When High Adoption Hides Churn Risk: A GUARD Health Check, Save Plan, Expansion Case and QBR on One Book of Accounts

In my view, most customer success dashboards lead with usage because it arrives every day without anyone asking, and it hides the most. In AI ticket resolution, high usage says the agent answered a lot of tickets, not that the answers were right, that anyone on the customer side still owns the result, or that procurement has not started a vendor review.

In my experience, a renewal lost late follows a path: the champion leaves, nobody replaces them, one ticket type keeps getting wrong answers, and the first warning is procurement asking for an outcomes report. The usage chart is still green.

A weighted health score fixes the reading if every account is scored the same way and gaps are flagged, not filled. The GUARD skills also refuse the wrong play: no expansion case below 60, and no discount for a problem a discount cannot fix.

The four skills run inside Claude once the plugin is installed, free, from GitHub. Each takes whatever you paste, scores what it can and marks the rest as missing. Two parts of the order are the skills' own rules; the rest is my sequencing.

The first rule: the health check comes first. The save plan skill says to run it if health context is missing, the expansion skill gates at 60 and the QBR skill says never to prep blind. The second: "Never pitch expansion to a renewal-at-risk account." Both save plans first, and the CRAFT GTM cross-check in Step 6, are my choices.

Step 6 uses a second connector on purpose: the CRAFT GTM retention playbook scores health with illustrative weights, which shows what a simpler score would miss.

Step 1. Score the whole book with Account health check

Input: one paste covering six accounts: ARR, renewal timing, usage against contract, CSAT on AI-resolved tickets, NPS where it exists, the people and what was said on calls. I left in Example Home Services Co, which had almost no data.

What came back: a ranked table, with $192,000 of the $522,000 book at risk (36.8%). Adoption, engagement and value score out of 20 each, sentiment and commercial health out of 15, and expansion readiness out of 10. Critical is 0 to 39, At risk 40 to 59.

How to read it: Example Travel Booking Co scored 18 of 20 on adoption and 41 overall. A champion who left in the last 90 days with no replacement costs 4 engagement points, taking it from 9 to 5, and a procurement vendor review sits in the lowest commercial band, so on-time payment still scored 3 of 15. The skill also named the conflict: high usage means the agent answers a lot of tickets, not that the answers are right.

A caution of my own: a score near a band edge is a judgement. 41 is one point above the At risk floor of 40. Two readings were boundary calls: engagement started at 9, though a skipped QBR also fits the 5 to 8 band, and sentiment got 4, though vendor review language fits the 0 to 3 band. That stricter reading gives about 33 to 39, so I plan the week as if the account were Critical.

Each diagnosis also carries an alternative. Here: a change in the customer's own cancellation policy never reached the agent's knowledge source, tested by tagging 50 low-CSAT tickets within 48 hours.

Example Home Services Co, with 3 of 6 signals scorable, got a partial assessment and no classification, which is not the same as safe.

Hand-off: Example Travel Booking Co goes to the save plan, Example Grocery App Co (86, Healthy, over its limit) to the expansion skill, and Example Telecom Co, the largest account, to QBR prep.

Step 2. Build the save plan with Renewal save plan

Input: the Example Travel Booking Co entry, its health result, a renewal date of 14 January 2027, and two facts the skill needs: no renewal conversation has happened, and procurement has not named another vendor.

What came back: 35%, At risk, with 80 days left and a countdown position of Day 100 of 180. The Day 180, 120 and 90 steps were all overdue, the last by ten days. It ranked three risk types: champion departure and value perception, both High, and possible competitive displacement, Medium, because a vendor review is open with no vendor named. The save plan runs the first two and says not to offer a discount, because a discount fixes neither.

How to read it: the playbook rates ten factors as high or low, and the run said it "gives no points per factor", so 35% is a judgement on the tally alongside. Read the tally and the overdue list, not the percentage. Read the countdown carefully too: "Day 100 of 180" counts days gone, while the steps at Day 180, 120 and 90 count days left, so with 80 days left the account is past all three. The playbook allows 30 days to build a relationship with a new stakeholder after a champion leaves; five weeks have passed, so week one is about finding that person.

Run the alternative test "within 48 hours", as the skill says: ask procurement whether the review covers every vendor. A routine review turns the save into a standard value recap.

Hand-off: Example Fintech App Co goes through the same skill.

Step 3. Run the same skill on the second At risk account

Input: the Example Fintech App Co entry: a score of 51, about 150 days to renewal, 38% of the contract in use, and a refund workflow waiting five months on the customer's own compliance review.

What came back: 50%, Uncertain, the band where the playbook asks for a formal save plan with CS leadership involved. At Day 30 of 180, the Day 180 baseline is about 30 days overdue and the Day 120 step about 30 days away. The tally was 3 high, 4 low and 3 unknown.

How to read it: same skill, different save. The weekly meeting with the Head of CX rules out champion risk; paying for the full contract, using 38% and measuring no outcome is what makes it value perception. The plan is a value audit of order-status answers, one outcome measure from now, a written answer for the compliance reviewer on refund approvals and the audit trail, and a 90-day plan to put refunds live on a limited set of refund types.

If compliance has already said no, the alternative is to prove value on order-status alone and discuss right-sizing, "framed as a positive step with a clear path back up". One line to remove before you use the opener: it speaks of "the refund workflow you bought this for", which my input never said, and the skill's own rule is "Only what you were told".

Step 4. Size the expansion with Upsell and expansion

Input: the Example Grocery App Co entry and its health result of 86: 23,600 resolutions a month against 20,000 contracted, a finance question about the last overage invoice, interest in the chat channel, and budget planning from the second week of December.

What came back: Warm, 3 of 4, health gate passed. Usage counted (open the talk above 70% of entitlement, urgent above 90%), and so did value: first response time from 5 hours to 4 minutes on the customer's own dashboard, shown to the COO. Timing counted under its 4 to 6 week rule, a slightly early read at 6 to 7 weeks. The Spanish-language queue did not: mentioned, not approved.

How to read it: the case is about predictability. At $0.40 a resolution, the volume they use comes to about $113,280 a year, about $17,280 above the contract and mostly paid as overage already. The skill labels this "for internal scoping", and the account executive owns final pricing.

The chat channel, a new module, goes to the AE with a four-part handoff brief.

Step 5. Prepare the review with QBR prep

Input: the Example Telecom Co entry, its score of 68, and the quarter's notes, including two success criteria agreed at signing, one escalation after a tariff update, and the Director of CX saying on the last call that they are "looking at what our help desk vendor's own AI add-on can do".

What came back: a 45-minute QBR brief in seven sections. Against the first criterion, 35% of chat tickets resolved without a human by month 6, it reported 41%. The second, outage-day backlog under 4 hours, was "not measured yet", so it proposed measuring it at the next outage.

How to read it: the risk section is where this QBR is won. The skill put the comparison on the agenda openly and offered to score both products on the same sample of the customer's own chat tickets, without commenting on the other product. It also gave the tariff change process an owner on each side.

Timing matters too. The account renews in about 4 months, roughly Day 60 of 180 with 120 days left, so the playbook's Day 120 step, confirming the sponsor and starting the business case, is due now. The skill said to hold the QBR within 3 weeks so the formal renewal conversation can follow at 90 days left.

Where it had no data it left a marker, such as "[CSM to add: one outage or peak day with before and after figures ...]". Fill each one or cut it.

Step 6. Cross-check the score with the CRAFT GTM retention playbook

Input to the retention playbook: Answerloop's segment, a usage-based model, an example churn rate of 8% a year, a small CS team, the six signals the team can track, and the three churn reasons this book shows, separated by semicolons. With commas, my first run merged them into one.

What came back: a play and a draft email for each reason, and a health score with weights it calls illustrations: usage volume 40%, usage trend 25%, feature breadth 15%, billing health 10%, support tickets 10%. Its one outreach touchpoint meant to save an account fires on a fall in volume.

How to read it: 65% of the weight sits on volume and trend, and CSAT feeds only the 10% support signal. The tool does not see the data or say how to scale each signal, so I scored Example Travel Booking Co myself, with the readings in the table alongside: 69 to 71, the monitor band or just over the healthy line, against 41 on the GUARD score. It falls to about 49 to 51 only if the two neutral signals are zero, and the fall alert never fires because volume never fell. The tool points to the fix: it listed five of my signals, including reopen rate and champion changes, as unmatched.

One draft to replace: the email for the competitor reason is a support apology ("Making things right"). Write that one from the QBR's risk section.

1. Paste your whole book into Account health check, with a quality measure next to usage, and run the 48-hour test for each At risk account.
2. Run Renewal save plan on every At risk account, and do its overdue steps first.
3. Run Upsell and expansion only on accounts that pass the health gate, and keep its figures internal until your AE checks them.
4. Add a quality signal and a champion change signal to any usage-weighted health score you run.

1. Reading usage as health. Pair it with a quality measure such as CSAT on AI-resolved tickets.
2. Waiting to meet the new owner after a champion leaves. The playbook allows 30 days at most.
3. Offering a discount to save an account whose problem is a departed champion or wrong answers.
4. Pitching expansion to an account at its limit before checking its health.
5. Leaving the risk section of a QBR thin. The customer already knows what went wrong.

• How I Check Product-Market Fit, Churn and ICP Gaps Before Scaling Paid Demand Generation
• One B2B Deal From MEDDPICC Discovery to a CFO-Ready ROI Case and a Dated Close Plan
• How I Stress-Test a B2B GTM Plan Before the Board Sees It, With a Pre-Mortem and a Bottom-Up Count
• All 15 guides