A new AI feature is not yet a business case.

Before a pilot

Five questions worth answering.

1. 01Which specific task would change, and who owns it?
2. 02Is the capability available in our actual plan and environment?
3. 03What information leaves our systems, and who can access it?
4. 04What errors are unacceptable, and how will we test them?
5. 05Does the likely benefit justify implementation, software and ongoing review costs?

After the announcement

Test a representative task, not a polished example.

Use a non-sensitive sample and explicit acceptance criteria. Record the configuration and limitations so the result can be repeated. A vendor demonstration and your independent test should always be labelled separately.