Ask a founder what AI has done for the business and you will hear a version of the same answer. It made us faster. It saved the team hours. We are more efficient now. The tools are impressive, the demos are convincing, and the feeling of speed is real. So the conclusion feels obvious. If we are faster, we must be saving money.
Here is the harder question. If AI is saving you money, show me the line on your profit and loss statement where a number went down. Not a slide about hours saved. Not a survey where the team says it feels more productive. An actual line item, this quarter against last, that is smaller because of the tool. Most founders cannot point to one. That gap, between the efficiency you feel and the efficiency you can bank, is the whole story.
The research is blunt about it. A widely cited MIT study of enterprise AI in 2025 found that ninety-five percent of corporate generative AI pilots delivered no measurable return. Not smaller returns. No measurable return at all. These were not toy projects run by amateurs. They were funded, staffed initiatives inside real companies. Ninety-five out of a hundred produced a feeling of progress and nothing you could take to the bank.
The efficiency you feel is not the efficiency you can spend
Start with the most counterintuitive finding, because it explains the rest. In 2025 the research group METR ran a controlled trial with experienced software developers working on real tasks in code they knew well. Some were allowed to use AI tools, some were not. Then they measured. The developers using AI were nineteen percent slower. Slower. And here is the part that should stop you as an owner. Afterward, those same developers estimated that AI had made them roughly twenty percent faster. They were off by about forty points on their own work, and they were off in the direction of optimism.
Sit with that, because it is the mechanism behind ninety-five percent of pilots going nowhere. The people using the tool are not lying to you. They genuinely feel faster. The output looks polished, it arrives quickly, and the friction of the blank page is gone. But the felt sense of speed and the measured reality of throughput came apart. If your read on AI's value comes from asking your team whether it helps, you are measuring the feeling. The feeling is unreliable by about forty percentage points.
The cost does not disappear. It moves.
So where does the money go, if not onto the P&L. It moves to a place most owners are not watching, which is the cost of checking the work.
Harvard Business Review reported on a pattern researchers named workslop. AI-generated output that looks finished but lacks the accuracy, context, or judgment to actually be useful. In a study run by BetterUp and Stanford, about forty percent of full-time desk workers said they had received workslop in the previous month. Each instance took nearly two hours to sort out. The estimated cost came to a hundred and eighty-six dollars per employee per month. For a ten-thousand-person company, that is roughly nine million dollars a year in productivity quietly eaten by work that had to be redone.
That is the part the demo never shows you. AI did not remove the labor. It moved it. The work shifted from creation, which felt slow and effortful, to verification, which feels like management and rarely gets counted as a cost. The junior analyst produces a report in ten minutes instead of two hours. The senior person now spends an hour finding the three numbers that are confidently wrong. On paper you saved time. In the building you traded an hour of cheap work for an hour of expensive work and called it progress.
AI rarely cuts a cost. It relocates one. Usually from a place you were measuring to a place you are not.
Why the pilots keep failing
This is why abandonment is rising, not falling, as the tools get better. S&P Global Market Intelligence found that forty-two percent of companies scrapped most of their AI initiatives in 2025, up from seventeen percent the year before. The average organization killed forty-six percent of its AI proof-of-concepts before they ever reached production. The models improved and the failure rate went up. That should tell you the problem was never the model.
The problem is that most companies bought AI as a feature and expected it to behave like a system. A feature sits on top of whatever you already do. If the process underneath is undocumented, inconsistent, and held together by a few people's memory, AI does not repair any of that. It runs faster on top of it and produces more of it. You do not get an advantage. You get a quicker version of the same mess, plus a verification bill nobody budgeted for.
Founders feel this as a paradox. We bought the tools, the team uses them every day, and nothing changed in the numbers. There is no paradox. The tool worked exactly as designed. It amplified the process underneath it. The process underneath it was never built to be amplified.
A test you can run this week
Skip the survey. Do not ask anyone how AI is going. Do three things instead.
First, open your profit and loss statement and find one line, any line, that is measurably smaller this quarter than it was before you adopted the tool, and that you can honestly attribute to it. Payroll. Contractor spend. The cost of producing a specific deliverable. If you can name the number, you have real savings. If the best you can do is that the team feels faster, you bought a feeling, not a system.
Second, pick one thing AI now produces in your business, a report, a draft, a client response, and trace what happens after it is generated. Who checks it. How long that takes. What breaks when they skip the check. That is your verification cost. It is real whether or not it shows up anywhere in your books, and it is the number that decides whether the tool actually paid for itself.
Third, ask whether the process you automated was written down before you automated it. If it lived in someone's head, you did not build an advantage. You made an undocumented process faster and harder to see. The right order is the reverse. Document the process, make it consistent, then let AI run it. Automation on top of a clean system compounds. Automation on top of a mess just ships the mess faster.
Where the tool finally earns its keep
Here is the part that separates a tool from an advantage. AI does not create value on its own. It creates value when it sits inside a business already run as one connected system, where how you are taxed, how capital moves, the partnerships you build, and the systems you automate are decided together instead of in four separate rooms.
An automation that speeds up billing is worth little if the entity underneath it is overpaying tax on every dollar it collects. A faster sales process is worth less than it looks if the capital to fund the growth is not already in place. AI is one lever. Run it alone and you get a quicker version of whatever you had, good or bad. Run it as part of the whole, on top of processes that are documented and inside a structure built to keep what it earns, and the same tool finally shows up where it was supposed to. On the line that got smaller.
So make the shift. Stop asking whether AI is helping. Start asking it to prove itself the way you would make any other investment prove itself, on the statement, in a number, this quarter. The tools that survive that question are worth keeping. The ones that cannot are costing you more than the subscription. They are costing you the hour on the other side of the work, and the belief that something changed when it did not.
If you are interested in exploring an engagement, the starting point is a 30-minute private call. There is no pitch and no pressure, it is a conversation to find out whether the work makes sense. You can book directly at calendly.com/chelsea-eba/30min.
About the author
Chelsea Michelle is the founder of Elevated Business Advisors, a private advisory practice for founders, investors, and family offices. She architects tax, capital, partnerships, and AI and systems as one integrated system for a deliberately small roster of clients, by application, across Florida and nationally. She also hosts The Power of the Pivot podcast.