Last year around this time, I sat through a savings review that had all the hallmarks of good analytics. Clean slides. Colour-coded categories. A year-on-year trend line moving in the right direction. The category lead walked the room through twelve months of numbers with the fluency of someone who'd rehearsed it — and probably had.
Then someone in the room — not senior, not trying to catch anyone out — asked a simple question. What was driving the movement in the third quarter?
The room went quiet. Someone reached for another slide. It didn't have the answer either. Nobody could explain the movement beyond describing the number itself again, slightly louder, as if repetition might pass for insight. The dashboard could tell you what had happened. Nobody in that room — including me, if I'm honest — could tell you why, or what was likely to happen next.
That's not a criticism of the people in the room. It's a pattern I've seen across a lot of procurement functions, and it deserves naming plainly: most procurement functions are analytics tourists. They visit the data. They don't live there.
The ceiling nobody names
Descriptive analytics is comfortable. It tells you what spend looked like last quarter, which suppliers took the largest share, where the variance sits against budget. It's real work, and it's not worthless. It's also, for most functions, where the analytical journey stops entirely.
Predictive analytics asks a harder question — what happens next, and how confident are we. Prescriptive analytics goes further still — given what's likely to happen, what should we do about it. Very few procurement functions operate at either level with any consistency. They report the past fluently and have almost nothing to say about the future beyond instinct dressed up as a forecast.
Nobody decided this deliberately. It happened because descriptive analytics is safe. It doesn't ask anyone to commit to a number they might be wrong about. It doesn't require sitting with uncertainty in front of a stakeholder. A trend line is defensible. A forecast is a target someone can hold you to.
Tourist, commuter, resident
I've started thinking about analytical maturity as three distinct relationships with the data, not a single line from bad to good.
The tourist visits. They receive a monthly pack, absorb the headlines, and move on until the next one arrives. The commuter has a regular routine with the data — they check the dashboard, they know where to look — but they're passing through, not building anything. The resident lives there. They interrogate the numbers, know what's missing, hold a working view of what's likely to happen, and use that view to make a call before anyone asks them to.
Most procurement functions I encounter are somewhere between tourist and commuter. Very few are resident.
Why this predates AI
None of this is new. Business intelligence tools have been sitting on procurement desktops for well over a decade, and the tourist pattern was just as visible then. It simply had lower stakes. A dashboard nobody interrogated properly produced a slide nobody interrogated properly. Mildly wasteful. Rarely dangerous.
AI changes the cost of the same gap. An AI-native tool doesn't wait to be interrogated — it produces a confident recommendation whether or not anyone in the room is equipped to challenge it. A tourist relationship with data, sitting underneath an agentic tool acting at speed and scale, isn't mildly wasteful any more. It's a function accepting decisions it has no ability to evaluate.
What a resident actually does
The shift from tourist to resident isn't a training course, and it isn't a certification. It's a set of working habits that show up long before any AI tool enters the conversation.
They interrogate the output, every time. A number arrives and the first instinct is "what would make this wrong," not "let's put this in the deck."
They're comfortable holding a range, not a false precision. "Between four and seven per cent, and here's what would move it" is a more useful answer than a single confident figure nobody can defend.
They know what data doesn't exist yet. A resident can tell you exactly which question they can't answer today and what it would take to answer it — which is precisely the muscle Part One and Part Two of this series were both circling.
They connect the number to a decision. Not "spend is up in this category" as a standalone fact, but "spend is up in this category, here's why, and here's what I'd do about it before the quarter closes."
The uncomfortable part
Becoming a resident is uncomfortable in a specific way. It means being willing to say "I think, with reasonable confidence, that X will happen" in a room where you might be wrong in front of people who will remember it. Descriptive analytics never asks that of anyone. That's precisely why so many functions have stayed there so long — not through incompetence, but because the alternative carries real professional risk.
The functions that make the shift build it into rhythm rather than treating it as an annual capability initiative. Weekly, not quarterly. Forecasts revisited and marked against what actually happened, honestly, in the room. Hypotheses tested rather than headlines repeated.
Before your organisation adds the next AI tool to the procurement stack, the harder question isn't whether the model works. It's whether anyone in the room actually lives in the data it's built on — or whether everyone's just visiting.
