Cookbook
Build the QBR deck before the agenda
Assemble the QBR deck from usage, tickets, and account history, with every claim tied to real data.
Customer Engineeringintermediate
What this does
- Pulls usage trend, ticket history, and contract facts since the last QBR
- Drafts the narrative, top wins, open risks, and the ask
- Drops any win not backed by a real usage or ticket data point
- Builds the deck and a text narrative, ready for the CSM to review
Workflow
- Trigger
QBR scheduled on the calendar
Or a manual kickoff two weeks out.
- Prompt
Pull the account story
Usage trend, ticket history, and value delivered since the last QBR.
- Prompt
Draft the narrative
Wins, risks, and the ask for next quarter.
- Shell
Build the deck
Turn the narrative into slides with the presentation skill.
- Result
Deliver for review
Deck saved to drive, summary posted to Slack.
Recommended skills
Reusable skills the recipe pulls in, add them once, every agent can use them.
QBR narrative structureWins, usage story, risks, and the ask, in that order.
Value-story groundingTie every claimed win to a real usage or ticket data point.
Presentation builderTurn the narrative into a formatted deck with the pptx skill.
How to use this recipe
- 1Connect PostHog, your support tool, Salesforce, and Slack
- 2Set the trigger to the QBR date on the calendar, or run it manually
- 3Paste the prompt and set what counts as a "win" for your policy
- 4Review the deck and narrative before the customer meeting
Recipe files
Copy the prompt
Install the "Build the QBR deck before the agenda" recipe from the Eluu cookbook. 1. Fetch https://eluu.ai/cookbook/qbr-pack/skill.md and save it into your skills folder as "qbr-pack". 2. Ask me its first-run questions and store my answers in memory. 3. Offer to schedule the recurring jobs listed in the skill; only create the ones I approve. 4. Then run the skill once end to end so we can review the output together.
eluu.ai/cookbook/qbr-pack.md
Paste it into any agent. It fetches the recipe file and pulls in the skills and prompts on its own, the prompt stays short, the detail lives at the URL.