Verified against ChatGPT · 2026-08-14
Write up a customer health assessment for a renewal decision that states its uncertainty instead of a false-precise score
Turns raw account signals into a plain-language health writeup for a renewal-risk conversation, distinguishing signals that actually predict churn from ones that just correlate with normal account activity, and stating explicitly how confident the assessment really is.
The prompt
Ready to copy — highlighted parts are example details you can swap.
Write a customer health assessment for a renewal-risk conversation, based on the signals below. This needs to hold up in a real conversation about whether to intervene, not just produce a score that looks precise. ACCOUNT SIGNALS Logins down 40% over the past 6 weeks; primary admin contact left the company 3 weeks ago per their LinkedIn; no support tickets filed in 2 months (previously averaged 3/month); NPS response 8 months ago was a 9. CONTRACT DETAILS Annual contract, $60k, auto-renews unless cancelled 30 days prior. RELATIONSHIP HISTORY Historically a steady, low-touch account with no prior escalations or complaints in 18 months as a customer. RENEWAL DATE Renews in 5 weeks; cancellation notice window closes in 25 days. RISK THRESHOLD DEFINITION Low risk: no major signal changes. Watch: one significant signal change (usage or contact change) without explanation. High risk: two or more unexplained significant changes, or any explicit dissatisfaction signal. Go through the given signals and separate the ones that actually predict renewal risk from ones that merely correlate with normal account variation — a drop in login frequency during a stated seasonal slow period for this customer's business is not the same signal as a drop in login frequency with no such explanation, even though both would look identical in a raw metrics dashboard. State which signals you're treating as meaningful and which you're discounting, and why. Apply the given risk threshold definition explicitly rather than producing a vague qualitative label — state where this account actually falls against that specific threshold, and what would need to be true for it to cross into the next risk tier, so the assessment is checkable against a rule rather than just an impression. State your actual confidence in this assessment, not just the conclusion — if the signals are ambiguous or thin, say plainly that this needs a direct conversation to confirm rather than presenting a guess with the same confidence as a well-evidenced read. An account with genuinely strong warning signals and an account with one ambiguous signal should not read as equally certain just because both produce a "risk" label. OUTPUT FORMAT 1. Signals treated as meaningful, and why. 2. Signals discounted, and why. 3. Risk tier against the stated threshold, and what would move it to the next tier. 4. Confidence level in this read, and what would raise or lower it. 5. The one recommended next action given the renewal timeline.
Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The instruction to separate signals that actually predict risk from ones that merely correlate with normal variation targets a real and common analytical error: raw account metrics like a login-frequency drop look identical whether they're caused by a seasonal business slowdown or genuine disengagement, and a model asked to assess health from the numbers alone, without being told to check for an ordinary explanation first, will treat every negative-looking number as an equally weighted warning sign, inflating risk assessments for accounts that are actually fine. Applying the given risk-threshold definition explicitly, rather than producing a qualitative label like "moderate concern," matters because a vague label can mean different things to different readers and can't be checked or compared across accounts, while a rule like "two or more unexplained significant changes" produces an assessment anyone reading it can verify against the same signals and get the same answer from. The explicit confidence statement is the most important structural element here because GPT-5.1, like most models, tends to present a conclusion in confident, uniform language regardless of how strong the underlying evidence actually is — an assessment built on one ambiguous signal (a stale NPS score from eight months ago) and one built on multiple concrete, unexplained changes (admin departure plus ticket silence plus usage drop) will otherwise read with identical certainty, which is actively misleading for a renewal-risk conversation where the appropriate response (a quiet check-in versus an urgent executive outreach) depends entirely on how sure the signal actually is.
What you get back
Meaningful signals: primary admin departure (unexplained, structurally significant — loses the account's internal champion) and the drop to zero support tickets alongside a 40% login decline (a genuine disengagement pattern, not explained by any known seasonal factor). Discounted: the 8-month-old NPS score, too stale to reflect current sentiment either way. Risk tier: High Risk against the stated threshold — two unexplained significant changes (admin departure, activity drop) with no explanation on file. Confidence: moderate-high: the pattern is concrete, but there's no direct recent conversation confirming intent, so this should trigger outreach to test the read rather than be treated as confirmed churn. Recommended action: a direct check-in call with a newly identified stakeholder within the next week, given the 25-day cancellation notice window.
Verified against
ChatGPT GPT-5.1 · 2026-08-14
Changelog
- 2026-08-14 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
