Auto-generated QBR

A draft QBR is useful when every slide cites usage, tickets, or words the customer already said, and you still own the story in the room. The job is quality of evidence, not speed of slide production. If the generator cannot show where a number came from, cut the number.

You are not asking the model to close the quarter for you. You are asking it to assemble a first pass from systems you already trust: product usage, support history, quoted outcomes, and commercial signals. Customer-success platforms in the Gainsight class, CRM records in the Salesforce class, and conversation captures in the Gong class each hold part of that pack. Treat them as one evidence layer, not as competing vendors and not as a ranked stack.

What "done" looks like: a deck the exec can challenge. Licensed versus working usage, ticket themes you will admit, outcomes in the customer's phrasing, expansion as questions rather than a SKU ladder, and speaker notes that name the source for each claim. You confirm every figure before anyone outside CS sees the file.

If health or usage is already pointing at risk, do not hide it behind a polished narrative. Read the draft next to your churn-risk prediction model and decide whether this meeting is a standard review, a risk review, or both. A risk score is context for you. It is not a slide title.

Feed the generator only what you would put on a shared screen.

Usage belongs in as licensed versus active, by product area, for this review window and the previous one. Prefer the same counts the customer can open in their admin view. Event vanity (logins, notification clicks, marketing-email opens) is not proof of value. If a module was provisioned and never used, keep it in the pack as unused, not as adopted.

Tickets belong in as themes, severity, open versus closed, and whether anything is still owed. A closed incident can live on a "what we fixed" slide. An open high-severity issue cannot live on a success slide. Do not let the model summarize a rough support period as "steady."

Quoted outcomes belong in as sentences the customer said on a call, in mail, or in a prior review, in their language. "We stopped reconciling in the spreadsheet" is a quote you can read aloud. A productivity multiple nobody in the account stated is not. Do not merge two periods into one triumphant sentence.

Expansion signals belong in as facts: seat requests, new teams asking for access, stalled pilots, a competitor named on a call, a budget date on the opportunity. These are inputs to a question you may ask. They are not a ranked list of SKUs to sell.

Bound the window with explicit dates. Strip internal-only comments. Strip unused modules from any "delivered" list before generation, or the model will copy catalog text into the win column. If the champion has gone quiet or left, pull champion-departure monitoring into the same pack so you do not present last quarter's sponsor as this quarter's buyer.

Keep the deck short enough to discuss. A working order:

Period and who is in the room, so the exec knows which window you are reviewing.

Outcomes they already claimed, each tied to a quote plus the usage or ticket that sits under it.

What they used, by area, including what they licensed and barely touched. Unused is a finding.

What broke and what you closed: ticket themes, not a dump of IDs.

Risks you will say out loud: adoption gaps, open issues, people changes.

Commitments you can staff next period, not a product-roadmap paste.

Expansion as questions. "Another team asked for seats; do you want a scoped conversation?" is a question. "Your top upsell is A, then B, then C" is a ranking. Do not generate that ranking.

Illustrative pattern (not a measured case, and not a set of results). You are preparing a quarterly review for an operations lead. They bought a core workflow SKU and an analytics add-on. Daily work shows up in the workflow area. The analytics add-on has licenses and almost no queries. Tickets in the period were about export formats and are closed. On the last review call they said they stopped a weekly spreadsheet reconcile because the workflow now holds the source of truth. CRM shows another regional team asked for seats. The opportunity is still unqualified.

A defensible draft puts the spreadsheet sentence on outcomes, with workflow usage underneath. It puts the export tickets on "what we fixed." It lists the analytics add-on under licensed but not in the working set. It raises the regional request as an open question, not as recommended SKU number one. It does not compute hours saved, cost avoided, or a productivity multiple, because nobody in the account quoted those figures this period.

If auto-renew is on and usage just dipped, pair the draft with your auto-renew and anomaly detector so the exec meeting is not the first time anyone names the dip.

Inventing a value metric. Generators fill empty outcome slots. If the pack has usage and no customer-stated result, the draft will reach for time saved, cost avoided, or ROI. Delete those slides. Either ask them for their metric in the meeting or keep a qualitative quote only. A figure you cannot open in their system is a figure you should not show.

Ranking expansion SKUs. Scoring "best next product" from fit, margin, or attach logic turns a review into a pitch. Executives feel that shift. Keep expansion as evidence: this team asked, this module sits unused, this competitor was named. Let them choose whether to discuss commercial next steps.

Copying unused features as delivered. "Enabled in the tenant" is not adoption. If the analytics add-on was provisioned at kickoff and never queried, it is not a value theme. Putting it under "what you got this quarter" teaches the exec to distrust every other slide. Put it in white space, with a question: keep, train, or drop at renewal.

Related traps: recycling old quotes; using a clip from a champion who left; calling a rough ticket stretch "calm." If the period was rough, say so, then show what closed.

When the account is already in a save motion, do not let the QBR draft compete with the save path or bury risk under unused-feature wins. Run the save on your save-play recommender track, and keep this deck honest.

Sit with the draft as you would sit with an analyst. For each slide: source, date, and whether you will say the sentence out loud.

Confirm usage against the admin view or export you trust, not against a description of a dashboard. Confirm ticket counts in the system of record. Confirm quotes by rereading or replaying the original; if the wording drifted, restore the exact sentence or drop it. Confirm expansion signals with the account executive. Unqualified interest stays a question. It does not become a commit on a slide.

Anything that fails confirmation comes out. Do not soften an invented ROI into "estimated value." Cut it. Do not keep a ranked SKU list or unused modules in the delivered column.

Freeze a version for the exec send. Keep a speaker-note copy with citations. You still run the room: which slides to skip, when to raise expansion, when to stop and listen. It does not replace your judgment or invent the customer's ROI. After they agree actions, fold those actions into the auto-drafted success plan so the next draft starts from commitments they heard, not from a fresh feature catalog.