Where Should Your Business Start With AI?

Look for friction you can describe

Listen for recurring complaints: we enter this twice, we answer this question every day, or we lose the details between people. Write down what happens before, during and after the task. Count actual examples over a representative period. This gives you a clearer starting point than choosing a tool and searching for something to do with it. The most useful opportunity may be a modest internal process rather than a public-facing assistant.

Compare frequency, effort and consequence

Consider three dimensions together. A frequent task offers more opportunities to help. A time-consuming task may release useful capacity. A consequential task requires more care because a wrong answer or action could be expensive. Do not collapse these into a single impressive score. A low-volume task with high consequences might deserve better documentation before AI. A frequent drafting task with easy human review may be a more practical first pilot.

Choose a bounded pilot

For example, a small team could test turning internal meeting notes into draft action lists. Keep the notes within an approved environment, require review and assign no tasks automatically. Compare the draft with what the team actually agreed. This is a proposed pilot, not evidence of a delivered result. Its limited scope helps reveal whether the output is useful before adding integrations or giving the system permission to act.

Set a baseline before the trial

Record how long the task takes today, how often it needs correction and where delays occur. Decide what an acceptable output must include. Use varied examples rather than only the cleanest ones. Keep the original process available while you learn. A pilot should be able to show that it helped, failed or needs revision; a trial designed only to produce a success story cannot guide the next investment.

Make the expansion decision explicit

At the end of a pilot, compare the result with the baseline and account for review time, operating cost and maintenance. Ask the people doing the work whether it actually makes their day easier. Expand only after identifying what changed and why. My preference is to make a small workflow useful enough that the team wants to keep it, then decide which adjacent step deserves attention.

Common questions