Draw the line between a data professional and a data user in the org chart, and you will get it wrong. You’ll centralise the fire warden because she owns an extinguisher, and leave the fire exactly where you found it.
The instinctive test is a skills test: if your output is reusable across any business area — a pipeline, a data model, a warehouse — you’re a data professional. If your data work only makes sense inside one function, you’re a data user who happens to touch data. It’s a clean test. It also fails the moment you apply it to the population actually causing organisations trouble: the finance analyst who taught herself enough SQL to stop waiting for a report, the marketing manager running her own attribution model in a spreadsheet nobody else can open, the supply chain planner who built a demand-forecasting macro because the official one shipped eighteen months late. Their skills transfer. Move them to another function and most of what they know still works. By the reusability test, you’d have to call them professionals. That should worry you, because it means the test is measuring the wrong thing.
What it’s actually measuring is a service failure, and the data backs this up with uncomfortable precision. BCG’s third annual AI at Work survey — more than 10,600 leaders, managers and frontline employees across eleven countries, published June 2025 — found that when employees don’t have the AI tools they need, more than half say they will simply find alternatives and use them anyway. That’s not rebellion. That’s the market clearing around a gap nobody official is filling.
Sisense’s State of Analytics 2025 research, a survey of 536 data professionals published in May 2025, shows what happens on the other side of that gap. Seventy-six percent of organisations admit they’ve made business decisions without consulting the data available to them. For most, that’s not indifference — it’s exhaustion. For three-quarters of organisations, the average employee spends two to ten hours a week just searching for the right data, and up to half of a working day can disappear into “digital friction,” bouncing between as many as ten applications to finish one analytics task. Faced with that, of course the finance analyst builds her own spreadsheet. It’s not a career choice. It’s the fastest way to stop losing a day a week to a broken service.
None of this is new, which is the part that should really give a centralisation programme pause. McKinsey flagged the shape of this problem back in 2018, when it estimated that US demand for “analytics translators” — people fluent enough in a business function to know what to ask for, and fluent enough in data to get it — could reach two to four million. That’s a role McKinsey explicitly located between the data team and the business, because the businesses weren’t being served directly and needed someone standing in the gap. Eight years on, most organisations still haven’t built that bridge as a formal service. So the business built its own, function by function, one self-taught analyst at a time.
I’ve spent the better part of two decades building the unglamorous layer underneath these numbers — the pipelines, the definitions, the governance nobody notices until it’s missing — and the pattern is always the same. You don’t get a shadow data function because people in Finance or Marketing wanted to become data people. You get one because the central data organisation had no service they could call, and the deadline didn’t wait for the org chart to catch up.
Which is why “does the skill transfer” is the wrong question, and “who does this person’s work actually serve” is the right one. A data professional produces something other people’s work depends on — a dataset, a pipeline, a definition that outlives the question that prompted it. A data user produces an output their own decision depends on, and it typically dies with the decision. Here’s the sharper version of that test, the one that actually survives a reorganisation: strip away the business context and ask whether the capability still has a job. Remove Supply Chain, and the data engineer’s pipeline skills still have a home somewhere in the company. Remove Supply Chain, and the demand-planning analyst’s SQL has no home at all — it was never the point, it was scaffolding for a service that didn’t exist.
That reframes the centralisation decision entirely. Pulling people out of the business because they’ve become data-competent doesn’t fix anything if the reason they became data-competent was a gap you haven’t closed. You’ll get the org chart tidier and the business exactly as under-served as before — minus the one person who understood both the numbers and the function they came from. The people worth centralising are the ones whose value was never tied to the function in the first place. Everyone else is evidence of a service you still haven’t built.
You can’t fix a drift by promoting the people who were quietly compensating for it.
Sources: BCG, “AI at Work 2025: Momentum Builds, but Gaps Remain” (June 2025); Sisense, “State of Analytics 2025” (May 2025); McKinsey, “Analytics Translator: The New Must-Have Role” (February 2018).
Suggested category: Entropy Management (secondary: Data and Artificial Intelligence)
Mode: Argument Essay
Note on scope: The centralisation angle in paragraphs 11–13 reflects the practical situation you described (your organisation centralising data roles) — reworded to stay general rather than naming your employer or team. Happy to sharpen or soften that if you want it closer to, or further from, your own case.
Leave a Reply