AI mistakes mid-sized companies make in their first year

Mistake 1: starting with the tool, not the problem

A director sees a demo, buys a licence, and then looks for somewhere to use it. The result is a pilot nobody needs. Reverse the order. Identify a specific process with a clear cost, then ask what would improve it. Sometimes the answer is AI; sometimes it is a simpler automation or a process change.

• Write the problem in one paragraph before talking to suppliers.
• Define what better looks like in measurable terms.

Mistake 2: underestimating data and process quality

AI and automation work with the information you give them. If records are inconsistent, documents are scattered or the process differs by person, the technology will inherit the mess. Check data quality and agree the real process early, and treat clean-up as part of the project rather than an afterthought.

Do not wait for perfect data, though. Many firms improve their data progressively as a side effect of a well-chosen first project.

Mistake 3: no owner and no sponsor

Projects owned by everyone are owned by nobody. Name a business owner who understands the process and a director who sponsors the work and clears obstacles. Without them, decisions stall and adoption fades once the supplier leaves.

Give the owner protected time. Treating a project as something to squeeze in around the day job is a reliable way to lose momentum.

Mistake 4: skipping governance and security

Staff adopt tools informally, and confidential or personal data ends up in places nobody approved. Put a short AI policy in place early: approved tools, what data may be used where, human review of important outputs, and supplier checks. Where personal data is involved, consider a data protection impact assessment and check current ICO guidance.

Mistake 5: trying to do too much at once

Ambitious programmes across every department drain attention and make it hard to learn what works. Pick one or two processes, deliver them well and use the results to decide what follows. Early wins build credibility with staff and the board.

1. Choose one process with a clear owner.
2. Build a small version and test it with real work.
3. Measure against the baseline.
4. Decide: improve, extend or stop.

Mistake 6: forgetting people and measurement

• Staff are told after the fact, so they distrust or work around the new system.
• No baseline is taken, so nobody can show whether it helped.
• Training is a single session and then forgotten.
• Nobody maintains the automation when software or processes change.
• Exceptions are not planned for, so edge cases pile up on one person.

Frequently asked questions