Operational Productivity, Part 2: The Dashboard Arms Race

Somewhere in your company there is a screen with a number on it that nobody has acted on in a year. It refreshes every morning. Someone built it, someone maintains it, and it rolls up into a weekly deck, which rolls up into a monthly review, which rolls up into a quarterly board slide. The number moves. The work it claims to describe does not.

That screen is the whole story of forty years of enterprise software, and it is why the productivity paradox refuses to die.

Part 1 of this series, Two Kinds of Productivity, drew the line the rest of it stands on. Management productivity is the work of seeing and reporting what happens. Operational productivity is making the actual work faster, cheaper, and better. Here is the claim this part defends, stated flat: we did not fail to improve the work by accident. We turned the computer into a measurement machine instead of a doing machine, and then we rebuilt the org chart to serve the measurement.

Call it the dashboard arms race.

The computer showed up everywhere except in the work

In 1987 the economist Robert Solow wrote a line that has outlived almost everything else written about technology that decade: “You can see the computer age everywhere but in the productivity statistics.” Companies had spent the 1970s and 1980s pouring money into information technology. Mainframes, then minicomputers, then the PC on every desk. The spending was real and enormous. The measured productivity gains were, for years, almost invisible.

Erik Brynjolfsson gave the puzzle its name in a 1993 paper, “the productivity paradox,” and spent the following years trying to explain it. Part of the answer was measurement lag and mismeasurement. But part of it was simpler and more damning, and it is the part this series cares about. Much of what the computer got aimed at was not the work. It was the record of the work.

Think about what the first waves of business software actually did. Payroll. General ledger. Accounts receivable. Inventory records. Order entry. Every one of those is a system for writing down what happened, faster and more legibly than a clerk with a paper ledger could. That is genuinely useful. It is also, precisely, management productivity. We digitized the reporting layer with extraordinary energy and left the production layer, the place where value is actually created, running more or less the way it always had.

“

We did not compute the work. We computed the paperwork about the work, and then mistook a faster ledger for a faster factory.

”

A faster ledger is not nothing. But it has a ceiling, and the ceiling is low. You can make the record of the work perfect and the work itself stays exactly as slow, as wasteful, and as error-prone as it was. Solow saw the computers on every desk. He could not see them in the output, because most of them were not touching the output. They were touching the description of it.

A good idea that turned into an arms race

The measurement instinct got its intellectual charter in 1992, when Robert Kaplan and David Norton published the Balanced Scorecard in the Harvard Business Review, later expanded into their 1996 book. The original idea was sane and even humane. Companies were steering by financial numbers alone, which are lagging indicators, the equivalent of driving while staring only at the rear-view mirror. Kaplan and Norton argued you should also measure the things that produce those financials before they show up in the accounts: the customer perspective, the internal-process perspective, and learning and growth. Four perspectives, connected to strategy, so that what you measured actually mapped to what you were trying to become.

Read in its own moment, it was a corrective. The problem is what it became once it met cheap software.

The Balanced Scorecard turned into the template for the modern dashboard, and the modern dashboard turned into an arms race. Once measuring is cheap and building a new report is a ticket in a backlog rather than a month of a clerk’s life, the number of things you measure only ever goes up. Every executive wants their view. Every function wants its own scorecard. Every incident spawns a new metric so that “we are now tracking it” can be said in the next review. Nobody is ever fired for adding a chart.

So the reporting layer grows without limit while the work underneath it sits untouched.

Here is where it stops being harmless. When a real production problem shows up, a plant losing yield, a fulfillment process leaking days, a support queue drowning, the reflex of a measurement culture is to build a report about it. Hire a developer, wire up a dashboard, add a column to the weekly deck. Now everyone can watch the problem in high resolution. The problem is still there. You have spent real money and real engineering hours, and the output of that spending is a clearer picture of a thing that is exactly as broken as before. That is not zero productivity. Measured honestly, against the cost of the people and software consumed, it is negative.

We measure what is easy, not what matters

There is a second, quieter distortion, and it decides which parts of the business get all this attention. We instrument what is easy to instrument, not what moves the money.

Look at where the measurement machine points its full firepower: sales and marketing. Every click, every open, every scroll, every step of the funnel is tracked to the decimal. There is a good reason for it, and it is not that clicks matter more than everything else. It is that clicks are trivially easy to capture. The event is already digital. The instrumentation is a snippet of code. So the funnel gets measured to death, A/B tested, dashboarded from a dozen angles, because the data collects itself.

Now look at the factory floor, the warehouse, the clinical workflow, the claims process. Production yield. Scrap and rework. Machine downtime. Inventory sitting as dead capital. Cycle time on the step that actually gates throughput. These frequently hit the bottom line harder than another point of funnel conversion, and they are chronically under-measured, because measuring them is hard. The data is not already digital. It lives in a machine that does not talk, a step a human does by hand, a form filled out after the fact. Instrumenting it means sensors, integration, and messy physical reality, so it does not get done.

“

The metric that is easiest to collect is almost never the metric that most needs fixing. We measured the funnel because it was made of clicks and left the factory alone because it was made of atoms.

”

The result is a company that knows its cost-per-click to four decimal places and cannot tell you, without a special project, what its true scrap rate was last month. The parts of the business that were already legible got more legible. The parts that were opaque and expensive stayed opaque and expensive. Four decades of software widened that gap instead of closing it, because software naturally flows toward the data that is already sitting there and away from the work that would be hard to reach.

How the work got boring

The arms race has a human cost, and it is the least discussed part of the whole story.

Every dashboard needs to be fed. A number on a screen exists because someone, somewhere, is producing that number. When the reporting layer grows without limit, the demand for inputs grows with it, and that demand lands on people. Knowledge work quietly reorganized itself around the reporting machine. Large numbers of capable people spend their days moving information from one system into another, reconciling a figure that two databases disagree about, updating a status field, chasing a colleague for the input the Monday deck needs, cleaning a spreadsheet so it can become a slide.

Notice what the output of that work is. It is a number on someone else’s screen. Nothing in the world got made, moved, fixed, or improved. A person’s day was consumed producing a description of work for a review that produces a description of the review.

This is how forty years of the most powerful tools ever built made a great deal of work more boring rather than more meaningful. The promise of the computer was that it would take the drudgery and leave people the judgment. What often happened instead is that it created a new drudgery, the endless feeding and reconciling of systems of record, and handed it to humans, because the software could see the work but it could not do it. It could hold the number. It could not change the thing the number described. So a person had to sit in the gap, and the gap turned out to be enormous.

The paradox comes back, wearing new clothes

Here is the part that should make anyone building with AI right now sit up.

In early 2026, Fortune reported on CEO surveys in which leaders admitted that, so far, AI has shown little measured impact on productivity. Economists were quoted reaching straight back for Solow, invoking the same paradox by name, forty years later. Enormous investment, real excitement, and the aggregate numbers stubbornly flat. If that sentence sounds familiar, it should. It is 1987 with a new acronym.

“

The productivity paradox is not a fact about computers or about AI. It is a fact about where we point them. Aim the most powerful tool of the age at the boardroom, and you will see it everywhere except in the output, again.

”

The risk is not that AI is weak. The risk is that we do to it exactly what we did to the PC. The path of least resistance for any powerful new tool inside a large company is to feed the existing measurement machine. Point AI at the dashboard and it will write you better summaries of the work, prettier board narratives, faster reconciliations of the numbers nobody acts on. It will make the reporting layer even more magnificent while the work underneath stays exactly where it was. That is the dashboard arms race with a language model bolted on, and it will reproduce the paradox with uncanny precision. You will see AI everywhere except in the productivity statistics, and the economists will dust off Solow one more time.

There is a reason this trap is so easy to fall into. Every incentive inside a measurement culture rewards the visible artifact. A new AI-generated report is demoable in a meeting on Friday. A quiet improvement to a production process that nobody was measuring is invisible, precisely because nobody was measuring it. The arms race selects for what shows up on a screen, and AI is very, very good at producing things that show up on a screen.

The way out is not another dashboard

So the dashboard arms race is the name for what went wrong: a measurement culture, armed with cheap software, that grew the record of the work without limit and left the work itself alone, then staffed the gap with people whose only output was a number. It happened once with the PC. The productivity paradox is the scar it left, and the 2026 CEOs are watching it try to happen again.

Which raises the only question that matters. What would it look like to break the pattern instead of repeating it? If the failure was aiming intelligence at the measurement, the fix is aiming it at the work. Not a system that watches the process and reports on it, but one that sits at the point of production and acts, in real time, on the thing itself. That is the shift the next part of this series is about. Part 3, The Ideas Only Possible with AI and Agents, is where the second chance gets specific: what changes the moment intelligence can do the work and not just describe it.

The stakes are simple. We are standing at the exact spot we stood at in the 1980s, holding a more powerful tool than we have ever held, with every organizational reflex pulling us to point it at the boardroom one more time. The companies that win the next decade will be the ones that refuse. They will aim it at the work.

Sources

• Robert Solow’s 1987 remark and Erik Brynjolfsson’s 1993 coining of “the productivity paradox”: Productivity paradox.
• Robert Kaplan and David Norton, “The Balanced Scorecard,” Harvard Business Review, 1992, expanded in their 1996 book.
• Fortune, 2026, on CEO surveys showing little measured AI productivity impact and economists invoking Solow’s paradox: The AI productivity paradox.