Compare equivalent requirements. Include assistance used in purchased-product configuration as well as custom development.
Workflow / pilot scope / owner:
Evaluator (independent of the builder where possible) / date:
Shared acceptance criteria, including access, export, recovery, and handover:
Data readiness (is the data the workflow needs available, clean, and permitted for this use?):
Skills available to review, operate, and change what is built:
Comparison
| Measure | Product configuration | Conventional build | AI-assisted build |
|---|---|---|---|
| Requirements effort | |||
| Implementation effort | |||
| Review and correction | |||
| Integration and deployment | |||
| Quality / security findings | |||
| Test coverage achieved | |||
| Remaining release work | |||
| Second-person handover result | |||
| Estimated recurring effort | |||
| Direct cost / comparison period, including model or tool usage fees | |||
| Cost per unit of work at expected volume |
AI tools / configurations / assistance used, with versions and dates:
Prompts or instructions retained for reproducibility (where):
Sample and experience limitations:
Checks on the generated result
- [ ] Generated code and configuration were reviewed by a person who can explain them.
- [ ] Dependencies added by the assistant were checked for maintenance status, license, and known vulnerabilities.
- [ ] No credentials or sensitive data were embedded in generated code, prompts, or logs.
- [ ] Automated tests exist and were run by someone other than the builder.
- [ ] Licensing and provenance of generated material are acceptable for the firm's use.
- [ ] The result still works if the AI tool used to build it is unavailable.
Ownership exercise
Have a second person operate the result, diagnose a known failure, make a small change, and explain the verification. Record dependence on the original builder.
Observed gaps:
Decision and supporting evidence:
Operating owner / remaining conditions:
Reconsideration triggers / next review (tool or model change, volume change, security finding, builder departure):
Common failure modes
- Speed of first draft is measured; time to a reviewed, tested, released result is not.
- The assistant adds dependencies nobody chose or checked.
- Nobody on the team can change the result without the same assistant and the same prompts.
- Pilot effort by an expert is projected onto the whole team.
Keep observed pilot effort separate from projected future savings.
This evaluation covers the build method. It does not decide whether to build at all; use the build, buy, or combine memo for that.