Anna Totterdell
Projects Director
Every vendor pitch starts the same way. "Your data is your most valuable asset." Then they show you a demo built on a perfectly clean dataset that looks nothing like yours.
Here are three tests you can run this week - without buying anything or hiring anyone - to find out whether your data is actually ready for AI. If you fail any of them, you do not have an AI problem. You have a data problem. And fixing that is where the real return starts.
Sign 1 - You cannot produce a single customer list
Ask your team to produce one accurate, deduplicated list of all your customers. Names, addresses, contact details, account status. One list. From your existing systems. Without anyone manually cleaning it.
If the CRM says 4,200 customers and the ERP says 3,800 and the finance system says 4,600 - you have a data and systems integration problem that no AI tool can solve. AI needs a single source of truth. If three systems give three different answers to the same question, any model trained on that data will produce confident, detailed, wrong outputs.
The test is not whether you have customer data. It is whether your systems agree on what that data is.
Sign 2 - Your reports require manual assembly
Look at how your monthly board pack gets built. If someone is exporting CSVs from three platforms, pasting them into a spreadsheet, reconciling the numbers, formatting the output, and emailing it around - your data is not structured. It is scattered.
AI does not read spreadsheets that someone built by hand. It reads structured, connected, consistently formatted data that flows automatically from source systems - the plumbing that business automation is supposed to put in. If a human has to assemble your reports before anyone can read them, your data pipeline is a person - and that person is a bottleneck, a single point of failure, and a cost you have never measured.
Count the hours. Most mid-market businesses spend between 20 and 60 hours per month on manual report assembly. That is your data readiness gap measured in time.
Sign 3 - The same field means different things in different systems
Pick any entity - customers, products, orders - and check how it is categorised across your platforms. Is a "product" in the ERP the same as a "product" on the website? Does "active customer" mean the same thing in the CRM as it does in finance? Are your date formats consistent? Your naming conventions?
If your teams have to translate between systems - if they know that "Category A" in one platform maps to "Type 1" in another, but only for certain product lines, and only after a certain date - your data is not integrated. It is manually interpreted. And manual interpretation does not scale, does not automate, and does not support AI.
What to do next
If you passed all three, you are in better shape than most. Start scoping AI use cases with confidence.
If you failed one or more, do not panic - but do not buy an AI platform either. The first project is not AI enablement. It is data. Map where the inconsistencies live, agree on standards, connect the systems, and build the foundation that makes everything else possible.
The businesses that get value from AI are not the ones with the best tools. They are the ones with the cleanest data. That work is less exciting than an AI demo - but it is where the return actually lives.
