The AI readiness checklist

Most AI projects that fail were the wrong pick, not the wrong technology. A process that isn't ready stalls however well it's built. Run your candidate through the questions below before you spend anything, and count your yeses as you go. There's a quick way to read the total at the end.

The process itself

• It's repetitive and rules-based. The same task, done often, mostly the same way every time. Lots of volume and little variation is where automation earns its keep.
• You can describe it step by step. If you can write down how it works, it can probably be automated. If it lives in one person's head as "it depends", writing it down is the first job.
• It's stable. Nobody's planning to rebuild the steps next quarter. Automating a process that's about to change is money spent twice.
• The pain is real and countable. You can point to the hours it eats or the errors it causes. "It'd be nice" isn't a business case.

Your data

• The information it needs already exists digitally. It sits in a system, a spreadsheet or an inbox, somewhere software can reach. If the key input is a shoebox of paper, digitising it comes first.
• It's reasonably consistent. The same field means the same thing each time. It doesn't need to be perfect. It needs to be findable.
• The tools have a way in. The software offers an API (a way for other programs to read and write its data) or at least an export, so systems pass data along instead of a person re-keying it.

Your people

• Someone owns it. One person understands the process and will help shape the automation and sign it off. Builds without an internal owner drift.
• The team is on side. The people doing the work today see it as their afternoon back, not their job under threat. That decides whether it gets used more than any feature does.
• There's an appetite to learn. A well-built system still needs people who are comfortable using it. If AI is new to the team, a little training up front pays for itself.

Risk & guardrails

• A mistake is survivable or catchable. Either an error is low-stakes, or a person reviews the output before it counts. AI shouldn't be making high-stakes calls you can't undo with nobody watching.
• You know where the human stays. You can name the step a person keeps: the judgement call, the approval or the final send. Good automation takes the grind and leaves that step alone.
• Sensitive data stays controlled. If the process touches confidential or personal information, you're willing to run it on systems you own instead of pasting it into a public tool.

Reading your score

Count the boxes you could honestly tick.

• 11 to 13: This process is ready. It's the kind we'd happily scope: pick it, build it properly and measure the hours it hands back.
• 7 to 10: Close. There's usually one fixable gap, such as messy data, a process nobody has written down, or no owner. Sort that and it's ready.
• Under 7: Not yet, and that's worth knowing before you spend. Either the process needs to settle down first, or there's a better candidate elsewhere in the business.

Most businesses have several processes in the top band without realising it, plus one or two they're keen to automate that sit at the bottom. Sorting which is which, in priority order, is what an AI audit does. If you'd rather talk it through, the form below is there for that.

In practice, when more than two answers come back as a no, we say so and suggest fixing those first instead of starting a build. Automating a process nobody has agreed on only makes the disagreement happen faster. That conversation costs nothing and saves a fortnight.