Is your data ready for AI? A practical check before you automate

Most AI projects that stall in smaller companies do not fail because the model was wrong. They fail because the data underneath was a mess. The technology gets blamed, but the real problem was sitting in five spreadsheets, an inbox, and someone’s head the whole time.

This is the question people skip. You have picked a process to automate, you are excited, and you want to start building. Before you do, stop and ask one unglamorous thing: is the data this process runs on actually good enough to automate? A few minutes of honesty here saves months of frustration later.

What “ready” actually means

Data readiness is not about having perfect, pristine data. Nobody has that, and waiting for it is just another way of never starting. Ready means good enough for this specific task. The bar is lower than people fear, but it is a real bar, and it is worth checking deliberately rather than discovering halfway through a build.

Run your process through four practical questions.

Is the input structured and consistent?

Does the work start from data in a predictable shape, or does every case look different? An order that always arrives as a PDF with the same fields is workable. An order that comes sometimes as an email, sometimes as a phone note, sometimes as a scribbled attachment, is much harder. You do not need rigid uniformity, but the more consistent the input, the less the automation has to guess, and guessing is where AI projects quietly go wrong.

Does it live in one accessible place?

This is where most SMEs stumble. The data exists, but it is scattered. Order information lives partly in email, partly in an Excel sheet, partly in the ERP, and the full picture only exists when a person stitches it together in their head. If a human has to open five systems to answer one question, an automation will struggle with the same fragmentation. The data does not have to be in one perfect database, but it has to be reachable without a treasure hunt.

Is it complete and current?

Half-filled fields and stale records sink automations fast. Supplier information that nobody has updated in two years, customer records missing half their phone numbers, product data that contradicts itself between systems. A human compensates for these gaps automatically, often without noticing. An automation takes the data at face value and produces confident nonsense. Before you automate, look honestly at whether the underlying records are complete enough and current enough to trust.

Does someone own it?

Data with no owner drifts. If no single person is responsible for keeping a dataset clean and current, it will degrade, and any automation built on it degrades with it. This does not need a formal role. It needs one named person who notices when something is off and fixes it. Ownership is the difference between a clean-up that lasts and one you repeat every six months.

You do not need perfect data, just a clean-up

Here is the reassuring part. When a readiness check turns up problems, the fix is usually small and targeted, not a giant data project. Standardising how orders come in, consolidating two overlapping spreadsheets, filling the one field that actually matters for the task. A focused clean-up of a few days often unlocks a quick win that would have been impossible the week before. The point of the check is not to find reasons to stop. It is to spend a little effort in the right place so the automation works the first time.

I have seen the opposite too often: a promising automation built on shaky data, shipped, and quietly abandoned within a month because nobody trusted its output. The data check is what prevents that. It is cheaper to discover a gap on paper than to discover it in production.

Where this gets assessed honestly

Judging whether your data is ready for a specific automation is exactly the kind of question the AI-waardescan answers per opportunity. For each process we consider, we look not just at the potential time saving but at whether the data underneath can actually support it, and what a realistic clean-up would involve. That honesty is the point. It is better to know up front that an opportunity needs a week of data tidying than to find out after you have paid to build on sand. Start with the data, and the automation has somewhere solid to stand.