Lifecycle programme measurement

Measure a lifecycle programme by asking who was eligible for help, whether they made the intended progress and whether the programme contributed. A send count cannot answer those questions alone. Define the customer outcome, counting unit and observation period before reading the report.

Build a measurement map

For each journey, record the customer task, entry rule, response and the outcome that would show useful progress. Label whether each measure counts people, accounts, orders or agreements. If messages count recipients but purchases count accounts, do not divide one figure by the other.

The first four rows describe what happened. The last asks a causal question and needs stronger evidence.

Funnel analysis in Amplitude counts users who complete a defined sequence of events, in order, within a window you set. Before comparing rates, check event order, filters, grouping and conversion window; a tool's conversion definition may not match the business definition.

Key metrics for lifecycle programme measurement

Define outcomes before reporting them

Choose an outcome that fits the journey's job. Onboarding might support a confirmed first useful task; renewal communication might support a verified decision before the relevant deadline. An email open can diagnose engagement, but it does not establish either result.

Record the event definition, source system, identity, occurrence time and treatment of corrections. State whether an attempt counts and whether the result belongs to a person or an account. A missing event means the result is unconfirmed; it does not always mean the customer failed.

Keep definitions and windows stable within a reporting series, and mark any change. Amplitude's default conversion window is one day in UTC, so a quoted rate may reflect that default rather than your intended outcome window.

Compare entry cohorts fairly

Group customers by when they became eligible, and give each cohort the same opportunity to reach the outcome. Mark a recent cohort whose window remains open as pending. Report eligible units, confirmed results and data exceptions alongside rates.

For time to value, show both the share reaching the result by a fixed point and elapsed time among those who reached it. A faster median among completers can coexist with fewer customers reaching value. For stage conversion, label whether the denominator is the previous stage or the original entry cohort; those rates answer different questions.

Amplitude defines total conversion as users who triggered every step in the funnel divided by users who triggered the first step, effectively an entry-cohort rate. Label stage rates separately so readers can distinguish the denominators.

Diagnose exceptions and implementation

A process evaluation examines how a programme operates as implemented, including the activities actually conducted. Pair every outcome measure with an implementation measure, so a weak result can be separated from a journey that never ran as designed.

Use conversion to assess completion, conversion over time to track rate trends, time to convert to examine elapsed time, and frequency to assess repeat completion. Use the largest drop-off view to locate the greatest absolute loss; use the slowest-conversion view to identify the step with the longest median transition time.

Report what was filtered out beside the rate. Check whether a condition narrows the population entering the measurement or only a single step; the two change what the denominator means.

Separate contribution from attribution

A platform may credit a conversion to a message under its configured rules. That is an attributed association. Customers may have progressed without the message, and people who engage with a journey may differ from those who do not.

Where feasible, assign eligible people or accounts to the journey or a suitable holdout before the marketing intervention. Keep necessary service communication available to both groups. Compare the same verified outcome over the same period; check assignment, exclusions and other campaigns before interpreting any difference. If no fair comparison is available, report the trend and plausible competing explanations without calling it an effect.

Finish the review with a recorded decision: keep, revise, investigate or stop the journey. Keep the definitions and evidence beside that decision so the next review remains comparable.

Statsig describes a holdout as withholding a small percentage of users, typically 1% to 2%, from a set of new features so their combined effect on a metric can be measured over months.