Score your book so the color drives the work.
The inputs, the weights, the green/yellow/red thresholds, and one specific action per tier.
Inside: The input set, the weights, the thresholds, and the action per tier.
Install it in one line, or paste it in.
~/.claude/skills/ and runs automatically when it is relevant.Connect your context. Set it to your motion.
it reads the book, ARR, renewal dates, and open cases automatically.
real adoption becomes the strongest input instead of a guess.
turns ticket volume and sentiment into a real risk input.
This was built for a B2B SaaS org scoring a renewal book. Set these to your stack:
Pick inputs you can actually see. A model with a 40% weight on a signal you never measure is a model that lies confidently.
| Set this | What it is | Default / Example |
|---|---|---|
| ADOPTION input | usage signal and weight | product analyticsweight 40% |
| RELATIONSHIP input | sponsor / champion status and weight | multi-thread countweight 20% |
| SUPPORT input | ticket load and sentiment and weight | ticketing toolweight 15% |
| ENGAGEMENT input | last touch, QBR cadence and weight | activity datesweight 15% |
| COMMERCIAL input | renewal proximity, growth and weight | ARR trendweight 10% |
| THRESHOLDS | green / yellow / red cutoffs | 70+ green40-69 yellow<40 red |
| TIER ACTIONS | the one move per tier | green nurtureyellow planred escalate |
Everything the skill does, in full.
Builds a health-score model for a whole book of business: the inputs that actually predict retention, the weight each one carries, the thresholds that split green from yellow from red, and the one specific action attached to each tier. The output is a scored book where every color tells the CSM what to do next, not just how to feel.
- 1Pick the inputs
Choose four to six signals that predict retention for your product: adoption, relationship depth, support health, engagement cadence, commercial trend. Fewer strong inputs beat many weak ones. Any input you cannot measure gets dropped or flagged, never faked.
- 2Set the weights
Weight by predictive power, not by what is easy to pull. For most B2B SaaS, adoption carries the most and commercial the least, because usage moves first and the renewal number moves last. State the weights out loud so anyone can challenge them.
- 3Draw the thresholds
Green, yellow, red are decisions about attention, not math for its own sake. Set cutoffs so green means "leave it running," yellow means "there is a plan," red means "escalate this week." Re-tune once you see the spread; if 80% of the book is red, the thresholds are wrong, not the book.
- 4Action per tier
Each tier gets exactly one default action. Green: nurture and look for expansion. Yellow: a written recovery plan with an owner. Red: escalate, involve a leader, set a save play in motion. A score with no action is a dashboard nobody uses.
- 5Explain every score
Every account shows the two inputs pulling it up and the two pulling it down. A CSM should never ask "why is this yellow." The model answers before they ask.
- Every input is measurable, or it is dropped from the model and noted.
- Weights are stated and sum to 100. No hidden factors.
- No account gets a color without the top reasons for it shown.
- Thresholds are sanity-checked against the book's spread, not set once and forgotten.
BOOK HEALTH · 5 sample accounts Account Score Tier Up Down Account A 82 GREEN adoption, sponsor none material Account B 61 YELLOW adoption champion quiet Account C 58 YELLOW engagement support spike Account D 34 RED none material no usage, no sponsor Account E 73 GREEN adoption, growth one open case Tier actions: GREEN (A, E): nurture, scout expansion. YELLOW (B, C): written recovery plan, owner named, this week. RED (D): escalate to leader, open a save play now.
The scores, weights, and thresholds come from the inputs you paste or the tools you connect. Nothing above is a real customer number; it is illustrative. The default weights (adoption 40, relationship 20, support 15, engagement 15, commercial 10) are a starting point tuned for one SaaS book, not a law. When an input is missing, the model redistributes weight and tells you which signal would sharpen the score.
Where an operator takes this next.
The model works from a pasted book. Here is where the score stops being a snapshot and starts being a live signal.
A score nobody acts on is a dashboard. A score that triggers the next move is a system.
A scheduled Claude task recomputes every score against live Amplitude usage and Salesforce data overnight, so a color never lags reality by more than a day.
Have it DM the CSM in Slack just the accounts that changed tier since yesterday, not the whole book, so the signal doesn't get lost in the noise.
Feed every RED tier into an escalation or save-play skill automatically, so the score doesn't just sit there, it triggers the next move.
One skill is the on-ramp.
A single skill does one job. Chained into a playbook, or run as a full build, it becomes a system. Here is where this one plugs in.