04 · Drill 03 — Bedrock vs Vertex vs First-Party

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

What this drill is. Pick the deployment surface for an Indian BFSI customer and defend the choice for 5 minutes against a skeptical CIO + CISO simulated panel.

What this drill is NOT. A vendor recommendation. Not a cloud sales pitch.

Prompt

"You're at a decision meeting with a large Indian BFSI customer. Their primary cloud is AWS Mumbai (~80% of workloads), but they have a strategic GCP relationship growing in their data team (~15%). Recently a new CIO has signaled willingness to evaluate first-party SaaS for AI workloads. The CISO wants to know which Claude deployment surface you'd recommend, and the CIO wants procurement story + roadmap velocity. You have 5 minutes."

Time box

Speak: 4:00–5:00.
Q&A: 3 minutes after.

The Strong-Hire structure

Beat 1 — Frame the decision (30 sec)

Don't rush to a recommendation. Name the three concerns the room will weigh: residency + IAM/audit, procurement + commercial, feature velocity. The right answer must satisfy all three.

Beat 2 — Walk the three options against the three concerns (2:30)

A 3×3 spoken matrix.

Surface	Residency + IAM/audit	Procurement	Feature velocity
First-party API	Verify residency for Indian BFSI; not cloud-IAM-native	Adds an Apic contract	Newest features first
Bedrock Mumbai	Strong: ap-south-1, IAM-native, CloudTrail	Rolls into AWS spend	Feature lag possible
Vertex Mumbai	Strong: asia-south1, IAM-native, Cloud Audit Logs	Rolls into GCP spend	Feature lag possible

"Given 80% AWS, fresh CIO appetite for new vendors, and CISO weight on residency: Bedrock Mumbai for the API workloads, and Claude for Work for the end-user productivity layer. The reasons — IAM-native against existing CloudTrail and KMS, residency anchored, procurement clean. The trade-off I'd surface explicitly is feature lag — newest Claude features land first-party first. We mitigate by designing the eval framework to accept model substitution, which means new capabilities can be validated and rolled in within a sprint when they reach Bedrock."

Beat 4 — Address the CISO + CIO concerns explicitly (1:00)

CISO: "CloudTrail-native logging, KMS for encryption, IAM-native access control. We won't introduce a new logging pipeline you have to audit separately."
CIO: "Procurement is one Bedrock line-item rather than a new vendor contract. Roadmap velocity gap is real but bounded — we'll show you the substitution path when newer capabilities ship."

Rubric

Strong Hire

Decision framed before recommended (don't rush).
3×3 matrix audible (3 surfaces × 3 concerns).
Recommendation specific (surface + product layer combination), not generic.
Trade-off named and mitigated, not hidden.
CISO and CIO addressed in their own concerns, separately.
4:00–5:00 wall clock.

Hire

Recommendation lands but rationale is generic ("Bedrock is what most banks use").
One concern (CISO or CIO) under-addressed.
Feature-lag trade-off named without mitigation.

Lean No

"It depends." (No recommendation.)
Generic vendor comparison without BFSI-specific framing.
Skipping the procurement axis.

Strong No

Recommending something the customer can't actually do (e.g. residency-incompatible).
"Apic is the safest cloud" — wrong axis.
Disparaging a hyperscaler.

Q&A defense — likely follow-ups

Q — "What if our GCP team pushes back?"

A — Open the door for the data team to use Vertex Mumbai for their workloads — multi-surface Claude is fine if eval framework is consistent. The decision isn't AWS-or-GCP company-wide; it's per-workload.

Q — "How do we avoid Apic vendor lock-in?"

A — Three layers of substitutability. Same answer as discovery objection 3 in Module 03. Reference Module 03 Drill 03 — Three Discovery Objections.

A — Customer is multi-cloud / cloud-agnostic and wants the newest capabilities (extended caching, just-shipped tool features) for a flagship use case. Less common in mature BFSI; more common in Tier-2 / new-AI buyers.

Q — "How does Claude for Work fit in if we're using Bedrock?"

A — Different layer. Bedrock is the API/integration layer; Claude for Work is the end-user productivity layer (Projects, Desktop). You can run both — most mature BFSI customers do.

Common Lean No traps

Trap 1 — Cloud loyalty / rivalry

"AWS is better than GCP." Strong No. Not the question.

Trap 2 — Single recommendation, no trade-off

"Bedrock Mumbai." Period. Lean No.

Trap 3 — Skipping Claude for Work

The CIO's "end-user productivity" concern goes unaddressed.

Trap 4 — Marketing-speak feature velocity

"Apic ships fast." Without the substitution-path answer, this is just a hedge.

Trap 5 — Ignoring the CISO

Speaking to the CIO and not the CISO. The CISO is the most likely veto.

How to run this drill

Cold record. Speak the 5-min answer.
Have a partner play CIO + CISO and run 3 Q&A follow-ups.
Score against rubric.
Re-run with a variant (GCP-shop instead, or first-party-leaning new buyer).
Application trigger: 2 Strong Hires across 2 variants.

Cross-references

Strong-Hire bar for this drill

3×3 matrix audible without reaching for slides.
Recommendation specific (Bedrock Mumbai + Claude for Work).
CISO and CIO addressed in their own languages.
Feature-lag trade-off mitigated, not hidden.