AI projects often start with a model demo before anyone has agreed what problem it should solve. A better first question is: which repeated task is costly, slow, or inconsistent today—and what would a useful improvement look like?
Map the current work
Write down who performs the task, what information they need, which systems they touch, where exceptions happen, and how success is checked. This reveals whether the real bottleneck is search, decision making, data quality, approvals, or handoff. AI may help with one step, but it may not be the right answer for the whole process.
Check the foundations
Confirm that the needed data is accessible and permitted for the intended use. Identify a workflow owner, an operational owner, and a human fallback. Decide what actions the system may take and which require approval. Define a baseline for quality, time, cost, and failure rate before comparing solutions.
Run a bounded pilot
Choose a narrow workflow and representative cases. Test the model and non-AI alternatives against the same outcome criteria. Include edge cases, failure recovery, privacy, and integration effort in the estimate. Agree on a stop condition before expanding scope.
A readiness review turns “we should use AI” into a decision the team can validate. It can also show that better process design or a small automation will deliver value sooner.
I share practical product and AI delivery notes at https://umer-amin.vercel.app/. What workflow would you assess first in your organization?
Top comments (0)