Three serious studies measured the return on artificial intelligence over the same pair of years and arrived at 95%, 85%, and 80%. None of the three lied. This session shows why the numbers diverge by seventy-five points, and hands over the three questions that settle the next statistic to land on your desk.
Every number cited comes with who measured it, on whom it was measured, and who paid for it — including the ones that hold up this session's own argument.
This is not a debate between good research and bad research. All three are defensible, all three were published by serious institutions, and even so the distance between the first and the last is seventy-five percentage points. Understanding why is more useful than picking a side.
The test is the ledger: the money has to show up. It is the hardest yardstick of the three, and MIT does not sell artificial intelligence.
Not "there was no return" — "we could not measure it". This is the most honest yardstick of the three: it admits its own difficulty instead of forcing a number.
More than five hundred technical leaders at companies that have already deployed agents. Those who tried, failed, and gave up are not in the sample.
You do not need to know the studies to take them apart. The same three questions work on any statistic that lands on your desk after the session, including the ones you have already used.
Impact on the ledger and return self-reported by those who deployed are not the same test. They are different questions, not different answers.
Surveying satisfaction only among those who are still customers always gives a good result. Survivorship bias is not bad faith, it is how the sample was cut.
Interest is not fraud, but it is a fact about the fact. It does not invalidate the study; it invalidates using the study alone.
The figure that holds up the title — 15% versus 3% — goes through the same three questions in front of the room, with the funder's interest stated out loud.
A hundred and twenty minutes do not replace months of work. Saying so at the opening is part of the method, not modesty.
The session shows why seeing the cost is hard — a thousand times more consumption on an agentic task than on a chat, and thirty times of variation between two runs of the same task. Building that visibility is a project, not a slide.
Five times more chance of a return is still little chance. Anyone who leaves thinking they installed a dashboard and solved the problem understood the opposite of what was said.
It works for a group from a single company or for an open audience. The four-minute exercise, in which each person applies the three questions to a statistic they themselves have repeated, is what changes the conversation afterwards.
It fits into a board-level calendar window, without requiring half a day to be blocked.
No technical background required. What it requires is having to defend or challenge a number about AI in front of someone.
The webinar hands over a yardstick for reading numbers. The Executive Judgment Masterclass takes on the next decision: how far AI can go in each process.
Which AI metric did you institute this year, and which of your own values is it contradicting right now?
A hundred and twenty minutes, and the next statistic that reaches the table is read with a yardstick. Tell me whether it is for you or for your team and I will reply with the possible dates.
I want to attend this webinar