How to choose your first AI deployment

Name the unit of work

Describe one request moving through the business: an inbound service inquiry, an estimate follow-up, or an employee question. Record the trigger, systems touched, decisions made, exceptions, and final record. “Improve customer experience” is an ambition. “Route a service inquiry to the correct next action with its context intact” is a workflow you can examine.

Find the cost of today’s process

Ask the people doing the work where time, rework, uncertainty, and avoidable handoffs occur. Capture the current volume and the effort for a comparable group of requests. Identify missing records during discovery. Include the work required to review AI output and maintain the system.

Prefer bounded decisions

Choose a first scope whose actions and consequences can be described. A system that drafts an answer for approval has a different risk profile from one that changes a contract or commits a customer to a price. Make human approval, escalation, and the fallback process explicit. Complexity should earn its place through an operating need; it is not itself evidence of value.

Make the funding decision reversible

Write down the success criteria, test cases, required access, accountable owner, and the conditions that would stop the pilot. Agree on what must be learned before expanding the scope. A discovery outcome may be to proceed, narrow the workflow, improve the data, or defer the deployment. Each is a useful answer when it prevents an unsupported implementation commitment.