Measuring customer analytics: the contract that has to come first

Begin with the decision the number must support. Customer analytics needs a repeatable contract that carries the same customer definition from source records through analysis to action, while setting limits on the claim. Write that metric contract before choosing retention, conversion, value, or a dashboard.

If nobody can name the decision

If nobody can name the business decision, do not choose a KPI yet. State the business question and the claim the evidence could support, then choose the customer unit and decision together before settling on a rate or dashboard.

Use the permitted claim to set the boundary for interpretation. Then order the work from customer and identity rules to event definitions, eligibility filters, reconciled totals, cohort or funnel logic, quality review, an owned action, and a later outcome check with correction.

• Rule of thumb: Set the decision and permitted claim before selecting a metric.

If a tool and the business count different people

If a reported total comes from a tool identifier, check whether it matches the company's governed customer before calling it a customer count. Tool-defined users and governed customers can both be useful, but they answer different questions in routine reporting.

Google Analytics documents a user-lifetime exploration that reports behavior and value across a user's lifetime in that product, subject to documented data and date requirements. Its user construct belongs to that product; do not assume it matches a company's governed customer definition.

Adobe Analytics, Amplitude, Mixpanel, and customer data platforms such as Segment each establish a user or profile identifier, then report according to that definition rather than a governed customer. Those figures can serve product and campaign questions; the customer and identity contract reconciles them to the number the business is prepared to defend.

• Rule of thumb: Treat a tool count as a governed customer count only when the identity contract supports the match.

If privacy rights limit who can be counted or contacted

If rights or permissions determine who may be counted or contacted, record those limits in the identity contract. Document the applicable basis, customer rights, and fields that must be removed or suppressed before activating a result involving personal information.

Data protection requirements differ by jurisdiction. In the European Union, the GDPR sets the lawful basis, consent, and deletion rights a customer record must respect. In California, the CCPA as amended by the CPRA adds disclosure, opt-out, and deletion duties, while many US states have their own privacy statutes.

GAO's evaluation design guide explains how evaluation questions, designs, data collection, and analysis should align in federal program work. It is not a customer-analytics rule; it matters here because evidence should fit the question before a customer claim is made.

• Rule of thumb: Record the applicable rights and permissions before activating customer-level results.

If a number must be compared across periods

If a number will be compared across periods, publish its rules before interpreting movement as performance. Make the contract a short written specification with a version and named owner, not a legal agreement prepared for each release.

Capture the terms that determine how the measure is produced and interpreted:

A data contract between a table producer and consumer is a useful analogy; dbt model contracts offer one way to record it. Make the customer, household, or account unit explicit, alongside event time, processing time, time zone, and the treatment of late data.

Activation is the share of new customers completing a defined first-value event. Retention measures the share of a base-period customer group still active later; churn is the share of that group that stops. Repeat rate counts customers buying again within the window, while customer lifetime value is observed or predicted contribution over a stated horizon.

Publish a compact metric card with the definition, source, refresh time, owner, known gaps, and permitted use. Preserve the prior version when identity rules or denominators change. An alert should state its review condition and identify who decides whether publication or activation continues; dashboard placement affects whether people see it.

• Rule of thumb: Publish the unit, denominator, window, and maturity rules with every comparable rate.

If the decision depends on proving an action's effect

If the decision depends on whether an action caused an effect, add random assignment or another credible counterfactual. Preserve the assigned, treated, eligible, and analyzed populations, then compare incremental contribution with the action's cost and capacity instead of relying on an observed difference alone.

Track profile match rate, unresolved conflicts, missing periods, late outcomes, cohort maturity, denominator movement, model error, segment size, treatment delivery, suppression success, and correction rate. These indicators expose process limits, but none by itself proves customer impact. Make those limits visible during quality review.

Industry bodies can sharpen a definition without settling it. The Media Rating Council accredits audience measurement and sets minimum standards; the Interactive Advertising Bureau publishes digital measurement and ad-viewability guidelines; and the Digital Analytics Association represents analytics practitioners. None defines your customer, which remains an internal decision.

Attribution also depends on a stable contract: last-click gives all credit to one touch, linear divides credit evenly across touches, position-based weights the first and last touches more heavily, and data-driven models allocate credit by modeled contribution. Each model needs a stable customer key and one defined conversion, so an unstable customer unit makes every attribution number unstable.

• Rule of thumb: Match the evidence and customer identity to the claim; a process indicator alone does not prove impact.

If another person or tool must act on the result

If a person or tool must act on the result, make every handoff observable. Assign each step to a named role, require the analyst to reproduce the result, and require the decision owner to explain the action and its stop condition.

Carry definitions, source, code or rules, quality limits, results, action, approval, and the scheduled later review into the handoff. That material lets the next owner reproduce the measurement and understand known limits, rather than receiving only a headline number or an undocumented instruction.

Test one broken-data case and one revoked-access case during handoff. Record the repair, elapsed time, and any vendor help so the written process reflects ordinary operation rather than a prepared demonstration. Keep both cases with the code or rules used to produce the result.

GOV.UK technology-selection guidance recommends choices that can change over time, preserve data control, address security risk, and include ownership cost. Those public-service principles can inform buying questions, but they are not private-sector mandates or product endorsements. Treat them as considerations rather than obligations when evaluating customer analytics tools.

CISA's software acquisition fact sheet points buyers toward development practice, supply-chain exposure, deployment, and vulnerability management. Add those topics to a customer-measurement tool review. The government context does not replace a buyer's own technical testing, contract review, or risk decision.

• Rule of thumb: A handoff is ready when another owner can reproduce the result, explain its limits, and identify when to stop.

If the observation or outcome window is still open

If observation or outcome windows are still open, label the result incomplete rather than final. A cohort is mature only after both the observation period and the outcome-maturity window have closed for the defined metric.

Choose the measure closest to the named decision and governed outcome, often activation, repeat use, retention, service resolution, or contribution. Keep recent cohorts marked incomplete until the relevant windows close; do not compare them as though they had equal time to produce an outcome.

Set the next review date and name the change that would trigger an earlier check. Measures that survive review are often the marketing KPIs the company already reports upward. Keep the prior contract version available when a denominator or identity rule changes, so the decision owner can see exactly what changed.

• Rule of thumb: Do not call a cohort mature before its observation and outcome windows close.