The easiest way to waste time with AI is to begin with a tool and search for a problem. A better starting point is a repeated business task with a clear owner, reliable inputs and a result that people can review.
This checklist helps a team evaluate an opportunity before automating it.
1. Is the problem clear?
Describe the task in one sentence. Good early candidates are narrow and frequent: summarising a standard report, drafting a first version of routine communication, classifying enquiries or identifying missing fields.
Avoid broad goals such as “improve marketing with AI.” Define the user, current process, delay or error and desired output.
2. Is the current process understood?
Document how the work happens today. Identify the trigger, inputs, steps, decisions, exceptions and final reviewer.
Automating a confused process usually makes the confusion faster. Simplify unnecessary steps before adding AI.
3. Are the inputs suitable?
Check whether the source information is available, consistent and permitted for the intended use. Ask:
- Are required fields present?
- Do teams use the same definitions?
- Is the information current?
- Does it include personal, confidential or commercially sensitive data?
- Is there permission to send that information to the chosen system?
If the inputs are unreliable, improve collection and governance first.
4. Can a person verify the output?
AI can produce a fluent answer that is incomplete or wrong. Define who reviews the output and what evidence they use.
Early projects should have a clear human checkpoint. The reviewer must be able to correct the result before it reaches a customer, employee, partner or public channel.
5. What happens when the system is uncertain?
List common exceptions and decide how they are handled. The process should route unclear cases to a person rather than forcing an answer.
For customer communication, set boundaries around refunds, health claims, contracts, prices and commitments. These areas need approved information and accountable review.
6. How will value be measured?
Choose a small number of measures before the pilot. Depending on the task, they may include:
- time saved per report;
- percentage of outputs accepted after review;
- error or rework rate;
- response time;
- completion rate;
- user satisfaction.
Include quality and risk, not only speed. A faster process that creates more corrections is not an improvement.
7. Can the team run a limited pilot?
Test one task, one team and a defined period. Keep a comparison sample from the current process. Record failure cases as carefully as successful examples.
At the end, decide whether to continue, revise or stop. A stopped pilot can be a useful result when it prevents a larger investment.
Takeaway: A practical AI opportunity has a clear problem, understood process, suitable data, human verification, exception handling, measurable value and a limited pilot.
Simple readiness decision
- Proceed to a pilot when all seven areas have a clear answer.
- Prepare first when the task is useful but data, ownership or review is weak.
- Do not automate yet when the risk is high, the decision is poorly defined or no one can verify the output.
AI readiness is mainly an operating question. Tool selection comes after the team understands the work, the evidence and the responsibility for the result.
Next step: Use the readiness questions on one recurring report or communication task and document what prevents a safe pilot.
Related: How to Ask Better Questions Before Reading a Business Dashboard
Last reviewed: 13 September 2026
