Retention Audit Checklist
A retention audit is a structured examination of every stage that turns first visits into a durable customer base, data, measurement, action, and ownership. Work through the five stages below in order, marking each check as holding, weak, or missing. The pattern of misses is the diagnosis: most businesses fail the same two stages, and knowing which two is worth more than any single metric. No scores are invented here; your evidence is your own records, read honestly.
Stage 1 (CONNECT): is your data worth analyzing?
Data completeness. Do visit records cover the full base and the full lookback (at least two full cadence cycles)? Gaps: a migrated system, an unlogged location, invalidate everything downstream built on them.
Customer identity quality. One client, one identity: duplicates merged, name changes resolved, household vs individual decided. Duplicate identities inflate acquisition and deflate retention simultaneously: the worst kind of error, wrong in both directions.
Visit history quality. Completed visits distinguished from bookings, no-shows, and cancellations; dates in the business reporting calendar; attendance status trustworthy. If “booked” and “kept” are the same field, stop the audit here, measurement built on this cannot distinguish intention from behavior (see why the distinction matters).
Revenue linkage (if LTV matters to you). Spend per visit joinable to visit records, margin known. Without it, lifetime-value claims are estimates, label them so (see honest LTV).
Stage 2 (UNDERSTAND): do you know your return reality?
First-to-second performance. Measured with explicit window, eligibility, and maturity rules, not vibes. See the framework.
Cadence quality. Per-service expected windows defined, due/overdue thresholds set, exceptions documented. “We sort of know” is missing.
Rebooking measurement. Before-departure vs eventual tracked separately, denominator rules stated. See what good analysis covers.
Overdue and lapsed visibility. A live list of who is past deadline, ranked, with evidence, not a monthly report nobody opens.
Segmentation. Return behavior cut by location, practitioner, service, and cohort. Blended-only measurement is a finding in itself: it means you cannot know where the problem is.
Benchmark posture. Own baseline trended under frozen definitions; no dependence on quoted industry averages of unknown provenance (see benchmark method).
LTV context. Retention read alongside value where revenue data exists, are you keeping the right clients, or just the easy ones?
Stage 3 (ACT): does understanding become action?
Follow-up coverage. What share of misses gets a specific touch within 48 hours? Below ~full coverage, the queue is theater.
Win-back logic. Segmented audiences, brand-safe bounded offers, waves not drips, or an honest admission that win-back is currently “whoever we remember.” See win-back strategy and reactivation discipline.
Ownership of actions. Every queue item has a named owner and a deadline. “The team” owns nothing; name names.
Automation guardrails. Consent current, approval tiers written, per-client caps enforced, kill switches tested, monthly audits read. See automation guardrails. Unreviewed bulk sending discovered here is a stop-everything finding.
Stage 4 (MEASURE): do you know what worked?
Outcome tracking. Acted-on cohorts compared against holdouts over full cycles, not before/after anecdotes. Without holdouts, every improvement claim is a story.
Definition stability. Same numerators, denominators, and windows as last cycle. Redefinition between measurements is how programs manufacture success.
Reconciliation. Start − churned + new = end, every period. Numbers that do not reconcile are not measurements.
Lag respect. Process metrics judged in weeks, business outcomes in full cadence cycles. Monthly verdicts on quarterly phenomena are noise-chasing.
Stage 5 (IMPROVE): does learning change operations?
Location and service differences acted on. Audit findings converted into scripts, schedules, staffing, or capacity, with dates. Findings without operational changes are a book club.
Exit learning loop. Departure reasons captured lightly, aggregated quarterly, reviewed by someone empowered to change things.
Loop cadence. The full CONNECT → UNDERSTAND → ACT → MEASURE → IMPROVE cycle runs at least quarterly, with weekly ACT cadence inside it. Annual audits discover last year's problems; loops prevent next quarter's.
Reading your results
Misses concentrated in CONNECT: stop analyzing and fix data first. Every downstream number is currently suspect, say so explicitly rather than acting on it.
Misses in UNDERSTAND: you are flying blind with instruments. Define windows, segment, baseline: one focused month of measurement work.
Misses in ACT: the classic profile, good dashboards, no decisions. Name owners, start the weekly review, approve the first ten actions personally.
Misses in MEASURE: effort without evidence. Add holdouts and reconciliation before expanding anything.
Misses in IMPROVE: learning without change. Tie one operational change per cycle to one audit finding, minimum.
Retake quarterly. The audit is not the product: RetentionTek's workspace exists to make the passing state permanent rather than periodic, but a rigorous annual-or-better audit is the cheapest retention investment a business can make.
Frequently asked questions
What is a retention audit?
A structured examination of every stage that turns first visits into a durable base: data quality, return measurement, action discipline, outcome tracking, and operational learning. The pattern of weak stages is the diagnosis.
How often should you audit retention?
The full cycle at least quarterly, with weekly action cadence inside it. Annual audits discover last year’s problems; continuous loops prevent next quarter’s.
What is the most common audit failure?
Strong dashboards with no action discipline: understood numbers, no named owners, no weekly review, no approved next steps. Measurement without the ACT stage is overhead.
Where should a failing audit start?
At the earliest failing stage, in loop order. Data failures invalidate everything downstream, so CONNECT problems outrank all others; within action, ownership outranks tooling, name owners before buying software.