The Customer Health Scoring Reboot — 60-Min Training
Direct Answer
Section 1 — Open & Frame the Vanity-Score Problem (5 min)
Open cold. Pull up the current health-score dashboard on screen and ask the room one question: "What did anyone do differently last week because of a color on this screen?" Wait for the silence. Lincoln Murphy's *Customer Success* (Wiley, 2016) named this the "watermelon account" problem — green on the outside, red on the inside — and Nick Mehta's Gainsight benchmarks show roughly 60% of B2B SaaS health scores fail to predict churn 90 days out.
The score is not the product. The *action it forces* is the product. Frame the hour: we are going to (1) pick the five inputs that actually predict, (2) decide weighting, (3) write the three playbooks, and (4) walk out with one account each, scored and assigned a color today.
Section 2 — The 5 Inputs That Actually Predict (15 min)
Whiteboard these five. No more, no less. ChurnZero's 2025 benchmark report and Tomasz Tunguz's Redpoint analyses converge on this set as the durable signal:
- Product usage — not logins, but depth of feature adoption and DAU/MAU among the licensed seats. A 40-seat account with 6 DAUs is a fire, regardless of MRR.
- Support volume & sentiment — ticket *velocity change* matters more than count. A quiet account that suddenly opens four P2s in a week is louder than a chatty account at steady state.
- NPS / verbatim sentiment — last survey score *plus* the verbatim. A 9 with "we're evaluating alternatives" beats a 6 with "great team, minor bugs" — every time.
- Executive engagement — has a VP+ replied to email or attended a QBR in the last 90 days? Bain's *Loyalty Effect* (Reichheld) data shows exec-replaced accounts churn at 3-4x the baseline.
- Contract risk — months to renewal, auto-renew status, multi-year vs annual, and whether procurement has been pulled in early.
Drill: in pairs, score one shared sample account on each input 1-5. Force disagreement out loud. The point is calibration, not consensus.
Section 3 — Weighted vs Un-Weighted: Pick Your Model (10 min)
Un-weighted (simple average of the five) is the right starting point if you have under 50 accounts or under 24 months of churn data. It is transparent, every CSM can compute it in their head, and it forces conversation about which input is dragging.
Weighted is correct once you have enough churned-logo history (Tunguz suggests 40+ churn events as the regression floor). Fit a logistic regression with churn as the dependent variable. In most B2B SaaS books, exec engagement and product-usage depth carry 50-60% of the predictive weight combined — but let *your* data tell you, not a blog post.
Re-fit quarterly.
The trap to avoid: weighting by *gut* without data. That is how you end up overweighting NPS (the easiest to measure) and underweighting exec engagement (the hardest), which is exactly backwards.
Section 4 — Kill the Vanity-Score Trap (10 min)
A vanity score has three tells. Read them aloud and have the room raise hands if any apply to your current dashboard:
- No one's calendar changes when the color changes. If a green-to-yellow flip does not auto-create a task, it is decoration.
- The CSM who owns the account cannot name the input that moved. If they say "it just turned yellow," the model is opaque.
- The score lags the save. If accounts only turn red *after* the churn notice arrives, your inputs are trailing, not leading.
Fix all three by enforcing the "so what" test: every score change must trigger a named action, owned by a named person, with a due date. Murphy frames it as "Desired Outcome ÷ Appropriate Experience" — the score exists to flag a gap between the two, then close it. If the gap-close is not on someone's Monday calendar, kill the score and start over.
Section 5 — The Three Playbooks (Verbatim) (15 min)
This is the section the team will quote back to you for the next year. Make them write these down word-for-word.
RED Playbook — "48-Hour Exec-to-Exec"
- Hour 0-4: CSM posts to #at-risk Slack with one-paragraph diagnosis (which input flipped, root cause hypothesis).
- Hour 4-24: CSM manager joins. Pre-mortem call with sales counterpart — what do we know, what do we need to know, who owns the relationship inside the customer?
- Hour 24-48: Exec-to-exec outreach. Our VP/C-suite to theirs. Subject line: "Personal check-in on the partnership." No deck. Listening posture.
- Day 3-7: Joint success replan — written, signed, dated. Specific outcomes, specific dates, specific owners on both sides.
- Day 14: Mandatory check-in. No silent reds. A red account with no touch in 14 days is a churned account that hasn't told you yet.
YELLOW Playbook — "14-Day Structured Business Review"
- Day 0-3: CSM runs root-cause diagnosis on the input that moved. Pull usage data, last 3 support tickets, last QBR notes.
- Day 3-7: Schedule the business review with the customer's champion. Frame as proactive, never alarmist: "We want to make sure year two is bigger than year one."
- Day 7-14: Deliver the review. Mandatory slides: value delivered to date, gaps observed, joint plan for next quarter. End with one specific ask of the customer (a new use case, a new team to onboard, a referral).
- Day 14-30: Re-score. If still yellow, escalate to red playbook. Yellow is a *transient state*, not a parking spot.
GREEN Playbook — "Quarterly Expansion Motion"
- Week 1: CSM and AE co-build the expansion hypothesis — which adjacent team, which adjacent use case, which adjacent product.
- Week 2-4: Champion-led intro to the next buying center. Never cold. The current champion makes the warm hand-off.
- Week 4-8: Run the expansion as a mini-sales-cycle. Discovery, demo, business case, procurement — yes, even with an existing logo.
- Week 8-12: Close, expand, or document the no. Green accounts that get no expansion motion for two consecutive quarters silently drift to yellow. Movement is the moat.
Section 6 — Walk-Out Drill & Close (5 min)
Every CSM picks one account. They score it on all five inputs in 60 seconds. They compute the composite.
They name the color out loud, then commit the *first action* from the matching playbook to the room. The manager logs the commitment and the 14-day check-in date. Close with the line that should be on the wall: **"A health score is not a metric.
It is a verb."**
FAQ
Q: How often should we re-score? A: Inputs refresh weekly (usage, tickets), composite recomputes weekly, full model re-fit quarterly. Gainsight's benchmarks show weekly cadence catches roughly 70% of churns earlier than monthly cadence does.
Q: What if our data is incomplete — no NPS, no exec-engagement tracking? A: Start un-weighted with the 2-3 inputs you have. A partial score with action beats a complete score that sits on a dashboard. Add inputs quarterly.
Q: Should the customer ever see their health score? A: No raw color, yes the underlying conversation. Murphy is firm here — sharing the color invites gaming. Sharing the *gap between desired outcome and current experience* invites partnership.
Q: How do we handle a "watermelon" — green on paper, churning in reality? A: That is a *model failure*, not an account failure. Add the missed signal (usually exec engagement or a specific feature-adoption depth metric) to the input set immediately and re-fit.
Q: Who owns the score in our org — CS, Sales, or RevOps? A: RevOps owns the *model*. CS owns the *score per account*. Sales owns the *expansion motion on greens*. Three owners, one number.
Sources
- Murphy, Lincoln. *Customer Success: How Innovative Companies Are Reducing Churn and Growing Recurring Revenue.* Wiley, 2016.
- Mehta, Nick et al. *Customer Success* (Gainsight). Wiley, 2016 — and Gainsight Pulse benchmark reports 2023-2025.
- Reichheld, Frederick F. *The Loyalty Effect.* Bain & Company / Harvard Business Review Press.
- Tunguz, Tomasz. Redpoint Ventures blog — SaaS retention and churn-prediction analyses (2020-2025).
- ChurnZero. *2025 Customer Success Leadership Study & Health Score Benchmarks.*
- Gartner. *Magic Quadrant for Customer Success Management Platforms* (2024-2025).
- Bain & Company. *Net Promoter System* research library.
- OpenView Partners. *SaaS Benchmarks Report* (annual) — retention and expansion cohort data.