CMSIT BUSINESS GUIDE

How to scope an AI workflow pilot for your business

By CMSIT · Published 26 October 2026

Start an AI project with one business task and a measurable outcome. A useful pilot tests whether the workflow improves in practice, including the time people spend checking and correcting the result. This checklist helps a business buyer prepare an AI integration brief.

1. Choose one workflow with an owner

Describe the current input, steps, output and person responsible. Examples to assess include drafting a response for review, finding approved information or preparing an internal document. These are possible use cases, not promises that a particular tool will suit your business.

Record the current workload, time per task, error rate and what a successful result looks like. If the process itself is unclear, clarify it before commissioning automation.

2. Map the systems and permitted data

• List the applications the pilot must read from or write to, and who owns each one.
• Identify the information needed and what must stay outside the pilot.
• Ask where data is processed and retained, who has access, and what the supplier terms allow.
• Check whether the relevant integration and access permissions are available.

Use approved sample material for early assessment. Do not send passwords, personal records or confidential business documents through a website enquiry.

3. Set human review and stop conditions

Decide which outputs need checking before use and who can approve an action. Agree what happens when information is missing, an answer is wrong or a connected service fails. Keep a workable manual path while evaluating the pilot.

NIST's AI risk management framework provides a useful structure for mapping a task, measuring performance and managing risk. It supports a considered evaluation; it does not certify that a particular pilot is safe or effective.

4. Measure the complete workflow

Compare a representative set of tasks against the existing process. Include preparation, checking, corrections and ongoing operating costs. A fast draft is not a saving if it takes longer to verify.

• Agree an acceptable output quality and a practical review method.
• Count successful tasks and failures, not only attractive demonstrations.
• Measure the time people actually spend from start to finish.
• Decide who can stop, change or approve expansion of the pilot.

5. Compare implementation and ongoing responsibilities

Ask for separate scope and costs for discovery, integration, configuration, evaluation, staff training and support. Clarify model or software charges, access changes, maintenance, documentation and exit arrangements. Discuss hosted and private options against your requirements rather than assuming either is automatically suitable.

A practical first project brief

Tell CMSIT the task, its frequency, the systems involved, the current difficulty, the project owner and the result you want to measure. A plain-language example without sensitive records is useful. We can then discuss whether an AI pilot, a simpler automation or a process change merits further assessment.