Operations
Part of Customer analytics: what to keep and what to drop
Nine customer analytics patterns with questions, actions and data checks
Nine customer analytics patterns, each with a guiding question, an action and the data checks to run before anyone acts on the result.
What to take away
- Each pattern uses a different customer unit, outcome and action.
- Paths show observed sequences, not motives or causes.
- Public demographic tables can add context but should not be attached to individuals.
Customer analytics work should start from a decision pattern, not a promised result. The nine patterns below can be adapted only after the customer unit, data rights, data quality, economics and delivery limits are checked.
| Pattern | Question | Action |
|---|---|---|
| Onboarding cohort | Who reaches first value? | Improve one step or message |
| Repeat purchase | When do buyers return? | Time replenishment help |
| Subscription renewal | Which signals precede renewal? | Prioritize review or test |
| Service recovery | Which issues recur? | Fix process and follow-up |
| B2B account growth | Which accounts adopt another use? | Coordinate product and sales help |
| Feature adoption | What sequence precedes durable use? | Change onboarding |
| Store and online | How do covered channels differ? | Adjust local experience |
| Donor retention | Which entry cohorts continue? | Test stewardship |
| Marketplace health | How do buyer and seller outcomes interact? | Balance both sides |
Data sources and checks for each pattern
- Onboarding cohort:Use product or Google Analytics 4 event data. Define first value as one named event before the analysis, and start each cohort on the account creation date.
- Repeat purchase:Use store order exports. Remove refunds and returns from the count, and match guest and account buyers with one identity rule.
- Subscription renewal:Use Stripe subscription lifecycle events. Code cancellations, failed payments and pauses as separate outcomes.
- Service recovery:Use ticket exports from a helpdesk such as Zendesk or Freshdesk. Count a reopened ticket once, and measure resolution time from one agreed start point.
- B2B account growth:Use HubSpot or Salesforce account records. Take seat and product counts from billing, not from the CRM.
- Feature adoption:Use event data from Amplitude or Mixpanel. Set the activation event before the analysis starts, never after.
- Store and online:Join point-of-sale exports to online orders using a shared customer ID. If no shared ID exists, label the join as approximate.
- Donor retention:Use the organization's own donor database. Count gifts by donor and entry cohort, and use only data the donor provided for that purpose or the law permits.
- Marketplace health:Use payout exports from a marketplace payments platform such as Stripe Connect. Use the same active-period definition for buyers and sellers.
No benchmark is assumed here. Set a local baseline from four to eight recent periods before judging any change. Ranges quoted in earlier drafts had no source, so they are not used.
Demo accounts are for practice
Public demo accounts, such as the Google Analytics 4 demo account for the Google Merchandise Store and Amplitude's demo environment, are useful for practising a query. They say nothing about any company, so nothing in them should be reported as a result.
Nine Customer Analytics Patterns
Pattern
- Onboarding cohort
- Who reaches first value?
- Repeat purchase
- When do buyers return?
- Subscription renewal
- Which signals precede renewal?
- Service recovery
- Which issues recur?
- B2B account growth
- Which accounts adopt another use?
- Feature adoption
- What sequence precedes durable use?
- Donor retention
- Which entry cohorts continue?
Question
- Onboarding cohort
- Improve one step
- Repeat purchase
- Time replenishment help
- Subscription renewal
- Prioritize review or test
- Service recovery
- Fix process and follow-up
- B2B account growth
- Coordinate product and sales
- Feature adoption
- Change onboarding
- Donor retention
- Test stewardship
Action
- Onboarding cohort
- Repeat purchase
- Subscription renewal
- Service recovery
- B2B account growth
- Feature adoption
- Donor retention
Add population context carefully
The Bureau of Labor Statistics publishes Consumer Expenditure Survey tables with aggregate expenditure information across published population groups. These tables can provide broad context. They do not identify a company's customers and do not authorize individual-level enrichment.
- Write the outcome and the observation window
- Identify who is absent from the data
- Set the action and the stop condition
- Measure treatment delivery and suppression
- Use an experiment for effect claims
- Review mature contribution and unintended outcomes
Avoid one generic dashboard for all nine patterns. Service recovery may use an issue as the unit, B2B growth an account, and renewal a subscription. Mixing them can create impossible denominators and misleading averages.
Document one falsifying observation for each pattern. If the data cannot tell a product defect from a customer trait, do not label the customer. Escalate the process problem instead. Labelling a person for a defect the product caused is among the costlier customer analytics mistakes.
Turn patterns into test cases
Record each pattern's audience, source data, identity rule, business constraint, decision and outcome window before transfer. Replace the example inputs with local data and test a failure case. If a pattern depends on rights, staff or product functions the organization lacks, label it unsuitable.
The NIST Privacy Framework starting guide describes an organizational process for identifying and managing privacy risk. Use it to assign owners, describe data processing and record responses. The framework is voluntary and does not establish legal compliance for any marketing use.
Keep the evidence record beside the decision so a reviewer can reproduce the reasoning. Set the next review date and name the change that would trigger an earlier check.
