In June, I wrote that vibe coding was dead and that ROI-driven analysis of AI was about to go from a nice-to-have to a necessity.
Here's what AI ROI is and why it's tricky to see in GDP numbers so far:
At its simplest, we need to understand two metrics: cost and value.
On the cost side, Ramp's AI Index tracks what more than 70,000 US businesses spend on AI. The median company spends $12.50 per employee a month. Against an average employee cost of about $8,500 a month, AI only has to make that person ~0.15% more productive to pay for itself. That's about three additional productive minutes a week.
So, for most companies, the token bill is close to a rounding error. But other costs are intangible: change management, re-training, and the rest of the corporate machinery.
Value is harder to pin down, so let's use productivity growth as a stand-in. At P&G, one person using AI produced product proposals as good as a two-person team without AI. MIT economists showed that automation pays when AI steps are grouped so a person checks the work once, not after every step.
So AI mostly speeds up production, but most of a corporate employee's week goes to coordination, meetings, alignment, process sign-offs, not to producing work. (And I have yet to see my Claude make meetings 10X faster!)
Which raises the question: what if humans are the bottleneck to AI ROI? AI can hand back those three minutes, but it can't decide what happens to them. If they flow into another meeting or another week waiting on legal, the return is zero.
That may also explain why the gains seem so uneven. AI may multiply output rather than add to it. If so, the best firms would pull further away from everyone else.
A 2026 experiment with 515 startups gives us a glimpse of that gap. Every firm got the same AI tools, training, and support. A random half also saw how other firms reorganized their work around AI. For the typical startup, revenue barely moved in either group. But almost all of the gains came from the top 10% of firms, and the treatment group pulled furthest ahead, enough to push its revenue to 1.9x the control group overall.
The tools were identical. What separated the leaders was how they rebuilt their work around them. Which brings us back to the process itself.
AI lowers the cost of doing things, but does nothing about whether they were worth doing in the first place.
In 1995, Bill Gates made the same point: automation applied to an efficient operation magnifies the efficiency, and applied to an inefficient one magnifies the inefficiency.
Elon Musk has a five-step process for this:
- Question every process and requirement
- Delete what you can
- Simplify what’s left
- Speed it up
- Only then automate.
Elon has said one of his biggest mistakes at Tesla was doing those steps backward and automating processes that later turned out to be unnecessary. Many firms that say they're “leveraging AI” are making the same mistake.
And this may be why the AI ROI debate talks past itself. It expects productivity numbers to move at the speed of AI, when they move at the speed of big companies adopting it. That wouldn't be the first time.
In 1987, Robert Solow said: "You can see the computer age everywhere but in the productivity statistics." Productivity grew about 1.5% a year from 1973 to 1995, then about 3.3% a year from 1995 to 2003, once work was reorganized around the machines.
So what separates the firms actually getting a return from everyone else? Where does the productivity gain disappear between the employee and the P&L? And how much of the problem comes down to the way the work itself is designed?
Our Deep Dive works through those questions and offers a practical guide to improving AI ROI, including what to delete, what to automate, which AI steps to chain together, where a person still needs to check the work, and when a cheaper model is good enough.
Read or listen to the full deep dive here: https://research.socialcapital.com/p/ai-roi
Good luck to all the players!
