Operated function · renewals
The champion who bought it moved teams in March. Usage in two of the four departments drifted to nothing over the summer. Nobody logged a complaint, nobody escalated, and the account looked healthy right up until a procurement email in October asked what exactly this was for.
The person who championed the purchase leaves, changes role, or gets a new manager with different priorities. Nobody on your side notices for a quarter, because the account is not in trouble by any measure you watch — the invoice is paid, the seats are provisioned, the support queue is quiet.
Usage narrows. What was bought for four teams is used properly by one. That team loves it and says so, which is why the health score looks fine: it is measuring the enthusiasm of the survivors rather than the width of the deployment.
Quiet reads as healthy. An account raising tickets is an account still trying; an account that has gone silent may be an account that stopped trying, and most health models score those the same way or score the silent one better.
Then the calendar surfaces the renewal at sixty or thirty days. Now there is a scramble: reconstruct the value story, find out who the new decision-maker is, book a meeting into a quarter that is already full, and negotiate from a position where the other side has all the leverage and a budget cycle to enforce it.
And the reason for a loss, when it happens, is recorded as whatever was said on the last call. Price, usually — because price is the polite reason, and the real reason was decided in March.
A renewal outcome is largely determined by changes — in sponsor, usage width, and engagement — that occur months before the renewal date, and the renewal date is the only thing most operations monitor.
This is why renewal effort has such poor leverage. Work applied at sixty days is applied after the decision, into a conversation where your only remaining instrument is price. Work applied when the sponsor changed is applied while the outcome is still open, and it usually costs an introduction rather than a discount.
The signals are not exotic and they are mostly already in your systems: who logs in and from which parts of the organisation, whether the sponsor’s email still resolves, whether the department that drove adoption still appears in the data, how long since anyone from your side had a substantive conversation with someone who is not the day-to-day user.
What is missing is that nothing watches them continuously and nothing raises them while they are still cheap to answer. Each one individually looks like noise; together, and with elapsed time attached, they are the earliest honest read on whether the account renews.
And the discipline that matters most is recording the loss reason at the moment of the signal rather than at the moment of the loss. A reason captured in March is evidence. A reason captured in October is a summary of what the customer was willing to say.
How early an at-risk account becomes visible — measured by days between the first recorded risk signal and the renewal date, against a baseline where the first signal was the renewal date itself.
Sponsor changes caught while the outcome is still open — measured by sponsor departures detected within a cycle, versus discovered during the renewal conversation.
Deployment width, as distinct from user enthusiasm — measured by active departments or teams against the number the contract was scoped for.
Renewals decided on evidence rather than on discount — measured by renewals closed without a price concession, as a share of all renewals, against your own prior period.
Loss reasons that are causal rather than polite — measured by losses attributable to a signal recorded before the renewal window, versus reasons captured at the final call.
Expansion identified separately from risk — measured by expansion conversations initiated outside a renewal window, which is where they are not crowded out.
that an account with a genuine product or budget problem renews, and no commercial decision. Nothing here prices a renewal, offers a concession, negotiates, or decides to let an account go. Those are yours. What changes is when you find out and what you know when you walk in.
Usage and identity from the product itself, contract and date state from your CRM or billing system, and conversation history from the channels your team already uses. Nothing migrates and no second account record is created.
The sponsor signal usually comes from the plainest sources available: an email that starts bouncing, a login that stops, a title that changes in a directory you already sync. It does not require a data-enrichment purchase, and where a signal genuinely cannot be read, that is reported as a blind spot rather than estimated.
Access follows your identity provider, so an account team member who leaves loses access here when your directory says so.
What is read is deployment shape — which parts of a customer organisation are using what was bought, and whether the sponsor relationship still exists. It is not individual behavioural monitoring, and it should not become a report about how many hours a named person spent in a product.
That distinction matters commercially as well as ethically: a customer who discovers their vendor is profiling individual employees has learned something about you that no renewal signal is worth.
Every signal raised carries its basis — what changed, when, and read from where — so an account owner can judge it rather than receive a score they cannot interrogate.
Operational access is not permission to train. Customer usage and relationship data does not become material improving anything serving another organisation, including one that competes with your customer.
The privacy question here is about your customers’ employees, and the answer that satisfies it is the scope: deployment shape rather than individual monitoring, stated before anything is connected and narrow enough to describe in a sentence.
Your CS leadership owns the thresholds. What counts as a risk signal, how loud it should be, and what an owner is expected to do about it are their decisions — a vendor-set threshold produces alerts nobody trusts and everybody mutes.
Where an obligation attaches through your customer contracts or a jurisdiction, it is marked applicability-gated rather than presented as standing.
A single cohort of accounts that already renewed or churned in the last year — observed read-only, with no live account touched — to establish which signals actually preceded which outcomes in your book.
Start in the past. Running the signal set against accounts whose outcome is already known tells you which signals have predictive value for your business rather than for a vendor’s case study, and it involves no live customer at all.
That backward pass is the honest test. If sponsor departure and width contraction turn out not to separate your renewals from your losses, the model is wrong for your book and you should not deploy it forward — and you will have learned that from your own history in a couple of weeks.
Only where the backward pass separates outcomes does forward watching begin, on one segment, with thresholds your CS leadership set and every signal going to a human rather than into an automation.
Most health scores blend usage and support signals into a single number, and the specific failure this addresses is that a narrowing deployment with one enthusiastic remaining team scores well. Run the backward pass against accounts you already lost: if your existing score separated them from the ones that renewed, it is working and this is a duplicate. If the lost accounts looked healthy until the last quarter, that is the gap.
It would be if it reported on individuals, and it does not. What is read is deployment shape — which teams still appear, whether the sponsor relationship still exists — not how many hours a named person spent in a screen. That boundary is in the scope before anything is connected, and it is as much a commercial protection as an ethical one: a customer who finds out their vendor profiles their employees has learned something no renewal signal is worth.
Auto-renewal delays the conversation rather than removing it, and it makes the eventual non-renewal arrive with no warning at all — because the account had been renewing itself while quietly contracting. The backward pass is particularly worth running on an auto-renew book, since the accounts that eventually lapse usually show the same width contraction, just over a longer period.
The experienced ones usually do, and that is the problem: it lives with the people who have been there longest, it does not survive their departure, and it does not cover the accounts nobody has time to think about. The backward pass tests it directly — if your team’s judgement already separates outcomes as well as the signals do, then the honest conclusion is that your constraint is coverage rather than insight, and the scope should change accordingly.