The wrong recovery metric on a slide makes a good program look bad and a bad program look excellent. Here's the shortlist that survives scrutiny — and the ones that don't.
# The five metrics worth reporting
1. Involuntary churn rate (% of MRR)
Failed-payment-driven cancellations, as a fraction of active MRR, per month. This is the number a CFO or board member will actually recognize. Anything under 1% for B2B and under 2% for consumer is credible; well-run programs land at 0.5% and 1.2% respectively.
2. Recovered MRR (absolute dollars)
Not recovery rate. Not attempts. Dollars recovered per month, ideally with a clear counterfactual: "without the recovery program, we'd have lost $X this month; we recovered $Y of it."
3. Recovery rate (recovered ÷ eligible), by day 7
Recovered dollars ÷ total failed dollars, capped at a 7-day window. Adding a time cap is critical — reporting "89% lifetime recovery" is meaningless if half the money took 6 months to come back.
4. Time-to-recovery, p50 and p90
How fast recovered money is recovered. Fast money is more valuable than slow money for cash flow and for retention (churn probability rises with time).
5. Recovery-driven retention lift (cohort)
30-day retention of customers whose payment failed and was recovered, vs. customers whose payment succeeded on the first try. This tells you whether recovery is truly restoring the relationship or just deferring churn.
# The metrics that belong in engineering dashboards, not the board
- Retry attempts per invoice.
- Dunning message send counts.
- Email open rates and CTR by step.
- WhatsApp delivery vs. read rates.
- Failure code distribution.
- Retry timing histograms.
These are operationally important. They are not board-worthy on their own. They inform why the top-line metrics moved, not whether they did.
# The vanity metrics to retire
- "Total lifetime recovered." Grows monotonically. Says nothing about the current state of the program.
- Recovery rate without a time cap. Same problem — asymptotically approaches 100% given enough time.
- Number of emails sent. More is not better. In fact, more is often worse.
- "AI-personalized message count." Vendor slide filler. Doesn't map to money.
# How to structure the board slide
Two charts, one table:
- Involuntary churn rate, trailing 12 months, target line overlaid.
- Recovered MRR by month, trailing 12 months.
- Table: current-month recovery rate, day-7 time-to-recovery p50, cohort retention lift.
Three sentences of context. Names the top driver of month-over-month change. Names one specific initiative in flight.
# How to structure the ops dashboard
For the team running recovery:
- Real-time count of active failed invoices, by segment.
- Recovery rate by cadence step (T0, T1, T2, T3), by failure code, by plan value.
- Deliverability health: spam-complaint rate, bounce rate, DMARC alignment.
- LLM generation health: p50 latency, denylist strike rate, template fallback rate.
- Provider health: webhook delivery lag, retry queue depth.
Different audience, different metrics. The board metrics roll up from these but don't replace them.
# The measurement stack we recommend
- Warehouse first. Ship recovery events into Snowflake/BigQuery/Postgres. Every downstream metric derives from the same source of truth.
- DBT models for the top five metrics. Version-controlled definitions beat spreadsheet math.
- A single "recovery health" dashboard that both engineering and finance look at. Two dashboards means two truths.
# The one metric to lead with
If you get one line on the CEO's weekly review: involuntary churn rate as % of MRR, trailing 30 days. Everything else is scaffolding.
Related: Involuntary Churn Benchmarks 2026 for how your numbers compare to peers.