1. Start from a process, not a technology
The right question is not "which AI should we use?" but "which task costs us the most time?". List the processes where the team retypes data, looks for information, sorts requests or checks documents. They are the natural candidates.
2. Measure how it works today
For each process, note how many times it is repeated in a week, how many minutes it takes and how many errors it produces. Reasonable estimates are enough: they are for comparing opportunities, not for drawing up accounts.
- Volume: how many tasks a week.
- Time: minutes per task, including checks and corrections.
- Impact: what happens when something goes wrong.
You can use the time calculator to do the maths in a minute.
3. Assess value and feasibility together
A high-value process with scattered, non-digital data may need some groundwork first. A medium-value process with data already in order may be the best starting point. Rank the opportunities on both dimensions.
4. Decide who checks what
A good AI project establishes from the start where a person steps in: approvals, exceptions, important operations. That way AI speeds up the work without taking control away from the people responsible for it.
5. Start small, measure, then extend
A pilot project on a limited scope lets you verify the results on real data. A first pilot typically takes 1 to 2 months; the exact timings depend on the process and the integrations.
The most common mistakes
- Buying a tool before understanding the problem.
- Choosing the most complex process as the first project.
- Not involving the people who run the process every day.
- Not agreeing beforehand how to measure the result.
If you like, we can go through this together: it is our AI consulting, which typically takes 2 to 3 weeks and ends with a written proposal.