1. Define the first win as one observable sentence, dictated by the customer. "Li doesn't have to export that sheet by hand any more" is a win; "completed training" is not. Ask it at the kickoff: thirty days from now, what will have happened for you to feel this was money well spent? Write down their exact words — that is the acceptance test.
2. Configure only what that one win needs. Full-feature training is the single biggest killer of adoption: the volume of information puts people off, and what they actually need may be two buttons. The rest can wait two months.
3. Book a time, not an intention. Not "shall we start next week?" but "Tuesday at nine, when you run that batch, let's walk through it together" — tie the action to a specific time, place and trigger. There is hard evidence for that one edit; see below.
4. Contact density in the first fortnight matters more than the following ten months combined. Better to feel slightly pushy than to feel finished.
Mechanism: motivation is at its maximum on signing day and decays monotonically. Whether the action gets repeated into routine before that motivation runs out decides whether this customer later uses you out of willpower or out of habit. The willpower ones all churn eventually, because willpower always gets claimed by something else.
Boundary: not every business has an onboarding. One-off transactions — selling a used phone, a single service call — have no such window; the equivalent move is one proactive check-in within 48 hours of delivery. And this only holds if the deal should have been sold in the first place (the qualifying piece) — otherwise the smoothest onboarding in the world just delays the discovery by six months.
1. Your champion changes role or leaves. The earliest and the deadliest: your value lives inside that person's head, and once they go you revert to being an invoice nobody in the building recognises. Which is why you meet a second person from day one.
2. Usage narrows: five people using it become one. Total volume may hold, but it has become one person's private tool — and when that person takes leave, you disappear.
3. Your meetings start getting rescheduled and downgraded (the owner used to come; now an assistant does). That is a public readout of priority, and it beats any survey.
4. They abruptly stop asking for things and stop complaining. Don't read that as "finally stable".
5. New vocabulary appears: compliance, evaluation, benchmarking, central procurement. By the time this one shows up it is largely over; what's left is finishing well.
The silence rule: any account with thirty consecutive days and no substantive human interaction (automated mail doesn't count) gets flagged, and a person books a call. Don't wait for the system to tell you — by the time a dashboard notices, the signal has already become the outcome.
Mechanism: sentiment metrics record what someone is willing to say. Behavioural metrics record where they actually put their attention — and attention is both more honest and earlier. People stop using first, become dissatisfied second, and cancel last.
Boundary: a signal exists to trigger one conversation, not to auto-demote an account. See a signal, make a call — don't see a signal and blast out a retention email, which only reveals that you think of them when the numbers dip.
Line one: the problem you were brought in for — in their words from the kickoff, not paraphrased.
Line two: the baseline number against today's number. Conservative and precise: "3.5 hours a week" beats "a substantial efficiency gain" — a conservative exact figure is more credible than a big one, as the value-selling piece works through.
Line three: the single most critical moment of the period — that incident, that deadline. This is the peak, and it must be named out loud, because it is what they remember.
Line four: the one thing to get done together next. Point the document forward rather than making it an invoice for goodwill.
Opening line: "Over these six months we took X from A to B together. What I want to push next half is C — does that match how you see it?"
Mechanism: memory is not a recording, it is a weighted sum of a few fragments. Evaluating a long relationship, people don't integrate — they take the most intense moment and the last moment. Your steady day-to-day delivery barely enters that calculation, which is why "our service has been consistently reliable" says nothing at all in a renewal meeting.
Boundary: a value review presupposes there is value. When the numbers won't stand up, the correct move is to say so and then talk about the fix — not to wrap three vague positives into a slide. They have their own ledger, and one page of polish will burn your entire credibility in a single sitting. Also: "engineer a peak" never means letting things break so you can rescue them. That is manipulation, and customers can see it in their own review.
(Say it on the same call where they confirmed the win. Don't book a separate meeting — booking one turns it into a sales meeting.)
1. Pick up their own sentence: "The reason this went so fast is that the data was cleaned up in advance, right?"
2. Name the next bottleneck in the same machine: "Then by the same logic, what jams next quarter is the approval step — it takes three days today."
3. Offer one small, reversible step: "One of our customers handles it like this — let me show you for ten minutes, and if it isn't worth it, forget I mentioned it."
The line you never cross: never sell the second thing before they have got value out of the first. That isn't expansion, it is raising the stake on an unverified assumption; they will notice eventually, and they will return both. The test is simple — can they say, unprompted, what the first purchase gave them? If not, go back and do the previous move.
Mechanism: the first purchase faces unknown risk, the second faces known risk, and people price those two completely differently. Layer status-quo bias on top — carrying on needs no justification, switching does — and you are holding the position by default. But that discount only applies while they still think you're useful; once you aren't, the same inertia turns into a single clean write-off at year end.
Boundary: not every customer should be expanded. Some are already at their ceiling, and adding more only worsens their return and makes the next renewal harder to argue. For those, the right move is to ask for a referral, not an upsell — convert their satisfaction into new customers rather than into more of their budget.
Ninety minutes to convert "maintaining the relationship" into four things you can actually check.