Customer conversation guide — for AWS staff¶
Last updated: 2026-09 · owner: Youngjin · volatility: medium
L0 TL;DR: Establish task, cost, and operating requirements before helping choose technology. This public guide contains no confidential customer or sales records. Show the Executive Brief and complete Start with the customer.
1. Meeting flow¶
- Task: identify repetitive work whose time, failures, or interventions should decrease.
- Alternatives: compare existing automation, products, robot SI, and model adaptation.
- Experiment: confirm rights, staffing, and limits; test one bottleneck.
- Exit conditions: agree on success, stopping, handover, and evidence for the next meeting.
2. Key questions¶
| Question | Include in the answer | Tool |
|---|---|---|
| ROI? | Total hardware/SI/data/staff/safety/operations cost and realizable benefit, beyond GPU | ROI |
| Why AWS? | Reuse of existing infrastructure, measured bottlenecks, needed services/support boundaries | Decisions |
| Staffing? | Robotics, ML, cloud, and site safety roles; partner/SI scope for gaps | Owners |
| When will we see results? | Range based on readiness/evaluation; no universal one-day or 12–24-month promise | Execution |
| Safety? | Tool permissions plus independent physical safety, local control, cancellation/recovery tests, site approval | Operations |
| Processing within Korea? | Check storage, inference, Memory/Evaluations, and external-tool routes separately | Evidence |
| Are open models commercial? | Version-specific code, weight, base-model, and dataset terms | P2 |
| Partner selection? | Comparable site outcomes, robot support, integration/safety responsibility, recovery/maintenance scope | Pilot card |
3. Handover and support¶
AWS staff clarify service architecture, quotas, pricing, and service support. Robotics/ML teams own data, model/device compatibility, and evaluation. SI/site teams own process integration, safety, and operating recovery. Do not assume sample maintainers offer a customer operations SLA. Record owners and unresolved items.
4. Promises to avoid¶
| Avoid | Say instead |
|---|---|
| “Seoul GA means no residency problem” | “We will inspect storage and processing per feature/model” |
| “OpenVLA is MIT, so commercial use is allowed” | “We will check code and weight terms separately” |
| “100 demos gives 80% in one day” | “We will measure data/time needs under fixed evaluation conditions” |
| “Cedar guarantees robot safety” | “Independent safety, local control, and site validation are also required” |
| “Open models eliminate lock-in” | “We will compare migration costs for models, data, and operating tools” |
➡️ Next action: share a one-page decision, spend/time caps, owners, and evidence required for the next stage.
owner: Youngjin · updated: 2026-09 · volatility: medium