Executive Brief — business value and investment approval¶
Last updated: 2026-09 · owner: Youngjin · volatility: medium
L0 TL;DR: Choose one business task and compare existing automation and products. Expand investment after a bounded experiment establishes benefit, total cost, and operability.
① Why now?¶
Public models and simulation/training assets can be evaluated, but product releases and service GA do not establish customer profitability. Connect model choices, execution paths, and evidence to form a hypothesis for your environment.
② What does this mean for our industry?¶
| Example task | Measure first | Alternatives |
|---|---|---|
| Manufacturing inspection/part handling | Defects, rework, cycle time, interventions | Existing vision/control improvements, commercial cells, SI |
| Logistics transport/loading | Throughput, downtime/recovery, environmental changes | Existing AMR/fleet products, process improvements |
| Robot product development | Data preparation/training/evaluation time, physical failures | Existing development-stack improvements, cloud experiments |
These are discovery examples. A particular model or humanoid is not the default answer for every task.
③ What is real and what is hype?¶
| Decision | Required conditions | Action |
|---|---|---|
| Bounded experiment | Measurable baseline, usage rights, owners, spending cap | Approve the pilot card |
| Site pilot | Improvement on separate evaluation; latency/safety/recovery test plan | Limited scope under supervision |
| Expand operations | Site outcomes, actual operating cost, interventions/recovery, support owners | Expand in stages |
| Hold/alternative | Unclear rights/safety ownership, more economical alternatives, repeated missed targets | Resolve gaps or choose another approach |
Do not use a generic “invest now because GA” or “results in 12–24 months” timetable. Separate product release, reproduction, and customer-site validation.
④ Where do we start?¶
Complete the eight discovery questions and compare the current approach. If learning is needed, choose data collection, simulation, or fine-tuning for the present bottleneck. Separate preparation time from training runtime.
Calculate ROI on total cost: initial hardware, integration, data, staff, and safety validation; recurring maintenance, interventions, and cloud. Check payback in low/base/high scenarios using the ROI worksheet.
⑤ Why work with AWS?¶
Assess reuse of existing AWS data, security, and operations; measure small, then add services only as needed. Decisions covers single-GPU, managed, and multi-node choices. Assign cloud support, model-provider support, and robot/SI site responsibility separately.
Seoul availability does not guarantee Korea-only processing (AgentCore evidence). Code licenses alone do not establish commercial weight rights (OpenVLA evidence). Evaluate portability without promising “no lock-in.”
To begin¶
The business owner, technical team, and AWS contact complete the pilot approval card: one task, baseline, spending/time caps, operations/safety owners, success/stop criteria. Then follow execution paths and release gates.
owner: Youngjin · updated: 2026-09 · volatility: medium