduced published model. ^[1]

So What for Managers

• Use the map to decide which customer problem, stage, owner, and measure deserves attention before selecting a campaign or product change.
• Convert observed friction into a bounded hypothesis with a customer benefit, business outcome, guardrail, and evidence plan.
• Revisit the map after research, service changes, or experiments; a static workshop artifact should not become the operating plan by default.

Limits and Critiques

• A map describes an experience under a chosen segment, context, and time window; it does not prove emotion, causality, incremental demand, or financial impact.
• Author synthesis: Stated emotions and recalled journeys can reflect recall, selection, survey-mode, and researcher-coding bias; triangulate them with behavior and service evidence.
• Stage names and touchpoints vary by business, accessibility need, channel, and customer; do not treat the illustrative sequence as a universal funnel.

Connections

• Inputs include qualitative research from Jobs-to-be-Done (Framework 2) and quantitative behavior, service, and drop-off data.
• Outputs can generate hypotheses for A/B Testing (Framework 8), service improvement, product discovery, and the Product Roadmap.
• Use the map alongside Cohort Analysis (Framework 10) and CLV/CAC Analysis (Framework 4) when a proposed experience change has retention or economic implications.

Overview

Jobs-to-be-Done is a family of practitioner approaches associated with Christensen and collaborators, Moesta, and related customer-needs research. Across variants, it asks what progress a customer seeks in a particular circumstance. Treat interview accounts as hypotheses to triangulate with observed behavior, alternative explanations, market evidence, and experiments—not as direct access to a customer's true motive. ^{[2] [3]}

How to Apply

1. Conduct a switch or struggle interview: In the Moesta–Spiek practitioner variant, ask for the chronology of a recent purchase, switch, or non-consumption decision and probe the circumstances, alternatives, events, concerns, and desired progress. Do not rely on stated feature preferences alone. ^[3]
2. State a job hypothesis: Summarize what progress may have been sought and in which circumstance, then test competing explanations. For this chapter, [progress sought] + [circumstance] + [constraints or success evidence] is an author-created notation, not a standardized JTBD syntax.
3. Map one practitioner variant—the Forces of Progress: Moesta and Spiek describe four forces in a switching decision: ^[3]
   • Push: The pain of the current situation driving the customer to act.
   • Pull: The appeal of the new solution.
   • Anxiety: The fears and uncertainties about the new solution.
   • Habit: The inertia of their current solution.
   • The model proposes that change becomes more likely when push and pull outweigh anxiety and habit. Use it as an interview-coding hypothesis, not a quantitative law or proof that marketing caused the switch.

So What for Managers

• Use a recent choice, switch, or non-consumption story to define a customer problem and a testable progress hypothesis rather than collecting a feature wish list.
• Compare the proposed job with alternatives, constraints, accessibility needs, behavioral evidence, and commercial outcomes before changing product or messaging.
• Route the hypothesis to a bounded discovery or market test; a compelling interview account is not a demand forecast or a product requirement.

Limits and Critiques

• JTBD is a family of practitioner variants, not one universally standardized method; the Forces of Progress and the chapter's bracketed job notation are bounded teaching choices.
• The milkshake story is a commonly reported teaching illustration, not universal evidence; verify jobs, alternatives, constraints, and desired outcomes in the market being studied.
• Retrospective interviews are vulnerable to recall, selection, social-desirability, and researcher-coding bias; triangulate with behavior, market evidence, and experiments.

Connections

• Inputs include customer interviews, observed behavior, market and competitor evidence, accessibility research, and experimental results.
• Outputs can inform Go-to-Market Strategy (Chapter 14), product discovery, and messaging tests while leaving feature and investment decisions open.
• Combine job hypotheses with Customer Journey Mapping (Framework 1) and A/B Testing (Framework 8) to connect context, experience, and measurable behavior.

Overview

RFM segmentation is a descriptive method that groups customers by recency, frequency, and monetary-value measures derived from transaction history. Fader, Hardie, and Lee formally connect RFM inputs to CLV through a model, which also shows why a simple score or segment is not itself customer lifetime value. The resulting groups do not by themselves establish loyalty, churn risk, or response to an intervention. ^[4]

How to Apply

1. Score Each Customer: For every customer, assign a score from 1 to 5 for each of the three dimensions based on their purchase history (e.g., top 20% of recent buyers get a Recency score of 5).
2. Combine Scores: Concatenate the scores to create an RFM segment (e.g., a customer with a score of 5 for Recency, 5 for Frequency, and 5 for Monetary is in the "555" segment).
3. Map to Hypothesis Segments: Labels such as “recent/frequent/high-spend” describe observed transactions, not customer worth, loyalty, or treatment response. For each segment, state the proposed action, customer benefit, consent/fairness limits, expected incremental effect, cost, holdout, and stop rule.

So What for Managers

• Use RFM cells to describe recent transaction patterns and prioritize questions, service reviews, or consent-respecting audience hypotheses.
• Define the proposed customer benefit, incremental outcome, cost, fairness check, holdout, and stop rule before acting on a segment.
• Reconcile segment patterns with contribution margin, service burden, retention, and customer experience rather than treating high spend as customer worth.

Limits and Critiques

• Score cutoffs, observation windows, missing purchases, returns, and monetary definitions are design choices; the illustrative 1–5 scoring is not a benchmark. ^[4]
• RFM is descriptive and can reflect acquisition mix, seasonality, pricing, channel, or measurement coverage; it does not establish loyalty, churn risk, or treatment response.
• Transaction history can omit non-purchase value, shared accounts, accessibility needs, and privacy or fairness risks; minimize data and review proxy effects before targeting.

Connections

• Inputs are defined transaction, margin, service-cost, consent, and customer-identity data from Finance/Sales and analytics systems.
• Outputs can define candidate audiences for A/B Testing (Framework 8), service changes, or research and can inform Customer Journey Mapping (Framework 1).
• Use CLV/CAC Analysis (Framework 4) and Cohort Analysis (Framework 10) to test whether a descriptive segment difference matters economically or longitudinally.

Overview

Customer lifetime value (CLV) is the present value of expected customer net cash flows, and Fader cautions against treating a homogeneous-customer point estimate as precise. This chapter compares a cohort CLV model with allocated acquisition cost as one input to a marketing investment decision; that comparison does not prove profitability. Retention, gross margin, service cost, discounting, attribution, cash timing, fixed costs, and uncertainty still matter. ^[5]

How to Apply

1. Calculate CAC by Cohort and Channel: Allocate sales, marketing, creative, promotion, referral, and relevant overhead consistently; state the attribution method and timing.
2. Estimate CLV: Model survival/retention, contribution margin, service cost, expansion/contraction, discounting, and uncertainty. The shortcut (period revenue × gross margin) / churn assumes a stable constant hazard and should be labeled accordingly.
3. Test Sensitivity and Cash: Compare cohorts and plausible ranges, calculate cash payback, reconcile with fixed costs and capacity, and identify which assumption reverses the recommendation.

So What for Managers

• Use CLV/CAC to compare customer-acquisition and retention choices under explicit cohort, margin, service-cost, discounting, and cash assumptions.
• Show a range, payback timing, capacity constraint, and the assumption that would reverse the recommendation before reallocating material spend.
• Treat the comparison as one decision input alongside customer, market, risk, operating, and financing evidence; it is not an approval threshold by itself.

Limits and Critiques

• The shortcut (period revenue × gross margin) / churn assumes a stable constant hazard and omits important timing, expansion, contraction, service, and uncertainty effects.
• CAC allocation depends on attribution, cost boundaries, timing, discounts, returns, and shared overhead; changing the convention can change the conclusion.
• Cohort averages hide heterogeneity and can be biased by censoring, selection, reactivation, pricing, and measurement changes; compare plausible alternatives and sensitivity ranges.

Connections

• Inputs include Attribution Models (Framework 5), Cohort Analysis (Framework 10), RFM Segmentation (Framework 3), finance records, and customer-service costs.
• Outputs inform go-to-market choices and fundraising and capital planning alongside customer, market, cash, risk, and execution evidence.
• Use Pricing Strategy & Models (Framework 6) and Customer Journey Mapping (Framework 1) to test how price, experience, retention, and service burden could change the economics.

Overview

Marketing attribution allocates conversion credit under a stated rule or model. Credit is not the same as incremental effect: observed touchpoints reflect targeting, customer intent, missing exposure data, platform rules, and selection. Li and Kannan demonstrate one multichannel model with a field validation in one firm; Dalessandro and colleagues show why approximations should be interpreted as variable-importance measures when causal assumptions fail. Use attribution for description and reconciliation, and use experiments or justified causal designs for incremental budget claims. ^{[6] [7]}

How to Apply

1. Understand the Models:
   • Last-Click: Gives all credit to the final recorded touchpoint. It is simple but sensitive to measurement coverage and customer intent.
   • First-Click: Gives all credit to the first recorded touchpoint. It describes the first touch associated with the path; it does not establish what caused awareness or conversion.
   • Linear: Divides credit equally among recorded touchpoints; equality is a convention, not evidence.
   • Position-Based: Assigns chosen weights to first, last, and middle touches; the weights encode assumptions.
   • Model-Based: Estimates patterns in observed paths. Prediction or probabilistic credit still does not establish what would have happened without exposure.
2. Triangulate: Compare credit rules, randomized holdouts where feasible, justified quasi-experiments, and aggregate marketing-mix models. Reconcile each with total spend, revenue, margin, coverage, and uncertainty; move budget only when the evidence is decision-relevant.

So What for Managers

• Use attribution to describe paths, reconcile recorded touchpoints, and generate sequencing or channel hypotheses under a declared rule.
• Define the estimand, population, exposure window, outcome, cost boundary, data coverage, and uncertainty before comparing channels or changing spend.
• Escalate material incremental-effect decisions to randomized holdouts or a justified quasi-experimental design where operations and ethics permit.

Limits and Critiques

• Rule-based and model-based credit can reflect targeting, intent, missing exposure, identity resolution, platform definitions, and selection; it is not automatically causal.
• The cited multichannel evidence is bounded to its data, model, and field setting; prediction or field validation does not make the method universal.
• Privacy choices, consent, offline exposure, cross-device gaps, interference, carryover, and spillover can change both the observed path and the decision.

Connections

• Inputs include a defined Customer Journey Map (Framework 1), exposure and conversion data, consent controls, channel costs, and a stated measurement contract.
• Outputs can inform CLV/CAC Analysis (Framework 4) and journey operations, but channel CAC remains dependent on allocation assumptions.
• Use the broader triangulation section below to reconcile attribution with experiments, marketing-mix models, and finance rather than averaging incompatible estimates.

Overview

Pricing connects customer value, willingness to pay, demand, competitive alternatives, costs, channels, and strategic positioning. Test the decision for the target segment and context rather than treating one pricing method as universally best. ^[8]

How to Apply

1. Define the price decision: Specify the target segment, use case, value metric, competitive alternatives, objective, time horizon, and constraints before selecting a pricing model.
2. Estimate willingness to pay and economics: Combine customer and market evidence with cost, capacity, margin, tax, contract, fairness, and cash analysis; state uncertainty and alternative explanations.
3. Test and govern the offer: Use bounded price research or experiments where lawful and appropriate, monitor comprehension, access, churn, margin, bill shock, and competitor response, and stage changes with a stop rule.

Pricing-posture matrix: This matrix positions your price based on the perceived value you deliver versus competitors.

• Premium posture: Higher perceived differentiation and lower direct comparability may support a higher price, subject to willingness-to-pay evidence.
• Value posture: In a more comparable market, offer a defensible value-price relationship for the target segment.
• Economy posture: A lower price requires a cost and operating model that can sustain it.
• Penetration hypothesis: A temporary lower price may build trial or network participation, but future price response, customer fairness, cash, competition, and regulation must be tested.

Figure 5.2. Constructed pricing-posture map. The anonymous points illustrate how teams might discuss perceived differentiation and competitive intensity; the coordinates are not market evidence or recommended prices.

quadrantChart
    title Pricing Strategy Matrix
    x-axis Low Competition --> High Competition
    y-axis Low Perceived Value --> High Perceived Value
    quadrant-1 Value Pricing
    quadrant-2 Premium Pricing
    quadrant-3 Penetration Pricing
    quadrant-4 Economy Pricing
    Offer A: [0.3, 0.8]
    Offer B: [0.7, 0.7]
    Offer C: [0.8, 0.2]
    Offer D: [0.3, 0.3]

Text equivalent: The horizontal axis runs from lower to higher competitive intensity and the vertical axis from lower to higher perceived differentiation. Four anonymous, illustrative offers occupy the premium, value, economy, and penetration discussion regions; actual placement requires customer and market evidence.

Source note: Original teaching matrix informed by open marketing-principles material; positions and labels are illustrative rather than empirical. ^[8]

Modern pricing models for digital products:

1. Cost-Plus Pricing: Adds a margin to defined cost. It can support cost recovery or regulated/contract contexts but does not by itself measure customer value or demand.
2. Competitor-Based Pricing: Setting your price based on what competitors charge. A common starting point, but it anchors you to their value proposition, not yours.
3. Value-Based Pricing: Uses evidence about differentiated customer value and willingness to pay. Validate the counterfactual, attribution, segment, implementation cost, risk, and buyer budget rather than applying a fixed share-of-value rule.
4. Usage-Based Pricing: Customers pay based on a measured unit of consumption. It aligns price with customer value only when the unit is a credible value metric; it can also create unpredictability, bill shock, gaming, weak cost recovery, or procurement friction. Test the metric, meter accuracy, fairness, caps, customer control, margin, tax, and contract implications.
5. Tiered Pricing: Multiple packages can differentiate features, service, limits, or outcomes for distinct use cases. Tiering is a hypothesis, not proof of segmentation or an automatic upsell path; test comprehension, willingness to pay, fairness, migration, cannibalization, support burden, and unit economics.

So What for Managers

• Choose a pricing method that fits the segment, value metric, competitive context, operating model, and decision objective rather than copying a market convention.
• Make the customer trade-off legible and protect agency with clear terms, controls, access, fairness checks, and escalation for material harm or bill shock.
• Connect price changes to marginal contribution, retention, capacity, cash timing, and customer outcomes; monitor the realized result after launch.

Limits and Critiques

• Perceived value, willingness to pay, elasticity, and competitive intensity are context-dependent and must be measured; an illustrative matrix does not establish market position.
• Cost-plus, competitor-based, value-based, usage-based, and tiered pricing each encode assumptions and can fail through weak value metrics, gaming, unpredictability, cannibalization, margin leakage, or procurement friction.
• Pricing experiments can create fairness, disclosure, legal, privacy, and trust risks; do not infer long-run demand or customer welfare from a short-term conversion lift alone.

Connections

• Inputs include Jobs-to-be-Done (Framework 2), Brand Architecture (Framework 7), competitive analysis, customer research, cost and capacity data, and finance assumptions.
• Outputs feed CLV/CAC Analysis (Framework 4), channel and go-to-market choices, and the customer-data and experimentation governance boundary.
• Use A/B Testing (Framework 8) or a justified pricing-research design only when the treatment, eligible population, outcome, guardrails, and legal basis are explicit.

Overview

Brand architecture organizes a brand portfolio by specifying brand roles and relationships among brands, including sub-brand and endorsed-brand alternatives. It can clarify the intended structure, but the architecture alone does not guarantee a business outcome. ^[9]

How to Apply

1. Choose a Model:
   • Branded house: A master brand covers multiple offers. Potential benefits include recognition and shared investment; risks include spillover from a failure.
   • House of brands: Distinct brands serve different positions. Potential benefits include separation and targeting; costs include portfolio complexity and duplicated investment.
   • Hybrid: Endorsement and independence vary by offer. The design requires explicit naming, governance, migration, and risk rules.
2. Map Your Portfolio: Visually map your current products and brands to see which architecture you are currently using (or if you have an inconsistent mix).
3. Align with Strategy: Compare customer recognition, position, channel, portfolio relationships, reputation spillover, operating cost, acquisition integration, and governance. Select an architecture as a hypothesis and measure its effects rather than labeling one model “best.”

So What for Managers

• Make brand roles, naming, endorsement, migration, investment, and failure-containment decisions explicit across the portfolio.
• Treat architecture as a design hypothesis and measure recognition, consideration, choice, willingness to pay, retention, spillover, cost, and governance outcomes.
• Assign an owner for portfolio coherence and review architecture changes alongside strategy, product, channel, and acquisition decisions.

Limits and Critiques

• Branded-house, house-of-brands, and hybrid models involve trade-offs; no architecture is universally best and the architecture choice is not evidence of a market outcome.
• Architecture can clarify relationships without causing efficiency, customer-based brand equity, or competitive advantage; those effects require focal-firm evidence and alternative explanations.
• Author synthesis: Brand knowledge and survey measures can be noisy, culturally specific, and disconnected from behavior or cash flow; protect respondents and test the full evidence chain.

Connections

• Inputs include Corporate Strategy (Chapter 3), target Customer Segments (Framework 3), acquisition or portfolio plans, customer research, channel evidence, and governance constraints.
• Outputs inform naming, endorsement, migration, allocation, and risk decisions and connect to the brand-equity evidence chain below.
• Use Pricing Strategy & Models (Framework 6) and Customer Journey Mapping (Framework 1) to test whether architecture changes affect perceived value and experience rather than assuming they do.

Overview

An online controlled experiment randomly assigns eligible units to variants to estimate a treatment difference under the experiment's design and trustworthiness assumptions. Define the tested contrast explicitly. ^[10]

How to Apply

1. Formulate the Decision and Hypothesis: Define the eligible population, unit of randomization, treatment, primary outcome, rationale, and action the result can change.
2. Predefine the Design: Specify the minimum detectable effect, sample and duration, allocation, guardrails, segment analyses, multiplicity approach, stopping rule, and data-quality checks.
3. Run the Test: Randomize with an approved experimentation system; monitor sample-ratio mismatch, instrumentation, interference, novelty, operational incidents, and privacy/safety guardrails without opportunistic peeking.
4. Analyze Results: Report the estimated treatment effect and uncertainty, experiment-validity checks, practical relevance, and guardrail results; do not reduce the decision to a binary “winner.” ^[10]
5. Decide and Stage: Ship, reject, redesign, or declare the result inconclusive using the pre-specified rule. Roll out gradually when consequences or uncertainty warrant it.
6. Monitor Durability: Verify that the effect and guardrails persist after rollout; record learning and update the next hypothesis.

Figure 5.3. Marketing experiment decision loop. The flow separates validity and trustworthiness checks, decision relevance, guardrails, and staged rollout rather than equating statistical significance with shipping. ^[10]

flowchart TD
    A[Decision and Customer Evidence] --> B[Pre-specified Hypothesis and Design]
    B --> C[Randomize and Run]
    C --> D{Data and Design Valid?}
    D -->|No| E[Repair or Stop]
    D -->|Yes| F{Effect Detectable and Material?}
    F -->|No| G[Inconclusive, Reject, or Redesign]
    F -->|Yes| H{Guardrails Acceptable?}
    H -->|No| G
    H -->|Yes| I[Stage Rollout]
    I --> J[Monitor Durability]
    E --> B
    G --> B
    J --> A

    style A fill:#4ecdc4
    style D fill:#ffd93d
    style F fill:#ffd93d
    style H fill:#ffd93d
    style I fill:#95e1d3

Text equivalent: Define the decision and pre-specify the experiment, randomize and run, verify data and design, assess whether the effect is both detectable and practically material, check guardrails, then stage and monitor rollout. Invalid, unsafe, weak, or inconclusive evidence routes to repair, rejection, or redesign rather than an automatic winner.

Source note: Original decision loop informed by the Cambridge guide's treatment of trustworthy analysis, guardrail metrics, decision pitfalls, and ramping exposure. ^[10]

So What for Managers

• Define the decision, eligible population, treatment contrast, primary outcome, minimum detectable effect, guardrails, stopping rule, and action before exposure begins.
• Check randomization, delivery, sample-ratio mismatch, instrumentation, interference, novelty, multiplicity, practical materiality, and customer-safety constraints.
• Stage a supported change and monitor durability; ship, reject, redesign, or declare inconclusive according to the pre-specified decision rule. ^[10]

Limits and Critiques

• An experiment estimates an effect for its tested alternatives, population, implementation, and time window; it does not prove that the tested set contains the best feasible design.
• Statistical detectability is not business materiality, and a null or wide result is not proof of zero effect; power, attrition, interference, and measurement quality shape the conclusion.
• Qualitative research, Jobs-to-be-Done (Framework 2), Customer Journey Mapping (Framework 1), prototypes, and broader redesigns can generate alternatives, but none guarantees a “global maximum.”

Connections

• Inputs include opportunities from a Customer Journey Map (Framework 1), behavioral and qualitative evidence, operational constraints, risk review, and the customer-data boundary.
• Outputs can inform a staged change; any effect on conversion, retention, or the CLV/CAC ratio (Framework 4) requires follow-up measurement and cost attribution.
• Reconcile treatment effects with marketing attribution, marketing-mix models, and finance in the measurement-triangulation protocol below.

Overview

Net Promoter Score (NPS) is a widely used metric to measure recommendation intent. It is based on a single 0–10 question; respondents are grouped into Promoters (9–10), Passives (7–8), and Detractors (0–6), and the score is % Promoters - % Detractors. ^[11] A driver analysis can then examine potential explanations for loyalty or dissatisfaction without treating the score as a causal measure.

How to Apply

1. Ask a Bounded Follow-Up: When appropriate, invite respondents to explain the score without forcing sensitive disclosure or assuming the response is causal.
2. Tag Verbatim Feedback: Categorize the open-ended responses into themes (e.g., "Ease of Use," "Pricing," "Customer Support," "Feature X"). Use text analytics tools for large volumes.
3. Associate Themes with Scores: For each theme, summarize score, response, segment, and uncertainty patterns. Theme-score relationships may reflect selection and do not establish a driver.
   • Driver of Delight: A theme mentioned frequently by Promoters (e.g., "Customer Support" has an average NPS of 80).
   • Driver of Dissatisfaction: A theme mentioned frequently by Detractors (e.g., "Pricing" has an average NPS of -50).
4. Prioritize Learning: Combine frequency, severity, affected customers, accessibility, behavioral evidence, feasibility, and risk; test whether a proposed improvement changes the intended outcome.

So What for Managers

• Use NPS as one standardized sentiment signal, then inspect response coverage, verbatims, behavioral retention, repurchase, complaints, accessibility, and economics.
• Treat themes as hypotheses about experience or dissatisfaction; prioritize them using frequency, severity, affected customers, feasibility, risk, and a testable outcome.
• Pair any proposed change with an owner, guardrail, and follow-up measure rather than managing to the score alone.

Limits and Critiques

• NPS is a recommendation-intent score and is sensitive to response/nonresponse, survey mode, customer mix, culture, timing, and instrument design. ^{[11] [12]}
• Keiningham and colleagues did not replicate the claim that NPS is the single most reliable indicator of firm revenue growth; that study does not establish individual future behavior. ^[12]
• Theme-score associations can reflect selection, common causes, coding choices, and reverse causality; they do not identify a driver without stronger design and triangulation.

Connections

• Inputs include customer surveys, verbatim feedback, response/nonresponse diagnostics, behavioral retention, complaints, and service evidence.
• Outputs generate hypotheses for Customer Journey Mapping (Framework 1), A/B Testing (Framework 8), product improvement, and service recovery.
• Reconcile sentiment with Cohort Analysis (Framework 10) and CLV/CAC Analysis (Framework 4) without treating either relationship as automatically causal.

Overview

Cohort analysis groups customers by a defined starting event or characteristic and follows a consistently defined outcome over time. A retention table aligns cohorts by time since the starting event so unlike-aged groups are not blended. It reveals descriptive differences that can generate hypotheses; it does not by itself show that a product change caused retention or that the product is getting “stickier.” ^[13]

How to Apply

1. Group Users by Cohort: Typically, group by the month or quarter they signed up.
2. Create a Retention Table: Create a triangle chart where each row is a cohort and each column is the month since that cohort joined (Month 0, Month 1, Month 2, etc.). The value in each cell is the percentage of the original cohort that is still active in that month.
3. Analyze the Curves:
   • Curve shape: A flattening curve describes persistence under the chosen activity definition. Check censoring, seasonality, reactivation, customer mix, and whether the outcome represents value.
   • Comparing cohorts: Higher or lower curves can reflect product changes, acquisition mix, pricing, seasonality, measurement, or external conditions. Use an experiment or justified causal design before attributing the difference.

So What for Managers

• Define the cohort event, activity or value outcome, observation window, censoring rule, and decision before comparing retention curves.
• Use the pattern to identify where to investigate product, acquisition, pricing, service, or customer-mix differences and to size follow-up evidence.
• Translate a proposed retention action into a bounded test or justified causal design and monitor customer value, quality, and agency alongside retention.

Limits and Critiques

• Higher or flatter curves can reflect product changes, acquisition mix, pricing, seasonality, external conditions, reactivation, censoring, or measurement changes.
• Activity definitions are not interchangeable with customer value or product-market fit; a retention pattern is descriptive evidence, not a causal conclusion.
• Cohorts can be small, immature, selected, or unlike the target population; report uncertainty and avoid universal retention benchmarks. ^[13]

Connections

• Inputs include sign-up or starting-event dates, activity and transaction data, product instrumentation, cohort definitions, and data-quality checks.
• Outputs supply definition-dependent retention estimates for CLV/CAC Analysis (Framework 4) and product decisions, and can guide Customer Journey Mapping (Framework 1).
• Reconcile cohort patterns with NPS Driver Analysis (Framework 9), experiments, acquisition attribution, pricing, and finance before changing material spend.