Field KitCS / Onboarding Prompts
All prompts
AdoptHealth checkHuman review required

Account Health Summary

Compress a noisy account into a health call with the evidence behind it, so the score means something.

Health scores rot when they average vibes into a number nobody trusts. A useful health call reasons from the evidence and weighs it against what success actually means for that specific account.

01

The template

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

prompt.txt
Summarize this account's health. Give a single health call (Healthy /
Watch / At Risk), then the three signals that most support it and the one
counter-signal that argues against it. Do not average vibes into a number;
reason from the evidence.

Account: {{account_name}}
Usage signals: {{usage_signals}}
Relationship signals: {{relationship_signals}}
Commercial signals: {{commercial_signals}}
Support signals: {{support_signals}}
What "healthy" means for THIS account: {{health_definition}}

Rules:
- Weigh signals against what healthy means for this specific account, not a generic rubric.
- Always surface one counter-signal, even for a Healthy call. No account is all green.
- If two signals conflict, say so; do not smooth it over.
- Under 150 words.

Variables

{{account_name}}
Account namee.g. Cedar & Vine (hospitality group)
{{usage_signals}}
Usage signalse.g. logins steady; but only 2 of 9 locations have onboarded staff
{{relationship_signals}}
Relationship signalse.g. champion is engaged and responsive; no exec relationship above her
{{commercial_signals}}
Commercial signalse.g. paid annual up front; expansion conversation started by them
{{support_signals}}
Support signalse.g. low ticket volume; one unanswered feature request from Q1
{{health_definition}}
What healthy means heree.g. multi-location rollout is the whole point; single-location usage is a warning, not a win
02

Worked example

A CSM managing 40 accounts needs to triage her book for the week. Cedar & Vine looks green on the dashboard (steady logins, paid up front, expansion interest), but she knows the deal was sold on a nine-location rollout and only two are live. She wants a health call that reflects the actual bet.

filled-prompt.txt
Account: Cedar & Vine (hospitality group)
Usage signals: logins steady; but only 2 of 9 locations have onboarded staff
Relationship signals: champion is engaged and responsive; no exec relationship above her
Commercial signals: paid annual up front; expansion conversation started by them
Support signals: low ticket volume; one unanswered feature request from Q1
What "healthy" means for THIS account: multi-location rollout is the whole point; single-location usage is a warning, not a win
03

Good output vs bad output

GOODRefuses to let "steady logins + paid up front" mask the rollout stall, weighs signals against this account's actual definition of healthy, and names the commercial-vs-deployment conflict instead of averaging it away.
HEALTH: WATCH.

On paper this looks healthy, paid up front, engaged champion, they started an
expansion talk. But for this account the deal thesis is the nine-location
rollout, and only two are live. Steady logins from two sites is not success
here, it is a stall dressed up as stability.

Three supporting signals: (1) 2 of 9 locations onboarded against a
multi-location thesis, (2) no relationship above the champion, so rollout has
no executive push, (3) a Q1 feature request still unanswered.

Counter-signal: they opened the expansion conversation, which is a real buying
signal, but expanding before the base rollout lands could paper over the gap.

Conflict worth naming: commercial signals say green, deployment says yellow.
Trust the deployment.
BADLooks fine at a glance. Here is why it fails.
HEALTH SCORE: 82/100 (Healthy)

Cedar & Vine is a healthy account. Logins are steady, they paid annually up
front, support tickets are low, and the champion is highly engaged. They've
even started an expansion conversation, which is a great sign. Overall this is
a strong, low-risk account trending in a positive direction. Recommended
action: continue current cadence and explore expansion opportunities.
  • Buried ledeThe one fact that matters, 2 of 9 locations live against a multi-location deal, is completely absent, so a real stall is scored as healthy.
  • Generic fillerA precise-looking "82/100" is fabricated confidence; nothing in the input maps to that number, and it invites the CSM to stop thinking.
  • Missing contextIgnores the account-specific definition of healthy that was handed to it, applying a generic "logins + payment + low tickets = good" rubric.
  • No next step"Continue current cadence" is exactly the wrong action for a rollout that has stalled at 2 of 9 sites.
04

Guardrails

When NOT to use this

  • Do not treat the output as your health score of record. It is a reasoning aid; your CS platform owns the number.
  • Not reliable if you feed it only quantitative signals with no account context. The account-specific definition of healthy is what makes it useful.

Responsible use

  • The model cannot see your product telemetry directly. Garbage or stale signals in means a confident wrong call out.
  • Do not let a green summary substitute for actually talking to the customer, especially when relationship signals are thin.

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

05

Model + settings

Model
Claude Sonnet for routine portfolio triage; Opus when the account is large and the signals conflict.
Temperature
0.3 (low, you want disciplined reasoning, not a creative narrative)
Settings
  • The "what healthy means for THIS account" field is the highest-leverage input. Never leave it blank.
  • Ask it to always surface a counter-signal; it stops the model from writing you a comforting summary.