AI investment has reached the trillions, but most companies still measure AI through productivity. Jeremy Barnett argues the bigger return may come from turning years of business decisions into proprietary intelligence.
AI investment has moved into the trillions. The world’s largest technology companies are committing extraordinary amounts of capital to data centers, chips, energy and infrastructure years ahead of the demand they expect those investments to serve. Recent reporting found that five technology companies alone have roughly $1.09 trillion in future lease payments, largely tied to data centers that haven’t yet come online.
We’re spending at infrastructure scale and still measuring much of the return at task scale. Hours saved. Research completed faster. Better output from the same team.
All useful, but while productivity tells us how efficiently AI completed a piece of work it says far less about whether anything became more valuable inside the company because the work happened.
That’s the piece I think we’re missing collectively at the top. Businesses make decisions every day that cost real money. They form a view, commit capital, act and eventually find out whether they were right. If AI can make that decision history usable the next time the company has to make a call, the economics start to look very different.
There’s precedent for looking beyond the first measurable return. Electrification changed manufacturing over decades as companies learned what distributed electric power allowed them to do differently with machinery, capital and production.
The computer era produced a similar lesson. Research into corporate IT investment found that much of its value appeared alongside changes in how organizations worked and used information. In both cases, the technology was visible before all of its economic consequences were.
AI may have its own version of that lag. We can already see what it saves. It takes longer to see what a company gets to keep.
Your Decision History Is an Asset
Every business has a decision history, built through the markets it entered, the prices it set, the products it launched, the companies it acquired and the capital it committed. Before any of those decisions were made, there was a view of what would happen next. Someone had an opinion about the customer, the market, the economics, the competition or the people involved. Eventually the company got an answer.
Companies preserve that answer remarkably well. The thinking that came before it is easier to lose.
Six months after a major decision, the revenue number is easy to find. The assumptions behind it may be buried in an old presentation, an investment memo or somebody’s inbox. Even when the people involved are still there, memory isn’t a database.
Successful decisions tend to get remembered more cleanly than they happened. Failed decisions acquire competing explanations. Assumptions everyone felt certain about at the time become surprisingly difficult to reconstruct. That gets expensive.
An acquisition makes the problem easy to see. Before a deal closes, there’s an educated expectation about the market, customers, management, growth, integration and where the upside is supposed to come from. A model captures some of it, but anyone who’s spent time around deals knows there are dozens of judgments behind the numbers.
How durable are the customers? How good is the management team? Where can the business expand? Which risks can we live with?
What has to be true for the investment case to work?
Then the deal closes, and reality starts answering those questions. Two years later, you know how the company performed. That still doesn’t tell you everything you learned.
The market thesis may have been right while the cross-sell opportunity everyone loved during diligence went nowhere. Customer concentration may have consumed weeks of debate and caused very little trouble. A talent issue that barely registered before close may have slowed growth for a year.
The deal may outperform because of an opportunity nobody identified going in, or underperform even though several of the original assumptions were right. If all the company carries forward is whether the deal hit plan, it loses most of the useful information.
Take a company that’s completed 40 acquisitions. It should absolutely know a thing or two more than any company completing its first one. It should know where its own forecasts tend to be wrong. It should recognize the risks the teams routinely overestimate and the problems they have a habit of missing.
It should have a better view of which management characteristics show up later in performance and which sources of upside sound great in a boardroom but rarely develop the way everyone expected.
Because at the end of the day, that knowledge wasn’t free. The company bought it through capital, time, opportunities, and successful bets.
Every Function Is Paying to Learn
You see the same problem everywhere else in the business. Marketing teams spend money against assumptions about audiences, messages and behavior, then find out who actually buys. The audience that generates the most engagement may turn out to produce weaker customers than a group that barely registered early in the campaign.
Product teams build around one expected use case and discover customers value something entirely different. Sales teams become convinced one type of account represents the best growth opportunity, then watch another segment stay longer and expand faster.
A company is constantly correcting its own understanding of the world. Historically, keeping those corrections has been difficult. The information is scattered across financial systems, CRM records, presentations, research, campaign platforms, meeting notes and people’s heads.
Even when the data exists, connecting an assumption made three years ago to the outcome that eventually followed can require more effort than anyone has time to spend. AI gives companies a credible way to change that.
An investment team looking at its next acquisition can use current financial and market information while also seeing where similar assumptions failed across previous deals. A company entering its sixteenth market can bring forward what happened in the first fifteen, including where demand differed from the forecast or which early indicators turned out to be noise.
A marketing team can go beyond knowing which campaign produced the strongest return and understand whether the thinking behind those campaigns kept proving right or wrong.
The Return Compounds
That’s where the economics get interesting. Models will improve. Analysis will get cheaper. And competitors will eventually have access to many of the same tools and public information.
They can’t buy your decision history overnight. Ten years of decisions contain things no outside model knows, including where your company consistently gets the market right and where its own judgment has a habit of breaking down.
That experience can keep paying after the original decision is over. Saving an hour on an analysis creates value once. Learning that your team consistently overweights a certain variable in market-entry decisions can change every market you evaluate afterward.
Discovering that the customers everyone overlooks in month one become your highest-value accounts two years later can change how you spend acquisition dollars for years.
That’s the kind of return I’d expect to see emerge from this investment cycle. Five years from now, I’d expect AI to have changed the quality of judgment inside the company. A forecast will be faster to produce and happen in real time, with information updating second by second.
The thousandth decision should have an advantage over the first because the other 999 happened. Businesses have spent decades paying for that knowledge, then losing pieces of it before the next big call gets made.
AI gives us a way to keep more of what experience already cost us to learn and turn it into an advantage competitors can’t simply buy.
