A Claude Skill that takes an at-risk renewal account and produces a structured draft playbook: a renewal-probability band, stakeholder-by-stakeholder motions tied to a 30/60/90-day pacing plan, talk-tracks for the most likely objections, and an escalation gate. The CSM reviews the draft, edits it, and converts each motion to a Gainsight CTA. The Skill replaces the blank-page problem at T-90, not the CSM’s judgment.
When to use
Use this Skill at T-90 to T-120 days from renewal on any account a CSM has flagged as at-risk. Earlier than T-120 the signal is too noisy; later than T-90 there is not enough runway to act on what the playbook recommends. The Skill is built for B2B SaaS accounts above 50k ARR where the cost of a custom playbook is justified by the expected save value.
The reader should expect to spend roughly 15 minutes per account: a few seconds to invoke the Skill, ten to twelve minutes reviewing and editing the draft, two to three minutes converting motions into Gainsight CTAs. That is the “time saved” baseline — replacing the 60 to 90 minutes a CSM otherwise spends pulling Gainsight reports, scrubbing Gong call notes, and assembling a renewal plan from scratch.
When NOT to use
This Skill is not the right tool for several account shapes that look superficially similar:
- Auto-renewal accounts where the contract auto-extends without a CSM-led motion. The playbook assumes the renewal is a moment of choice. Auto-renewal accounts need an entirely different motion (de-risk the auto-renew trigger, monitor health, intervene only if the customer initiates an opt-out conversation). The Skill aborts and flags this if the contract has an active auto-renewal clause and the opt-out window has not opened.
- Anything customer-facing without CSM review. The Skill produces internal strategy and scaffolding for preparation. It does not produce customer-ready copy. Talk-tracks are structured objection-and-response frames the CSM rewrites in their own voice — not scripts to paste into an email or read aloud on a call.
- Contractually-binding emails, commercial proposals, or anything that touches commercial terms. Pricing, discounting, term length, and legal language stay with the CSM and Deal Desk. The Skill is forbidden from recommending specific discount amounts.
- SMB accounts under 25k ARR. The cost of generating, reviewing, and customizing a bespoke playbook exceeds the expected save value. Use a templated low-touch motion instead — the Skill returns a recommendation to that effect rather than a full playbook.
Setup
- Drop the Skill into your Claude config. The bundle is at
apps/web/public/artifacts/cs-renewal-playbook-skill/. CopySKILL.mdto~/.claude/skills/cs-renewal-playbook/SKILL.mdand thereferences/directory alongside it. - Customize the reference files. The bundle ships three template references the Skill reads on every invocation. Edit each to match your team:
references/1-segment-playbook-config.md— your ARR thresholds, per-segment section inclusion rules, and the “do not run this Skill” floor.references/2-stakeholder-motion-matrix.md— your canonical stakeholder roles, the motion per (role times archetype) cell, and the stakeholder-gap rules that cap the renewal-probability band.references/3-output-format.md— the literal Markdown shape the Skill emits. Edit only if you are extending the format itself; for wording inside motions, edit the other two references.
- Wire credentials. Set
GAINSIGHT_TOKENandGONG_API_KEYin your shell. The Skill uses Gainsight for health-score history, success-plan progress, and the stakeholder map; Gong for the last 180 days of calls. - Run on a flagged account. Invoke as
cs-renewal-playbook(account_id="...", renewal_date="2026-08-15", arr=180000, segment="mid-market"). Optional inputs:risk_archetype_hint,competitor,executive_sponsor_email. The Skill returns a Markdown draft. - Convert to Gainsight CTAs. The playbook is worthless if the motions stay in a document. Each row in the stakeholder-motion table is one CTA — owner, due week, success signal. Tracked weekly in your renewal pipeline review.
What the skill actually does
The Skill runs eight steps in order; the order matters because each step constrains the next. Pulling contract terms first means every downstream recommendation has to fit inside commercial reality. Diagnosing the risk archetype next means the segment-aware template has something concrete to tailor. Selecting the segment template before generating stakeholder motions means the motion matrix is filtered to the right shape before any output is assembled.
The risk diagnosis runs a Claude classification pass against five canonical archetypes: champion-lost, low-adoption, pricing-pushback, competitive, and value-gap. Each archetype has a concrete signal definition — low-adoption means weekly active users below the segment p25 for two consecutive months, not a vibes-based read of the dashboard. The Skill returns up to two archetypes ranked by confidence; if the CSM passed a hint that disagrees, the Skill surfaces the disagreement rather than overriding either side.
Renewal probability is bucketed into four bands (over 70 percent, 40 to 70, 15 to 40, under 15) rather than a point estimate. Point estimates here are spurious precision — nobody can tell the difference between a 32 percent and a 38 percent renewal in practice, and presenting a number invites overconfidence. The bands map to recognizably different motions: a 40-70% band runs a standard save play; a 15-40% band assumes executive escalation; an under 15% band shifts the playbook from save to de-risk-the-loss (data export, contract-termination terms, prevent-bad-references motion).
The segment-aware template is the difference between this Skill and a generic prompt. The same archetype demands a different motion at different deal sizes. A “low adoption” save on a 500k enterprise account warrants a CSM-and-CSM-leader joint executive business review and a 90-day pacing plan. The same archetype on an 80k mid-market account warrants a 30-minute working session with the end-user team and a 60-day pacing plan. The segment-playbook config in references/1-segment-playbook-config.md encodes these differences as section-inclusion rules.
The stakeholder-motion matrix in references/2-stakeholder-motion-matrix.md is the single source of truth for “which stakeholder do we engage, and how, given which risk pattern.” The Skill looks up each stakeholder in the account’s Gainsight map and assigns the recommended motion. If a required role is missing, the Skill caps the renewal-probability band — no economic buyer means the band cannot exceed 40-70% until one is identified, no champion means the band cannot exceed 15-40%. This forces the output to surface the gap rather than paper over it.
Talk-tracks are structured as objection (verbatim, in the customer’s likely phrasing), underlying concern, response posture (acknowledge then reframe then ask), and bridge question. They are explicitly labeled as “scaffolding for preparation, not a script” in the output. The Skill does not produce conversational copy at all — that is the CSM’s job.
The escalation gate specifies the explicit signal that triggers escalation to the CSM leader, and from CSM leader to VP CS. “If no new champion identified by end of W3” is a gate. “If renewal-probability band drops to under 15 percent after W4 motions complete” is a gate. Without explicit gates, CSMs grind through doomed save motions past the point of recoverability — the gate is what makes “stop, escalate, or write off” a decision the CSM has pre-committed to.
Cost reality
Token cost per playbook is in the order of 30k to 60k tokens (input plus output) on Claude — call it 10 to 25 cents at current Sonnet pricing, or under 50 cents at Opus. The input volume varies with how much Gong call transcript and Gainsight history is loaded; the output is consistently around 1500 to 2500 words.
Time saved per playbook is the load-bearing cost number. A CSM running a manual renewal prep on a flagged account spends 60 to 90 minutes pulling reports, reviewing call notes, and assembling a draft plan. The Skill compresses that to roughly 15 minutes of CSM attention: invoke, review, edit, convert to CTAs. On a CS team with 30 at-risk renewals per quarter, that is roughly 25 to 35 hours of CSM time reclaimed per quarter, or roughly 10 percent of one CSM FTE if you assume a 4-renewal-per-week steady state.
Net: the Skill pays for itself if it saves a single 100k renewal that would otherwise have been worked at a lower quality. The token cost is rounding error against the contract value at risk.
Success metric
Track three things over the first quarter of use:
- Time-to-first-draft per renewal. Should drop from 60 to 90 minutes to under 20 minutes (Skill runtime plus CSM review). If it does not, the bottleneck is either stale Gainsight data or under-customized reference files.
- CSM CTA conversion rate. Of the motions in each generated playbook, what percentage become tracked Gainsight CTAs within 48 hours of the playbook being generated. Target above 80 percent. Lower than that means the motions are not actionable enough — usually a sign the stakeholder-motion matrix is too generic.
- Renewal rate on at-risk accounts where the Skill was used vs not. This is the delayed signal that matters most. Compare quarter-over-quarter. The Skill is not the only variable, but if at-risk renewal rate does not move at all, the playbooks are being generated and ignored.
vs alternatives
- Gainsight Renewal Playbooks (the in-product feature). Gainsight ships templated renewal playbooks tied to CTA workflows. They are operationally tighter than this Skill — every motion is a tracked task in the same system as the rest of CS. They are also generic: the templates are not tailored per-account using call evidence, and the archetype diagnosis is whatever the CSM types in. The right combination is to use this Skill to generate the draft and Gainsight Playbooks to track execution. The Skill is the brain; Gainsight is the system of record.
- Catalyst (or other CS platforms). Catalyst’s renewal forecasting produces a single renewal-probability number per account from a model, no plan attached. Useful for prioritization, useless for “what do I do about it.” The Skill is complementary — Catalyst tells you which accounts to flag, the Skill tells you what to do per flagged account.
- Manual Quartet planning (a CSM and their leader spending an hour per account in a whiteboard session). Highest quality output, completely unscalable. Reserve for the top three accounts of the quarter; let the Skill handle the long tail. The Skill is not better than two experienced CS humans thinking carefully — it is better than the same humans skipping the analysis on accounts 5 through 30 because they ran out of time.
Watch-outs
- Over-confident probability bands. The classifier will sometimes confidently bucket an account into
over 70 percentbased on a strong recent health-score trend that masks a stakeholder loss the Skill weighted too lightly. Guard: the band must always be paired with at least three independent drivers; if the Skill cannot produce three, it downgrades the band one level. The CSM treats every band as a hypothesis, not a forecast — and the band that says “likely renews” warrants the same scrutiny as the band that says “likely churns.” - Stale stakeholder map. Gainsight stakeholder maps go stale fast. Sponsors leave, titles change, someone gets reorged into a different BU and nobody updated the record. Guard: the Skill compares Gainsight roles against LinkedIn current titles where possible and flags every stakeholder it cannot reconcile within the last 90 days. If more than half the map is unreconciled, the Skill aborts with “stakeholder map staleness over 50 percent; refresh in Gainsight before re-running.” This refuses to generate a playbook on a fictional org chart.
- Scripted talk-tracks losing trust. If the CSM reads the talk-tracks aloud as scripts, the customer hears it and the renewal save dies. Guard: every talk-track is labeled “scaffolding for preparation, not a script” in the output, and the CSM is instructed to rewrite each one in their own voice before any conversation. The Skill does not produce conversational copy at all — only the structured objection, underlying-concern, posture, and bridge-question scaffolding.
- Treating the playbook as the work. The playbook is worthless if interventions are not converted to tracked Gainsight CTAs and reviewed weekly. The output’s footer reminds the CSM of this; the setup step ties generation to a CTA-creation step.
Stack
- Gainsight — health-score history, success-plan progress, stakeholder map, CTA tracking
- Gong — last 180 days of calls for archetype diagnosis and citation
- Claude (Sonnet or Opus) — diagnosis, segment-aware template selection, motion matching, talk-track scaffolding
- Artifact bundle at
apps/web/public/artifacts/cs-renewal-playbook-skill/(SKILL.md,references/1-segment-playbook-config.md,references/2-stakeholder-motion-matrix.md,references/3-output-format.md)