5. Discover with the customer¶
Step 5 of 10 · Deliver to a customer
Step at a glance
🎯 Goal — Run the workshop (or /ingest-prd if a PRD/BRD/spec exists) and fill docs/discovery/solution-brief.md with numeric success criteria, KPI events, RAI risks.
📋 Prerequisite — 4. Clone for the customer complete; customer stakeholder workshop scheduled or PRD in hand.
💻 Where you'll work — Customer workshop room (or Teams) for the conversation; the cloned customer repo in VS Code + Copilot Chat to drive /discover-scenario after the workshop.
✅ Done when — docs/discovery/solution-brief.md has zero TBD; success criteria are numeric (baseline → target with %); 3–6 KPI events named; solution shape chosen; 3–5 RAI risks listed; accelerator.yaml updated by the chatmode.
Chatmodes used here
/ingest-prd (optional, if customer provided a PRD) · /discover-scenario
Full reference: Chatmodes overview.
The brief is the single source of truth for the engagement. Every downstream artifact — prompts, tools, retrieval schema, HITL gates, eval thresholds, dashboards — derives from it. A weak brief makes every later step harder.
The five discovery artifacts¶
Five artifacts ship under docs/discovery/. Use them in this order:
| # | Artifact | Purpose | Who fills it |
|---|---|---|---|
| 1 | use-case-canvas.md |
1-page exec alignment before you spend workshop time | Partner lead + customer sponsor, async |
| 2 | SOLUTION-BRIEF-GUIDE.md |
How to run the workshop that fills the brief | Read by partner lead; optional coaching for junior facilitators |
| 3 | discovery-workbook.csv — download |
Structured capture during the live workshop | Partner facilitator (typing) + customer SMEs (answering) |
| 4 | solution-brief.md |
Canonical engagement doc | Output of /discover-scenario |
| 5 | roi-calculator.xlsx — download |
Quantifies the hypothesis in Section 4 of the brief | Partner lead, after Section 3 of the brief is filled |
The flow¶
Before the workshop¶
- Use-case canvas — send to the customer sponsor after the first scoping call. If the sponsor can't fill it in ~30 minutes, the engagement isn't ready for a workshop. The canvas is a go/no-go gate, not a deliverable.
If the customer already provided a PRD / BRD / functional spec¶
If the customer handed you a written spec before the workshop, you can pre-draft the brief from it instead of starting blank:
…in Copilot Chat, pointed at the file (path inside the cloned repo or local path). Supported formats: .md, .txt, .docx, text-extractable .pdf. The chatmode runs scripts/extract-brief-from-doc.py, maps evidence to the 7-section brief schema, and writes a draft docs/discovery/solution-brief.md with a STATUS: AI-extracted draft banner and per-section <!-- evidence: ... --> comment blocks.
Treat every CONFIRMED field as a hypothesis (the LLM read the PRD, not the customer). Run the workshop on the remaining TBDs.
The full PRD-ingestion flow lives in Reference → Discovery how-to → "If the customer already provided a PRD".
During the workshop¶
- Discovery workbook — the facilitator types answers into the CSV live. One question per row keeps the conversation disciplined. The workbook is scaffolding; it gets replaced by the solution brief after the session.
Immediately after¶
- Solution brief — in Copilot Chat:
The chatmode produces the 7-section brief at docs/discovery/solution-brief.md and updates accelerator.yaml fields (solution.*, acceptance.*, kpis[]). It does not touch scenario: — that comes from /scaffold-from-brief in the next step.
If you ran /ingest-prd first, /discover-scenario detects the draft banner and enters gap-fill mode — asks only about TBD fields, preserves every confirmed field byte-for-byte, strips the banner and evidence comments on exit.
- ROI calculator — open
docs/discovery/roi-calculator.xlsx, fill blue cells on theInputssheet with numbers from brief Section 3 + Section 4. Read theROIsheet for annual savings and payback; read theKPIssheet for rendered baseline/target values.
Before scaffolding¶
- Walk the sponsor through the brief + ROI calculator together. If either has
TBDfields or the ROI doesn't clear the customer's hurdle rate, iterate before running/scaffold-from-brief. A scaffold is expensive to redo; a discovery redo is cheap.
What "good" looks like¶
- Every section filled — no
TBDleft when you exit the session. - Success criteria are numeric (baseline → target, with %).
- 3–6 concrete KPI event names picked — these become typed telemetry events in
src/accelerator_baseline/telemetry.pyand App Insights alerts later. - Solution pattern chosen: supervisor-routing (flagship default), single-agent, or chat-with-actioning.
- RAI risks listed as 3–5 concrete statements — these become redteam cases.
What to push back on (workshop facilitation)¶
- "We want it fast" → "What's the current time, and what does 'fast enough' mean to the executive sponsor?"
- "Just use AI Search" → walk through the Tool catalog to pick the right grounding tool (File Search vs Azure AI Search vs SharePoint vs Fabric — they each have different prereqs and auth stories).
- "Skip HITL, we trust the model" → walk the customer through the side-effect catalogue and ask which writes they're willing to defend without an approval trail.
Honesty notes¶
- The canvas and workbook are discovery capture; they don't replace the solution brief. If you stop at the workbook, downstream automation (lint, scaffold, evals) won't have the structure it expects.
- The ROI calculator is a hypothesis tool at discovery. Real numbers come from running the accelerator against the customer's data during Iterate & evaluate. Expect to reconcile against measured performance before production sign-off.
Continue → 6. Scaffold from the brief