I was in the room when the CRO pulled up the utilization dashboard.
Mid-market software company. Twelve months into an AI deployment. The Customer Advisory Board had just wrapped: a mix of RevOps leaders, a handful of AEs, one SE who looked like he hadn't slept since Q3. The CRO had the slide on screen before anyone sat down.
Fourteen percent.
Fourteen percent of licensed users had engaged with AI in a meaningful way in the prior thirty days. The other eighty-six percent had a tile on their desktop and a line item on the P&L.
Nobody said anything for a moment.
Then the RevOps lead said it. The one who had spent eight months building the enablement program. The one whose bonus had been cut the prior quarter.
"The support agents are worse. The ones running the AI-assisted tickets are closing at about a twenty-five percent error rate on tier-one issues. The ones doing it manually are at eleven."
The CRO nodded slowly. Put the deck down. And said: "We need to accelerate the roadmap."
More tools. More licenses. More acceleration.
The bonus stayed cut.
The macro picture is not an anomaly. It is the model.
A March 2026 survey of 866 U.S. business leaders by ResumeBuilder.com found that 54% of companies have reduced, or plan to reduce, employee compensation to free up capital for AI spending this year. Not just headcount. Bonuses, equity, raises, benefits, and base pay, all being cut simultaneously, across industries.
The companies doing the cutting are not apologetic about the trade-off. 88% of those surveyed said the weak job market makes it easier to reduce compensation without losing talent.
Meta eliminated 8,000 positions in May 2026. Microsoft offered voluntary buyouts to approximately 8,750 employees the same month. And TTEC, one of the largest customer experience outsourcers in North America, suspended its 401(k) employer match for approximately 16,000 U.S. employees through year-end 2026, citing the need to invest in AI tools and training. The memo from the chief people officer did not reach for euphemism. The match was suspended to fund AI. She said it plainly.
For every TTEC that said it plainly, there are a hundred companies doing it quietly: through bonus compression, merit freezes, and RSU reductions that never make the memo.
TTEC is not a cautionary tale. It is the version that got documented. Deloitte and Zoom cut parental leave and other benefits in the same cycle. The CFO is not asking whether AI is ready. The CFO is asking which line items can be reduced to fund the AI budget. The answer, in boardroom after boardroom, is the same: the people.
The Most Expensive Mistake in Enterprise Software History Is in Progress
In 2025, MIT Media Lab's Project NANDA published The GenAI Divide: State of AI in Business 2025. The finding that should be pinned to every CAB deck and every budget approval:
95% of companies deploying generative AI are generating activity. 5% are generating value.
5%.
The researchers identified four traits that separated the 5% from the 95%:
They started with a specific, high-value problem, not a deployment mandate.
They kept domain experts embedded in the AI development process, not siloed from it.
They measured outcomes against business results, not adoption metrics.
They iterated on failure. They didn't hide it in a utilization dashboard.
A Gartner survey of 350 executives corroborates the pattern: 80% of companies piloting AI reduced headcount, but high ROI was more common in firms using AI to enhance worker productivity than in those using it to replace staff.
Every trait that separates the 5% from the 95% requires human judgment. Specific problem identification. Domain expertise. Outcome accountability. Institutional learning. The 5% didn't win because they bought better tools. They won because they kept the people who knew how to use them.
Here is the pattern, plainly stated.
The CFO is cutting the people who know which problems are worth solving. The CFO is funding tools to replace them. The tools don't know which problems are worth solving.
The tools are not the team. The AI is the intern: fast, tireless, occasionally brilliant, and catastrophically wrong without supervision. You don't fire the manager and promote the intern. You use the intern to make the manager faster.
The companies in the 95% are not failing because they chose bad technology. They are failing because they confused adoption with value. A utilization dashboard that reads 14% is not an AI problem. It is a judgment problem. Someone deployed a tool before anyone defined what winning looked like.
They are paying for the lesson twice: once in the budget line that bought the licenses, and once in the budget line that cut the people who would have known better.
Use-case discipline is the moat. The CFO buying the most agents is not winning. The operator who knows which three workflows justify AI investment, and can articulate why to a board, is winning. That operator is a human being. Today, they are often unemployed.
What this means for you.
If you are still employed, you already know something the CFO doesn't: which problems in your organization are actually worth solving with AI. That is not a trivial asset. That is the moat.
The 5% didn't happen by accident. They happened because someone in the room, maybe a RevOps lead, maybe a field SE, maybe a CRO who had been around long enough to see a CRM rollout go sideways, had the credibility and the judgment to say: not that problem, this one.
That person is more valuable right now than any license.
The question is whether their company knows it.
Intelligence is cheap. Insight is not. They just proved it.
- R.W.B.
Forward this to one person who should be reading it.