Skip to content

03 · Case Study 01 — BFSI Discovery Script

Status: Outline. Body fills in Week 2. Voice: principal-level, BFSI-threaded, Apic-calibrated.

What this file is. The worked-out 30-minute discovery script for the running BFSI customer's call 1 — what the architect would actually say, in order, with the partner's likely responses and the architect's reactions.

What this file is NOT. A pitch script. Not a generic discovery template — calibrated specifically for the running BFSI customer's profile.

Customer recap

Indian BFSI institution, multiple regulated business units, ~50K employees, RBI scrutiny, DPDPA-2023, fresh CISO scar from a previous vendor's hallucination during a demo. Six candidate use cases (per Module 05 INDEX) but the customer hasn't decided which to pilot.

Room: CIO (host), CDO (champion candidate), CISO (skeptic). AE has already done warmup and hand-off.

→ Full customer profile: Module 01 BFSI Customer Intro.

Script anatomy (30 minutes)

Time Phase Architect Customer
0:00–0:03 Architect intro 3-min intro per Module 02 case study Listening
0:03–0:08 Customer context Open questions CIO does most of talking
0:08–0:18 Problem surfacing Probing, careful CDO leads; CISO interjects
0:18–0:24 Constraint enumeration Specific, technical All three speak
0:24–0:28 Pattern recognition 1–2 customer parallels, no proposals Listening
0:28–0:30 Hand-off "What should I have asked?" Surfaces the constraint they were waiting to volunteer

The script (placeholder)

Body to be filled in Week 2 with the actual question wording, anticipated customer responses, and architect's reactions. The skeleton below holds the slots.

0:03 — Customer context

  • "Walk me through what your team does today — what's in production with AI, what's in pilot, what's only an idea."
  • Likely response: CIO talks; mentions a previous internal-search POC (not the burned one).

0:05 — Use case discovery

  • "Of the six use cases on your radar — support agent assist, internal SOP search, RM copilot, compliance doc review, developer productivity, exec analytics — which one keeps coming back in conversations? Which one would your CFO notice if it shipped?"
  • Likely response: customer either picks one (clear champion) or says "all six matter."
  • If "all six matter": "Which one carries the lowest blast radius if it failed publicly? That's the one to start with."

0:10 — Surface the previous-vendor scar (if not volunteered)

  • "You mentioned earlier you'd worked with another vendor on something similar. What did the previous attempt teach you?"
  • Expected: CISO surfaces the hallucination story.
  • Architect's response: empathy, no competitor disparagement. "That's a class of failure we'd want to design against from the start — eval-gated rollout, citation-grounded retrieval, never autonomous on first deployment."

0:18 — Constraints

  • Residency: "Where does the underlying data live? Mumbai-only, or open to Singapore?"
  • Integration: "What systems would we be reaching into — ServiceNow, Salesforce, SharePoint, Confluence, Snowflake?"
  • Sign-off: "Beyond the room here, who else has to bless this before production?"
  • Timeline: "What's the time horizon for first-value? And what's driving that timeline?"
  • Budget shape: "Per-use-case budget, or pooled? Is this OPEX or CAPEX-shaped?"

0:24 — Pattern recognition

  • "Two parallels worth flagging, not as proposals but as patterns we've seen. First — the previous-vendor scar is recoverable; the architecture pattern that closes it is eval-gated rollout with citation-grounded retrieval. Second — the six use cases share more than they diverge; we'd suggest scoping them as a portfolio with shared eval infrastructure, not as six separate POCs."

0:28 — Close

  • "Two questions before we wrap. One: what should I have asked you that I didn't? And what's a good answer? Two: who else needs to be in the next conversation?"

What the architect would not do

  • Propose model selection in this call.
  • Compare Claude to OpenAI/Gemini.
  • Reach for slides.
  • Ask the CISO for a "list of compliance requirements." (You research RBI/DPDPA before the call; the CISO tests whether you know.)
  • Try to close on a POC scope by minute 30.

What this discovery output feeds

  • A 1-page problem-and-constraint definition (the gate to Stage 2 — Eval Design).
  • Module 04: Claude platform decisions calibrated to this customer's residency + integration shape.
  • Module 05: prioritized use-case sequence (probably internal SOP search first; lowest blast radius, clearest eval).
  • Module 06: eval framework kickoff for the chosen first use case.
  • Module 09: CISO architecture review prep for call 3.

Cross-references

Strong-Hire bar for this file

  • Script lands all 6 phases in 30 min, with the architect ≤30% speak-time.
  • Previous-vendor scar surfaced or addressed without disparagement.
  • Constraint enumeration covers residency + integration + sign-off + timeline + budget.
  • "Use cases as portfolio with shared eval infra" framing audible by minute 26.
  • Hand-off output is a 1-page problem-and-constraint definition, not a POC commit.