One of the actual retention moves you probably arent making.

By Ron Davenport. Published 2026-05-23.

Tbh, this is the part most teams skip. 20-40% of subscription SaaS churn is failed payments nobody is actually recovering. The other half is hiding in the cohort retention curve that most teams I work with never actually pull.

Heres what happens on every retention engagement Im in. Someone shows me a 12-month churn rate and asks what to ship. My answer every time: pull the curve. The number on its own isnt enough.

What follows: how to read each of the four shapes the curve can make plus the dunning math for the failed-payment side.

These four show up across pretty much every subscription SaaS. Learning to spot them quick is what separates a retention conversation that goes somewhere from one that turns into a quarterly ritual.

Cliff (week 1).

Your retention line drops 30-60% in the first 7 days, then plateaus or keeps bleeding a little slower. A cliff means activation is broken. People signed up, never hit the moment that shows them why the product is valuable, left. The work is upstream of the welcome series, which means welcome-series copy edits on a cliff curve are mostly wasted effort. You need to redefine the activation event (regression on retained vs churned cohorts), build event-triggered flows that branch on stall points, and stamp the event on the user record. The SaaS onboarding cornerstone covers the full methodology.

Slow bleed (weeks 4-12).

Healthy in week 1, no cliff, but the line keeps declining month over month for the first quarter. A slow bleed means value delivery broke after activation worked. The product was useful initially, then stopped being useful as user context shifted or needs grew. The work is lifecycle programs tied to product behavior. Behavior-triggered re-engagement when usage drops below cohort threshold. Feature-adoption nudges that surface adjacent value. Expansion triggers that pitch the next tier when usage signals saturation. Generic re-engagement (the calendar-based “we miss you” email) doesnt move slow bleed. The lifecycle marketing cornerstone has the trigger architecture.

Seasonal dip.

Retention drops on a predictable cycle and either recovers or doesnt. Some products show this around quarter-ends, holidays, summer, mid-month billing cycles. The diagnostic question is whether the dipped users come back. If yes, the dip is a usage-pattern thing, not actual churn, and the work is timing-aware lifecycle (pre-dip warnings, recovery-point re-engagement, retention-aware billing options like pause-and-return). If no, the dip is hiding actual churn correlated with the cycle and the work is usually upstream of lifecycle entirely.

Plateau and drop (month 6+).

Curve holds steady for 4-6 months and then drops sharply. Trickiest of the four because the early curve looks healthy until it isnt, and by the time you see the drop the cohort is mostly gone. Two common causes: the user hit a ceiling the product cant grow past (expansion failure), or got what they came for and theres no next milestone keeping them paying (engagement decay). The work is expansion lifecycle (tier-upgrade triggers, deeper-feature nudges, multi-product cross-sell, re-onboarding for stale users). Most teams underinvest here because the early curve looked fine.

These arent mutually exclusive. Your product probably has a cliff in one cohort, a slow bleed in another, and a plateau-and-drop in a third. The aggregate curve averages over all of them and shows you nothing useful. Segmenting by cohort surfaces the multi-archetype reality, and the fix depends on which specific archetype each cohort is actually exhibiting.

For each archetype the diagnostic surfaces, theres a specific class of work that actually moves it. Wrong fix for the right diagnosis is the most expensive failure pattern in retention work, which is why the matrix below is the one I run on every engagement. Identify the archetype, run the matched fix, skip the rest.

Cliff at week 1 → activation event redefinition.

The work is upstream of the welcome series. Activation event needs to be the right one, trigger logic needs to fire at the right stall point, data layer needs to surface where users are stuck. Welcome-series copy edits dont move a cliff. The SaaS onboarding cornerstone covers the methodology: pick the activation event with regression, validate with the matrix method, branch the flow by stall point, stamp the event on the user record.

Slow bleed weeks 4-12 → feature-adoption nudges tied to cohort signals.

Your product is working but users arent deepening usage. The work is product-event-triggered re-engagement: a sequence that fires when a key feature hasnt been used in N days, or when usage drops below a cohort threshold, or when an adjacent feature would unlock more value. Calendar-based re-engagement (the “we miss you” email) doesnt move slow bleed. Behavior-triggered does. The lifecycle marketing cornerstone covers the trigger architecture in depth.

Seasonal dip with recovery → anchored re-engagement at the recovery point.

When users predictably dip and predictably return, the work is timing-aware. Pre-dip warning sequences. Re-engagement at the recovery point. Retention-aware billing options like pause-and-return instead of cancel-and-resubscribe. The lift is small but compounds because the population is recurring.

Seasonal dip without recovery → upstream of lifecycle.

When users dip and dont return, the problem isnt lifecycle. Its product-market fit at the use-case level. No lifecycle work fixes a product that doesnt serve the users actual workflow. The honest move is to escalate to product or scope the engagement out, which is what the disqualifier section below is about.

Plateau and drop at month 6+ → expansion lifecycle.

Your user has gotten what they came for. The work is surfacing the next milestone before they hit the renewal decision. Tier-upgrade triggers tied to usage approaching ceiling thresholds. Deeper-feature nudges that reveal the next value layer before the user plateaus. Multi-product cross-sell when the primary product has hit its natural usage limit. Re-onboarding for users approaching renewal who havent engaged in 30+ days. Most teams underinvest here because the early curve looked fine.

Failed-payment involuntary → dunning rebuild.

Separate work from voluntary churn. The intervention class is payment recovery: retry logic optimization, dunning sequence cadence, card-update flows, pre-dunning warning emails. See the SaaS dunning playbook for the 7-stage sequence.

The matrix isnt elegant, its specific. Specific is what wins in retention work.

The metrics teams use to talk about retention arent interchangeable. Each one tells the truth about a different part of the business, and each one lies if read out of context.

Gross Revenue Retention (GRR).

Percentage of recurring revenue retained from a cohort over a period, excluding any expansion. Start a year with $1M MRR from a cohort and end the year at $850K excluding expansion, GRR is 85%. GRR tells the truth about pure retention. It lies if you read it as the only retention metric, because it ignores the offset effect of expansion within the retained customers.

Net Revenue Retention (NRR).

Same denominator (starting MRR from a cohort) but includes expansion the cohort generated. $1M cohort ends at $1.1M after $250K of expansion against $150K of churn, NRR is 110%. Anything over 100% means the cohort grew without you needing to acquire new customers. NRR tells the truth about the cohorts overall financial trajectory. It lies if you use it as a substitute for GRR, because healthy NRR with bad GRR means youre masking high churn with aggressive expansion, which usually hides product problems that catch up with you later.

Revenue churn.

Simplest of the three: percentage of MRR lost over a period. Useful for top-line dashboards and benchmark conversations. It lies if you use it to plan retention work because it doesnt distinguish voluntary from involuntary, doesnt surface cohort patterns, and doesnt tell you which segment is bleeding.

The operator practice is to report all three with cohort segmentation underneath each, since reading any one of them in isolation leads to a different wrong conclusion about whats actually happening inside your retention.