If you give a VP a dashboard, they will ask why it isn’t real-time. When you make it real-time, they will want it by region. When you give them the regions, they will want it by store, then by week, then by the segment someone named in the meeting. When you build all of it, they will say it has gotten too busy to read, and ask you to just tell them what is going on. When you give them the one number that matters, they will ask what is driving it. And if they want to know what is driving it, they are going to want a dashboard.
Every one of those requests was reasonable, which is what took me the longest to see. I was the one building them, by hand, every time, because the numbers underneath lived in a dozen systems that had never been introduced to each other, and there was nowhere to ask the question just once. I wasn’t analyzing the business. I was reassembling it, from parts, and then reassembling it again the next week.
I think about that job every time someone tells me AI is about to transform their company, because whatever they buy will be pointed at the same data I spent years failing to line up.
I went into data work because I wanted a technical skill I could carry across industries, one that would let me work on almost any problem as long as it had data underneath: how people behave, how a process breaks, where the money leaks out. In the mid-2010s, "big data" was the phrase and data science was the job everyone suddenly wanted. I majored in analytics at William & Mary, finished the degree in three years, and came out in 2018 certain the hard part would be the questions. My first job out of school, at Deloitte, was supposed to be answering them. Instead I spent most of my time compiling and cleaning the data before I could ask it anything. The analysis I had trained for was the last thing on the list, and most days I never reached it.
The numbers a company needs to answer a simple question are almost never in one place. At one consumer business, finance, operations, and marketing each kept their own version of the customer, and none of the three matched. Each team had built its own for its own purpose, and at the time the overlap did not look worth the trouble. It was faster to keep your own copy than to reconcile it with everyone else’s. In a 2024 survey of 1,050 IT leaders, the software company MuleSoft found the average organization running more than 1,000 separate applications, with fewer than a third of them connected. The rest sit in their own corners, and much of what matters never reaches a system at all. It lives in a spreadsheet on somebody’s laptop, which is to say it lives nowhere anyone else can get to. That is why there was no place to ask the question once. There was no once. There were a thousand elsewheres.
The strange part is that the fix was never in doubt. Pool the data in one place, put a real architecture under it, govern it so it stays clean, and teams can finally use it. For years at KPMG I ran workshops on exactly this for the executives of big companies. It was the standard answer, the thing my professors taught and every consulting deck sold. Gartner warned as far back as 2014 that a data lake without governance just turns into a data swamp, and mostly they were right. Everyone knew the answer, and knowing was never the gap.
The data is fragmented because the company is. Each function bought its own system to serve its own targets, and the walls between the systems are the walls between the departments, which are older and load-bearing. You cannot merge two databases whose owners are measured on different numbers. The merge is a negotiation, and the negotiation is the conversation no one wants to have. The shape of the data is the shape of the org chart.
So companies did the thing companies do with a problem that crosses every department. They made it a role. They hired a Chief Data Officer, stood up a data team, gave it a mandate to bring order to all of it, and then sat the new office off to one side and made it earn its way in. The Chief Data Officer now has the shortest tenure of anyone in the C-suite, a little over two years against a chief executive’s seven, and the story is always the same: hired to make a technical change, handed a cultural one, given no authority to force it. The fix for fragmentation became another fragment. A company that had genuinely made data everyone’s job would not need to appoint one person to go door to door asking for it.
The officer never did the reconciling. That fell to whoever was junior enough to be handed it, and for a while that was me. I would be sent to sit between the teams and translate, to get four departments to agree on a single number long enough to make one decision. It is real work, and it does not scale, because a human being is standing in for a bridge that was never built.
I had seen the same failure long before anyone had a model to blame for it, with plain analytics, in a federal agency and in ordinary companies alike, and the pattern never cared which.
Which is why I read this moment the way I do. AI does not route around fragmentation. It runs into it harder than anything before, because a model is only as good as its reach, and its reach stops at the first wall. Point one at a company whose data is scattered across a thousand systems and it will do what I did with the dashboards, faster and at greater expense. The wall was already there. AI is just the first thing expensive enough to make anyone look at it.
The convenient story right now is that the companies sitting on the most data will win by default. But a pile of data you cannot query is not an asset. It is a storage bill. You cannot see the value of what you have until you fix the infrastructure under it, and you cannot fix the infrastructure without fixing the organization that fragmented it, which is slow and political and has no launch date, so everyone skips it, and it is the part that decides. When every company has the same model, and most of them are sitting on the same unusable pile of data, the edge is whoever spent the unglamorous years building an organization that can actually use it. I spent my twenties making dashboards that were apologies for a foundation no one had poured. The tools have grown far more powerful since. The foundation is still the job.