A Claude Skill that turns a single Gainsight account into a QBR narrative plus a slide-by-slide outline: an executive headline, a usage-and-outcomes story keyed to the account’s success-plan goals, an open-risks table, and a ranked set of expansion plays. The CSM gets a structured Markdown draft — one block per slide, in the order the deck runs — that they edit and drop into the team’s deck template, instead of staring at a blank outline. The artifact bundle ships SKILL.md plus three reference files the CSM team adapts once and reuses across every account.

This is the narrative-and-outline cousin of the QBR prep skill, which pulls Salesforce and Gong as well. If you live in Gainsight and want the deck story straight from your CSP’s account record, scorecard, and success plan, start here.

When to use

You are a CSM prepping a QBR for one named account, Gainsight is your system of record for health, usage, and success-plan goals, and you want a populated narrative and slide outline you can edit rather than a blank deck. The Skill is built for the case where the QBR story has to braid four things together — where the account is against last quarter’s commitments, what the usage and outcome data say, what risks are open, and where the credible expansion is — and land it in a slide order a human can refine.

It produces the most useful output when the Gainsight scorecard is configured (not just the default health pill), the success plan has goals with target dates and status, and usage measures are populated against a real baseline. For accounts that meet that bar, the draft lands close to deck-ready. For accounts that do not, the Skill flags the gap by name rather than writing a confident draft on top of missing data.

When NOT to use

Do not use this Skill to auto-send a QBR deck. It drafts a narrative and an outline; it does not render slides and it does not replace the CSM’s read of the relationship. Every deck gets a human pass and an account-team sign-off on framing before the customer sees it.

Do not point it at an account with under 30 days of usage data in Gainsight, an empty or never-configured scorecard, or a success plan with no goals logged. The Skill is built to flag and refuse rather than pad with generalities — but only if you honor the refusal. Overriding the INSUFFICIENT_DATA flag produces a draft that reads fine and misleads the customer.

Do not use it as a renewal-likelihood or churn model. The risk table is QBR framing — what to talk about in the room — not a calibrated retention score. If you need a defensible health number with a why-changed explanation, build the composite health score in n8n instead and read this Skill’s risk table as commentary, not signal.

Do not use it for internal or board reviews. The narrative and tone passes assume an external customer audience and a renewal-or-expansion frame.

Setup

Roughly 45 to 90 minutes the first time, almost all of it spent mapping your deck’s slide order to the Skill’s outline vocabulary and dropping in real voice samples. After the first account, runs are a couple of minutes.

Install the Skill. Drop the bundle from apps/web/public/artifacts/qbr-deck-builder-skill/ into ~/.claude/skills/qbr-deck-builder/. It exposes one command, build_qbr_deck(account_id, quarter), plus internal helpers for the Gainsight pulls, success-plan parsing, and the two-pass Claude pipeline.
Wire the Gainsight credential. Set GAINSIGHT_API_KEY and GAINSIGHT_DOMAIN with read access to Company, Scorecard, Success Plan / CTA, and the usage/adoption objects your org populates. The Skill reads only; it never writes back to Gainsight. If your usage data lives in a separate adoption object, set GAINSIGHT_USAGE_OBJECT to its API name — the Skill validates the field names against references/1-deck-outline-map.md and refuses to fill the usage slide if they drift.
Map your deck outline. Open references/1-deck-outline-map.md and replace the slide manifest with your team’s actual slide order and titles. The default outline is executive-headline, where-we-were, usage-and-outcomes, open-risks, expansion-plays, asks-and-next-steps — re-order or rename to match the deck you actually present.
Pin the success-plan shape. Open references/2-success-plan-format.md and either adopt the schema verbatim or describe how your team logs goals in Gainsight (CTA type, the field that holds the target date, the picklist that holds status). The Skill needs a stable shape to parse goal-by-goal; mismatched shape is the most common cause of an empty outcomes story.
Drop in voice samples. Replace the placeholder in references/3-sample-deck.md with three to five anonymized prior QBR narratives from your CSM team so the tone pass has real material to mimic. With no samples the Skill writes in a neutral register and flags TONE_REVIEW_NEEDED.
Run for one account. build_qbr_deck(account_id="1P01...XYZ", quarter="Q2-2026", last_deck_path="..."). The Skill writes one Markdown file with a fenced block per slide, plus a one-paragraph executive summary as a separate file. Read it top to bottom, edit, and paste into the deck.

What the Skill actually does

The Skill pulls four things from Gainsight in one batch: the Company record (ARR, renewal date, segment, lifecycle stage), the current scorecard (each measure, its score, and its trend versus last period), the active success plan (each goal with target date and status), and the usage rollup against the account’s baseline. If any pull returns empty, the Skill records unavailable for that input and threads it through to an INSUFFICIENT_DATA flag on the affected slide rather than inventing content. Gainsight is the single source here on purpose — one credential, one account record, one scorecard definition — so the draft is reproducible and the data lineage is obvious to the CSM editing it.

It then runs two Claude passes, not one. Pass one is synthesis: Claude reads all four inputs plus the prior QBR deck and builds an internal scratchpad — the wins keyed to success-plan goals, the gaps, the usage-and-outcomes story against baseline, the open risks ranked red/yellow/green, and the expansion plays ranked by signal (usage trend, scorecard strength, contract headroom) and confidence (high, medium, low). Synthesis is its own pass because the next pass needs one coherent picture; folding synthesis and slide-writing into a single pass makes the slides over-weight whichever input Claude read last, and the expansion ranking degrades to a list rather than a ranking.

Pass two is narrative-and-outline. Claude reads the voice samples from references/3-sample-deck.md and rewrites the scratchpad into the team’s voice — neutral, data-led, no superlatives — then maps it onto the slide order from references/1-deck-outline-map.md, one fenced block per slide. Keeping the outline mapping in the second pass means re-ordering or renaming slides only re-runs this pass; the synthesis scratchpad is reused. The executive summary is generated last from the finished narrative so it never contradicts the slides beneath it.

The output is a single Markdown file — one fenced block per slide, in deck order, each block headed with the slide title from your map — plus a separate one-paragraph executive summary. The CSM always edits before the deck is built. The Skill is a draft engine, not a publisher.

Cost reality

A full run costs roughly 12,000 to 22,000 input tokens and 3,000 to 6,000 output tokens on Claude Sonnet — call it 5 to 12 cents per QBR at current Sonnet pricing. It is cheaper than the Salesforce-plus-Gong prep skill because there are no call transcripts in the input; the Gainsight pulls are structured records, not 60-minute transcripts. Wall-clock time is one to three minutes per account, dominated by the Gainsight API batch in pass zero; the two Claude passes add 30 to 60 seconds.

A CSM building a QBR narrative from scratch typically spends 60 to 120 minutes per account pulling the scorecard, reconstructing last quarter’s commitments, and drafting slide text. The Skill takes that to 20 to 40 minutes of editing, so the saving is roughly an hour per QBR. A book of 25 accounts at one QBR per quarter is about 25 hours saved per quarter per CSM — against an Anthropic spend under $3 per quarter for that book.

Success metric

Track time from “outline generated” to “deck sent for internal review” per QBR; the Skill should pull the median under 45 minutes within the first quarter of use. Watch the rate of drafts carrying INSUFFICIENT_DATA, SUCCESS_PLAN_STALE, or TONE_REVIEW_NEEDED flags — those are leading indicators of upstream hygiene problems (scorecards left on defaults, success plans not maintained, no voice samples) that the Skill surfaces by design; a healthy quarter trends them down. Also track the share of generated narrative that survives the CSM edit pass: aim for 70% or higher. Below that and the Gainsight data needs work; above 90% and the CSM is probably under-editing a draft that should never ship unedited.

Versus the alternatives

Versus the Gainsight-native QBR / Success Snapshot. Gainsight ships templated QBR and Success Snapshot exports that auto-populate scorecard measures, usage charts, and NPS into a slide layout. If you want fields rendered into a deck, that is less setup and the obvious default — it is the product you already pay for. The gap it leaves is narrative: it surfaces measures, it does not write the story that connects this quarter’s usage to last quarter’s commitments, and it does not rank expansion plays in prose. This Skill writes that story from the same Gainsight data. Use the native export for the charted slides and this Skill for the narrative slides; they are complementary, not competing.

Versus the Salesforce-plus-Gong QBR prep skill. That skill adds Salesforce account history and Gong call themes to the mix and is the right pick when call-derived color (executive quotes, competitor mentions) is load-bearing in your QBRs and you have the Gong coverage to support it. It costs more per run and needs three credentials. This Skill is the right pick when Gainsight is your source of truth, you do not have or do not want to wire Gong, and the scorecard plus success plan carry the story. Run the Gong-backed one for strategic accounts with rich call history; run this one for the long tail you steer from the CSP.

Versus writing it by hand. Manual prep produces the best QBR for a top-tier account because the CSM carries context no record holds — the off-record budget conversation, the champion who is quietly leaving. The cost is the hour-plus per account, which does not scale across a 25-account book. Use manual prep for the handful of accounts where the relationship context is the deck; use the Skill for the rest, and even on the top-tier accounts start from the Skill’s draft and edit harder.

Watch-outs

Default scorecard, confident story. An account whose scorecard was never configured past the default health pill produces measures that mean nothing, and a narrative built on them reads plausible and says nothing. Guard: the Skill checks whether the scorecard has more than the single default measure populated; if not, it writes the usage-and-outcomes slide as INSUFFICIENT_DATA: scorecard not configured rather than narrating noise.
Stale success plan. A success plan untouched for 60-plus days yields a confident-sounding but outdated progress story. Guard: the Skill reads each goal’s last-modified date; if the plan’s most recent edit is older than 60 days it prepends SUCCESS_PLAN_STALE to the where-we-were slide so the CSM has to confirm the goals still hold before presenting them.
Wrong prior deck. If last_deck_path points at another account’s or quarter’s deck, the “commitments then versus reality now” framing breaks silently and the narrative reports against the wrong promises. Guard: the synthesis pass extracts the account name and quarter from the prior deck’s first slide and halts with a mismatch error if either disagrees with the inputs.
Tone mismatch with the customer. The default voice is the CSM team’s, which may not match how an enterprise buyer expects to be addressed versus a startup. Guard: when voice samples in references/3-sample-deck.md include decks that customer actually received, the tone pass weights them over internal docs; with no samples it writes neutral register and flags TONE_REVIEW_NEEDED on the executive summary.
Expansion plays that outrun the data. Asked to rank expansion, a model will manufacture a plausible upsell even when no usage or contract signal supports it. Guard: the expansion ranking is constrained to plays backed by a named signal (a usage measure trending up, scorecard headroom, contract seats below entitlement); a play with no signal is dropped, not downgraded, and if none qualify the slide says so rather than padding.

Stack

Gainsight — Company record, scorecard, success plan, and usage rollup; read-only source for the entire draft
Claude — two-pass pipeline: synthesis (wins, risks, expansion, usage story) then narrative plus slide-outline mapping (Sonnet recommended for cost; Opus only if voice match matters more than budget)
Your deck tool — Google Slides or PowerPoint; the CSM pastes the slide-by-slide Markdown into the team template after editing

Edit this page on GitHub