When someone asks me where their company should start with AI, my honest answer is almost never about AI.
It is about the gap between the systems they already own.
A small example
A professional services firm I worked with had spent the better part of a year evaluating AI tools. They wanted help. Specifically, they wanted to understand why their delivery margins had been slowly compressing for two years even though revenue was growing.
They had a CRM. They had a project management tool. They had a time tracking system. They had a financial system. Each of them had data that, in isolation, was fine.
But no one in the company could answer the question that mattered: “which clients are we losing money on, and why?”
Not because the data did not exist. It existed in four places. It just did not exist in any place at the same time.
When we mapped it out, every person on the leadership team had a theory. The CFO thought it was scope creep on a specific kind of engagement. The COO thought it was over-staffing on early project phases. The head of sales thought it was discounting. They were all partly right. They were also all partly wrong. And the company had been making decisions for two years based on the version of the story each of them happened to believe.
The fix was not an AI tool. It was a connected layer where the data from all four systems could be aggregated, governed, and queried — and then an AI sitting on top that could answer the actual question.
When we turned it on, the margin pattern became obvious within the first week of looking. Three clients accounted for nearly all of the compression. Two of them had become unprofitable. One of them they had been planning to expand.
They did not need AI. They needed visibility. The AI was just the interface that made visibility usable.
The architectural shift
There is a useful mental model here. AI is not a product you buy. AI is a layer in a stack. Underneath it sits the layer that actually matters — the layer that connects the systems where your business records what it does.
If that underlying layer is missing, no AI tool will help you. You will be asking the smartest assistant in the world to reason about your business based on whatever piece of your business it happens to be staring at. Sales asks ChatGPT about pipeline. ChatGPT can only see what sales pasted into it. The financial picture is invisible. The delivery picture is invisible. The advice is, at best, partially informed.
If that underlying layer is present, almost any AI tool starts to feel transformative. The same questions get materially better answers, because the system has access to the full picture.
The work, then, is not in choosing the AI. It is in building the layer that makes the AI worth choosing.
Where to start
If you are wondering how to approach this without committing to a year-long platform initiative, the honest answer is: do not start with the platform. Start with the question.
Pick the one decision your leadership team makes regularly that they currently have to guess at. The one where everyone has a different theory and no one has the data to settle it.
Now ask: what would have to be true for an AI to answer that question correctly?
Whatever shows up on the other side of that question — that is your first piece of intelligence architecture. Not the whole stack. Not the platform vision. Just enough connective tissue to make the next decision smarter.
Then do it again. And again. The architecture builds itself in service of the decisions, not the other way around.
That is how intelligence actually gets built. Not in a slide deck. In a sequence of “we should know this and we don’t” moments, each one turned into a place where, from now on, the answer is in the system.