Five questions to ask before you start an AI project

Almost every organisation we speak to has an AI idea on the table. Some are ambitious, some are modest, and most are genuinely promising. Yet many AI initiatives never make it past an impressive demo. The reason is rarely the technology. It's usually that a few basic questions weren't asked early enough.

Before you commit budget and people, it's worth working through these five.

1. What problem are we actually solving?

Start with the problem, not the technology. "We should use AI" isn't a goal. "Our support team spends hours each day answering the same questions" is. A clearly described problem tells you who benefits, what good looks like, and whether AI is even the right tool. Sometimes a simpler change to a process or a product does the job better.

2. How will we know it worked?

Agree how you'll measure success before you build anything. That might be time saved, faster response times, fewer errors, higher conversion or happier customers. Pick one or two measures you can track from day one, and capture a baseline now so you have something to compare against. Without this, even a successful project is hard to defend when budgets are reviewed.

3. Do we have the data, and the right to use it?

AI is only as good as the data behind it. Check early whether the data you need exists, whether it's accurate and complete enough, and whether you can access it in a usable form. Just as important: check that you're allowed to use it for this purpose. If personal data is involved, data protection rules such as UK GDPR apply, and it's far easier to design for them from the start than to retrofit them later.

4. What happens when it gets something wrong?

Every AI system makes mistakes. The question is what those mistakes cost and how you'll catch them. A wrong product recommendation is a minor annoyance; a wrong answer in a regulated setting is a serious problem. Decide where a person needs to stay in the loop, how users can flag issues, and who is accountable for the system's behaviour. Good guardrails are what let you move quickly with confidence.

5. Who will own it after launch?

An AI feature isn't finished when it goes live. Models need monitoring, data changes, costs need watching, and users will ask for improvements. Make sure there's a clear owner and a team with the skills to look after it, whether that's your own people, a partner, or a mix of both. Projects without an owner tend to quietly decay.

Start small, prove value, then scale

If you can answer these five questions with confidence, you're in a strong position. Our advice is usually the same: pick one well-defined use case, run a focused pilot with clear measures, and only scale once you've seen real results. It's a less dramatic route than a big-bang transformation, but it's the one that tends to deliver.

If you'd like a second opinion on an AI idea, or help turning one into a working pilot, our AI Adoption and AI Software Development pages explain how we can help.