Reading Churn Signals and Making Save Offers Honestly

The save offer should match the reason for leaving, and the reason is usually visible before the cancellation: declining usage, a team shrinking, a feature never adopted, a billing surprise. Monitor a short list of signals, reach out with help specific to the signal (a training session for the unused feature, a plan downgrade for the shrunken team, a pause for the seasonal business), and reserve discounts for the cases where price is genuinely the stated, plausible reason. Blanket exit discounts train your best customers to threaten cancellation annually, and they save the accounts least likely to stay.

Four signals carry most of the predictive weight for a small SaaS: login and core-action frequency trending down over 30 to 60 days; seat or usage volume reductions; support contact that ends unresolved or turns frustrated; and payment failures or card expiries approaching. None requires a data science team - a monthly export and an hour of honest review surfaces them. Segment the watch: a seasonal business going quiet in its off-season is not churning, and treating it as such wastes the outreach it will actually need at season start.

The first save move is a question, not an offer: a short, human message from a named person asking what changed. The answers sort accounts into the real buckets - outgrown the product, never got value, price pressure, internal change (the champion left), or pausing. Each bucket has a different honest save, and two of them have none: an account that outgrew the product or never fitted should be helped to leave well, because a graceful exit preserves referrals, reviews and the occasional return, while a trapped account produces the opposite of all three.

Match the offer to the bucket. Never got value: an onboarding rescue - a session that does the setup with them, aimed at the activation they missed. Shrunk team: a downgrade to the fitting plan, offered before the renewal makes it a cancellation. Seasonal: a pause that keeps data and settings, with a stated restart. Price pressure with genuine fit: a discount can be legitimate - time-limited, explained, and recorded so the pattern is visible if it recurs. Champion left: an introduction offer to the successor, with a handover summary of how the account used the product.

A save is not a save until the account is alive and paying a quarter later. Report save attempts, acceptance, and 90-day survival by bucket and offer; the menu item that 'saves' everyone and retains no one is a delay machine, converting clean cancellations into resentful later ones plus refund risk. This longer window also disciplines the discount habit: discounts usually show the worst 90-day survival of any save, which is the finding that finally retires them as the default.

A large share of small-SaaS churn is involuntary: expired cards, replaced cards, bank declines. It deserves its own routine because it is fixable without persuasion: pre-expiry notices, clear in-product warnings, retry logic that respects the bank's timing, and a grace period that keeps the product usable while the card is fixed - with the state of the account communicated plainly throughout. Measure recovered revenue from failed payments separately; it is often the cheapest retention money the business ever finds, and it is invisible in cancellation surveys because the customer never decided anything.

Signals mislead when read account by account without the cohort behind them. An account whose usage dips in its second month is behaving normally for products with a learning curve; the same dip in month nine is a warning. Build the habit of comparing an account's curve to its signup cohort's typical curve: deviation from the cohort, not the raw direction, is the signal. This needs no tooling beyond the monthly export and a drawn average, and it prevents the two classic errors - panicking over normal settling, and missing the quiet fade of a mature account because its absolute usage still looks respectable next to new trials.