Over the last few years I’ve helped build AI-powered and automation workflows inside two AI products, and I’ve watched many more businesses buy AI tools that nobody ends up using. The pattern is consistent: the tool gets chosen before the problem is defined.
Here are the ten questions I work through with clients before any money is spent.
The problem
1. What specific task are we trying to improve?
“Use AI in marketing” isn’t a task. “Draft first-pass replies to the 200 inbound enquiries we get each week” is. The narrower the task, the easier it is to succeed and to measure.
2. What does that task cost us today?
Hours, delays, errors, lost leads. If you can’t roughly quantify the current cost, you won’t be able to prove the AI made a difference.
3. Would simple automation solve it without AI?
A surprising number of “AI” problems are really workflow problems. A well-built automation in a tool like Zapier is cheaper, more predictable and easier to maintain. Use AI where judgement or language is genuinely required.
The data
4. Where does the information the AI needs actually live?
In a CRM? Spreadsheets? People’s inboxes? AI is only as useful as the context it can reach.
5. Are we comfortable with where that data goes?
Check where the vendor processes and stores data, whether it’s used for training, and how that fits your GDPR obligations and client contracts. This is not optional in the UK and EU.
The people
6. Who owns this after launch?
Every successful AI workflow I’ve seen has a named owner who monitors quality and improves it. Without one, it quietly degrades.
7. How will the team’s day actually change?
If the new tool adds a step rather than removing one, people will route around it. Design the workflow with the people who do the work.
8. What happens when the AI is wrong?
It will be, sometimes. Decide in advance where a human reviews output, and what the fallback is.
The economics
9. What is the full cost, not just the licence?
Include set-up, integration, training, usage-based fees that grow with volume, and the time someone spends maintaining it.
10. How will we know in 90 days if it worked?
Pick one or two measures — time saved, response speed, conversion, cost per lead — and agree the target before you start.
AI should amplify expertise, not replace thinking. These questions are the thinking.
If you can answer all ten clearly, you’re ready to choose a tool — and you’ll probably choose a better one. If several are hard to answer, that’s exactly what an AI readiness assessment is for.