1. Describe the task before choosing the technology
Write down the trigger, inputs, steps and final output. For example: an enquiry arrives, staff collect missing details, check relevant information and prepare the next action. Count how often it happens and where delays arise.
If the steps are fixed and the information is structured, conventional automation may be sufficient. AI is worth exploring where language, documents or variable inputs make the task harder to express as rules.
2. Choose a result you can check
Set a baseline from representative recent work. Measure staff minutes per completed task, response time, rework or missing information. Keep quality alongside speed so an apparently faster process does not simply move the correction work elsewhere.
An illustrative pilot might aim to prepare an enquiry summary that a staff member can check. The threshold for accepting it should be agreed with that team, using realistic examples.
3. Check information and access
Identify the source of truth and who is allowed to use it. Outdated documents, inconsistent records and missing API access can stop a project before model choice matters. Keep personal or confidential information out of an initial demonstration unless appropriate handling has been agreed.
4. Make the human review point explicit
Decide which outputs can be used directly, which need checking and which actions remain with a person. Test unclear requests, unavailable services and information the tool should not reveal.
5. Agree a decision at the end of the pilot
Define the test period, owner, expected operating costs and acceptance criteria. At the review, decide whether to expand, revise or stop. Keep the decision tied to completed work and quality, rather than the number of prompts or messages.
Bring this checklist to a free Intellagens consultation and we can discuss a first pilot for your business.