The demonstration is convincing. A request arrives, the AI interprets it and a tidy response appears. Before agreeing the project, ask what happens when the request is incomplete or your operations lead is away.
An AI implementation partner should explain the operating process, evidence of quality, ownership and limitations in terms your team can review. Bring one real workflow to make the discussion specific.
Ask them to define the boundary
Which inputs will the solution accept? What will it produce? Which decisions remain with your team? Ask for examples deliberately outside scope.
A proposal to automate enquiries needs detail. Reading a message, suggesting a category, drafting a response and sending it involve different responsibilities. Ask where the proposed engagement starts and finishes.
Ask for evidence
Use these questions:
Which representative cases will we evaluate together?
What counts as an unacceptable error?
How are missing information and conflicting sources handled?
What can a reviewer inspect before approval?
How will changes be tested after handover?
Ask for a failure demonstration too. Supply an authorised example with an ambiguity and see whether the proposed solution exposes it or hides it behind a confident answer.
Establish what you will own
Agree who maintains instructions, documents, access requirements and connected workflows. Clarify which materials you receive and what staff need to operate them.
Ask about dependencies, ongoing charges, support boundaries and ending the engagement. Check the actual proposal rather than relying on a sales assurance.
Include your administrator or adviser where technical or contractual questions require them. A useful operational discussion makes those dependencies visible rather than pretending one person can settle everything.
Define the handover decision
Set acceptance criteria before work begins. Include staff training, documented exceptions, a named owner and a fallback if the solution cannot perform an agreed task.
Ask who resolves defects after launch and how you distinguish a defect from a new requirement. That distinction can prevent avoidable disagreement later.
Related guidance: guide to mapping the work before changing it.
If you want help turning an AI proposal into operational requirements, Hili Consulting can help prepare the questions and acceptance checks before you commit.
