RescueChurn triageHuman review required

Churn Signal Triage

Sort a week of at-risk accounts into who needs a call today versus who needs an email, with the reason.

Triage fails when it follows the loudest signal instead of the most reversible loss. The account quietly dropping usage with no complaint is usually more dangerous than the one filing an angry ticket.

The template

Copy it, swap the {{marked}} variables for your account, and run it.

prompt.txt

I have several at-risk accounts and limited time this week. Triage them into
three buckets: CALL TODAY, EMAIL THIS WEEK, MONITOR. For each account, give one
sentence on why it landed in that bucket and the single most useful first move.

Accounts and their signals:
{{account_list}}

My constraint this week: {{constraint}}
What I can realistically do: {{capacity}}

Rules:
- Prioritize by reversibility and value at risk, not by who is loudest.
- "Loud but low-value" is not the same as "quiet but about to churn." Flag the quiet ones.
- Do not put more accounts in CALL TODAY than my capacity allows.
- Be explicit when a signal is ambiguous rather than guessing the cause.

Variables

{{account_list}}: Accounts and signals (one per line)e.g. - Atlas Logistics ($120k): champion went quiet 3 weeks, renewal in 45 days - Brightline ($18k): angry ticket about a bug, otherwise heavy usage - Corian Foods ($90k): usage down 40%, no complaints, renewal in 6 months
{{constraint}}: Your constraint this weeke.g. I have room for two real save calls, not five
{{capacity}}: What you can realistically doe.g. two calls, plus emails, plus one exec escalation if needed

Worked example

It is Monday and a CSM has three accounts flashing red and time for maybe two save calls. One is loud (an angry ticket), one is high-value and quiet, and one is quietly bleeding usage with no complaint. She needs to spend her two calls on the right two.

filled-prompt.txt

Accounts and their signals:
- Atlas Logistics ($120k): champion went quiet 3 weeks, renewal in 45 days
- Brightline ($18k): angry ticket about a bug, otherwise heavy usage
- Corian Foods ($90k): usage down 40%, no complaints, renewal in 6 months

My constraint this week: I have room for two real save calls, not five
What I can realistically do: two calls, plus emails, plus one exec escalation if needed

Good output vs bad output

GOODSpends the two calls on value-at-risk and reversibility, explicitly flags the quiet 40%-drop account as the real danger, and calls out the loud-but-safe account so it does not steal attention.

CALL TODAY (your two calls):
- Atlas Logistics ($120k): Highest value, most reversible, and a 45-day
  renewal clock. A silent champion this close to renewal is the classic
  quiet-churn setup. First move: a direct call to the champion, and if no
  response in 48h, escalate above her.
- Corian Foods ($90k): This is the dangerous one, a 40% usage drop with zero
  complaints means they may have already checked out mentally. No complaint is
  not good news here, it is the absence of a fight. First move: a call to find
  out what changed before it hardens.

EMAIL THIS WEEK:
- Brightline ($18k): Loud but low-value and heavy usage suggests they are
  frustrated, not leaving. First move: a specific email with a real fix ETA;
  a call would spend a save slot on an account that is not actually at risk.

Note: Brightline is the loudest and the lowest priority. Do not let the angry
ticket pull your attention off Corian.

BADLooks fine at a glance. Here is why it fails.

Here's how I'd prioritize:

CALL TODAY:
- Brightline: they're angry and sent a ticket, so reach out ASAP to smooth
  things over.
- Atlas Logistics: champion went quiet, follow up.
- Corian Foods: usage is down, keep an eye on it.

All three accounts show risk signals and deserve attention. I'd recommend
reaching out to all of them proactively to demonstrate care and rebuild
engagement across the board.

Wrong registerPuts the loud $18k ticket at the top of CALL TODAY, it triaged by volume, not value or reversibility, exactly the mistake the prompt warned against.
Missing contextTreats Corian's quiet 40% usage drop as "keep an eye on it" when a silent usage collapse is the most dangerous signal in the set.
No next step"Reach out to all of them" ignores the stated capacity of two calls, so the CSM ends the exercise with no actual plan.
Generic filler"Demonstrate care and rebuild engagement across the board" is a platitude that does no triage work at all.

Guardrails

When NOT to use this

Do not use it to replace judgment on a single high-stakes account; triage is for a portfolio, not a decision you should own directly.
Skip if your signals are unreliable or stale; triaging bad data just launders it into false confidence.

Responsible use

The model does not know internal politics, contract clauses, or a save already in motion. Sanity-check every bucket against what you know.
Do not paste customer-identifying detail beyond what you need; account names plus signals are usually enough.

A human must review the output before it reaches the customer.

Model + settings

Model: Claude Sonnet, fast triage over a list is exactly its strength.
Temperature: 0.3 (low, you want consistent, defensible prioritization)
Settings: Always give it your real capacity constraint; without it, the model will tell you to call everyone, which is not triage.
Re-run it midweek as signals change; treat the buckets as a living list, not a Monday decree.

Back to all prompts