Photo: Google Gemini · AI-generated
Today is Digital X in Cologne. The stage banner reads “Ready for AI”. It is the right question. It usually gets the wrong kind of answer.
Ask a mid-sized company whether it is ready for AI, and the answer is about tools, licences, or a pilot with a vendor. Those are the easy parts. The hard part sits one layer down, in the systems the AI would have to read from and write to.
We are not at Digital X this year. But the question follows us into almost every first conversation, so here is how we answer it.
An AI feature is only as good as what it can reach
Every useful AI feature does one of two things. It reads your data and tells you something, or it reads your data and does something. Both start with reading.
So the honest readiness check is not “which model?” It is three plainer questions:
Where does the data actually live? Can anything read it without a person in the middle? And who owns it when the answer is wrong?
If any of those three has a fuzzy answer, the AI project will spend most of its budget finding out.
Where the data actually lives
A garment manufacturer we work with had stock in one system and sales in another. Both were fine products. Neither could see the other. Every question about replenishment was answered by a person exporting two files and joining them by hand.
We built a supply-chain intelligence platform that connected the two and gave them one consolidated view. The interesting part is what the project was mostly made of: not intelligence, but plumbing. The consolidation was the work. The insight came almost for free afterwards.
That is the usual shape. Before you can ask a model anything about your business, someone has to make the data reachable in one place, with one meaning per field. That job does not appear on the stage. It is the job.
Whether anything can read it
The second question is about interfaces.
A software vendor came to us with a warehouse and production tool running on Windows CE, talking to Sage 100. The hardware was dying. The tool was still exactly what the warehouse needed. We rebuilt it as a native Android app and kept it talking to Sage 100 as before.
What made that possible was that Sage 100 had a usable interface. If your core system has no clean way in and out, an AI feature has nothing to hold on to. It either gets a copy of the data that is already stale, or it gets nothing.
There is a quieter version of the same problem. A water treatment company had field staff typing measurement reports into the CRM by hand. That manual step was the data pipeline. We replaced it with an app that syncs the reports directly, with logging and retry. Until then, any analysis on top of that CRM would have inherited every typo and every missed afternoon.
The order of work
This is the part a keynote skips, because it is not exciting: for most mid-sized companies, “ready for AI” means three unglamorous things, in this order.
Make the data reachable. Give the core systems an interface that something else can call. Decide who owns each number.
Do that, and the AI question becomes small and cheap to test. Skip it, and the AI project quietly turns into an integration project with a worse name and a bigger budget.
Where to start
You do not need a strategy deck to find out where you stand. You need someone to look at your systems and write down what is reachable, what is not, and what it would take.
If you already have an offer or a pilot proposal on the table, a second opinion before you sign is the cheapest week you will spend on it. If you have one concrete initiative and no written requirements yet, a requirements workshop turns it into a list of interfaces and a budget corridor you can defend.
Either way, ask your systems first. The model can wait a few weeks. Your ERP has been waiting for years.
Talk to us if you want a straight answer on where you stand.