POV
AI investment For CEOs and CFOs

The companies that win with AI won't be the ones that spent the most. They'll be the ones who can tell, with evidence, what it returned.

Two teams have the same AI budget and the same usage slide, and opposite results. Here's the sum that tells them apart, and three questions to ask before your next budget meeting.

The short version

Don't ask how much you spend on AI. Ask what one finished piece of work costs now, with people and AI together: one customer problem solved, one feature delivered, one invoice processed. Ask what it was worth, and how sure you are of that. Then decide where the next dollar goes.

Skip the story, show me the three questions

Picture a budget meeting next month. Two team leads, Priya and Dan, walk in one after the other. Each runs a team that costs $100,000 a quarter. Each spent another $10,000 on AI tools. Each has a slide showing that almost everyone on the team uses AI every day, and each asks for more money next year.

You have to pick who gets it, and from those two slides you can't. They look the same.

They aren't the same. One of these teams is now paying more for every piece of work it finishes than it did before AI arrived. Neither slide shows it, and neither Priya nor Dan is hiding anything. Nobody has asked them for the number that would.

The sum that tells them apart

It takes a minute, and it needs one idea: a piece of work. That's one finished item of whatever a team produces. It could be a customer problem solved, a feature delivered, a deal closed or an invoice processed.

Before AI, both teams cost $100,000 a quarter and finished 100 pieces of work, so each piece cost $1,000. Then each team added $10,000 of AI.

Same AI spend, opposite results

A made-up example with round numbers, to show the sum.

Before AIPriya's teamDan's team
People cost$100,000$100,000$100,000
AI cost$0$10,000$10,000
Pieces of work finished100120100
Cost of each piece$1,000$9178% cheaper$1,10010% more expensive

Priya's team now finishes 120 pieces, so each one costs $917. Dan's team still finishes 100, so each one costs $1,100. That's 8% cheaper for one team and 10% more expensive for the other, from the same spend and the same slide.

If you run a team, you're already objecting, and you're right to. Pieces of work aren't all the same size. A team that gets funded on the count will start cutting its work into smaller pieces. And cheaper only matters if the extra 20 pieces were work the company wanted, done as well as before.

So count only work that someone accepted, whether that's a customer or another team. Check that the amount of work sent back to be redone didn't go up. Then ask what the extra work was worth, which is the second question further down.

Even with all that care, you now know something the slides couldn't tell you. Priya has earned a serious look at more money. Dan needs help before he gets any.

Why this meeting is happening everywhere

Companies spent about $33 billion on AI last year. This year it's $177 billion, more than five times as much. That's the count from Menlo Ventures, an investment firm that publishes it every year. I can't think of another line in a company budget that has grown like that without the CFO asking, in writing, what came back.

And what did come back? McKinsey, the consulting firm, asked people across many companies this summer. Eight out of ten said AI makes them faster at their own job. Only 37% said it had added anything to their company's operating profit, which is the money a business makes from its day-to-day work. That share was the same a year ago.

That's Priya and Dan's meeting, repeated in thousands of companies. Almost everyone feels faster, and most companies can't yet see it in the money.

We see it up close as well. At one of the larger companies we work with, keeping this record showed two things no usage slide could.

The first was about cost. AI tools run on different models, and some cost many times more than others. A large part of this company's AI bill was going to the most expensive ones, for everyday work that a cheaper one does just as well.

The second was about people. The teams getting the most done weren't the ones using AI the most. They used less of it and used it better, and once that was visible, other teams could learn how they worked. Nobody had hidden any of this. Nobody had been asked to look.

Who's getting results

It would be easy to conclude that AI isn't paying off. I don't think that's right any more. BCG, another consulting firm, reported this year that almost half of companies now get value from AI. That fits with McKinsey's number. The value is turning up in parts of the business before it turns up in the overall profit.

So some companies are pulling ahead, and I wanted to know if they'd simply spent more. They hadn't. The winners spend plenty, but so does everyone else. BCG found back in 2024 that the leading companies try about half as many AI projects as the rest. They also keep count.

Wharton, the business school, ran a survey with the research firm GBK this summer. It asked leaders how sure they are of what their company spends on AI in total, and then compared how each group was doing.

The ones who keep count are the ones who see results.

Share of leaders who say AI is paying off strongly for them.

Share of leaders who say AI is paying off strongly Among leaders who are less sure what their company spends on AI in total, 10 percent report strong results, against 56 percent of those who are very sure. Among teams that count only time and cost saved, 13 percent report strong results, against 35 percent of teams that also count growth. DO THEY KNOW WHAT THEY SPEND ON AI IN TOTAL? Less sure 10% Very sure 56% WHAT THE TEAM COUNTS Only time and cost saved 13% Also what the work brought in, like new sales 35%
Source: Wharton Human-AI Research and GBK Collective, "The Adaptive Advantage", October 2026. People rated their own results.
View as table
GroupSay AI is paying off strongly
Less sure what they spend on AI in total10%
Very sure what they spend on AI in total56%
Team counts only time and cost saved13%
Team also counts what the work brought in35%

Leaders who are very sure what they spend say AI is paying off strongly 56% of the time. For the rest, it's 10%. The lower pair matters too. Counting only the time and cost saved isn't enough, and the teams doing best also count what the work brought in.

A fair warning: this is a survey, and people are grading their own results. Wharton says itself that leaders who feel sure about their spending tend to feel good about AI in general. I'm also not telling anyone to spend less. But the pattern matches what the sum showed. The companies getting results can say where the money went and what came back.

The real mistake

The mistake isn't spending on AI. It's spending without keeping a record, because then next year's budget goes to whoever tells the best story.

Before the next budget meeting

Three questions to ask Priya and Dan

The sum gives you the first number, not the whole answer. Here's the full set I'd ask of any team, in any part of a company.

Three questions
  1. What does one piece of work cost now?
    People cost plus AI cost, divided by the work that was finished and accepted. Compare it with before AI.
  2. What was that work worth?
    Start with what the team sees, like faster delivery or fewer errors. Then look for what the company sees, like a sale or a renewal. Count only what you can trace back to the work.
  3. How sure are we?
    Is the number someone's guess, something we measured, or money that showed up in the accounts? Say which.

With those three answers the decision gets much easier. You put more money in, fix it, wait, or stop.

The method behind these questions is one we built at Garth, and it's our own. We call it Work Worth. It works out what a piece of work cost, what it was worth and how sure we can be, for any team in a company. I'll explain it properly in my next piece.

We started with engineering teams because every piece of their work leaves a record. Counting it fairly is still hard, and that's most of what we do.

Back to the budget meeting

Priya and Dan are still waiting, with the same slide. Ask each of them for one more: what a piece of work cost before AI, and what it costs now. One will be glad you asked. The other needs help, and you'll know it a year earlier than you would have.

Every company will keep spending on AI, and I think they should. The ones that pull ahead will be the ones that can answer that question team by team. They'll move their money toward what's working while everyone else is still arguing about whether AI works at all.

You can try it this month without buying anything. Pick one team. Take one quarter from before it had AI, and the most recent one. Add up the people and AI cost for each, and divide by the work finished.

If you'd like a second pair of eyes, bring me one team's numbers and we'll do the sum together. And if your numbers tell a different story from mine, I'd like to see them.

Next in this series

Work Worth explained: how to answer the three questions in any part of a company. After that, how to decide where the next dollar goes.

Sources
  1. Menlo Ventures, "2026: The State of Enterprise AI", 3 November 2026. Companies' spending on using AI reached $176.8B by September 2026, up from $33B in 2025. Menlo invests in AI companies.
  2. McKinsey, "The State of AI in 2026: On the Road to ROI". Survey of 1,719 people, May to June 2026.
  3. BCG, "AI Is Starting to Pay Off. Almost 50% of Companies Now Generate Value with It", 30 September 2026. Survey of 1,330 senior leaders.
  4. BCG, "Where's the Value in AI?", October 2024. Survey of 1,000 senior executives.
  5. Wharton Human-AI Research and GBK Collective, "The Adaptive Advantage: Who's Winning With AI and How", October 2026. About 1,400 people in US companies, surveyed July 2026.
Kumar Saurabh Johny
Founder and CEO, Darner. Maker of Garth.

Leaders talk to J at two points. Before they build, to decide which AI projects are worth doing and how soon they should show value. And once the money is being spent, to see what it's giving back and whether it's ready for the whole company.

Garth is the platform he and his team at Darner built to answer those questions with evidence. It shows what each piece of AI-assisted work cost and what it returned. It works for engineering teams today, and a version for CEOs and CFOs arrives in December 2026. Before Darner, J and MK Patil, who is Founder and CTO at Darner, built a company called Cathyos and sold it to Automation Anywhere.