A weekly read on what the AI noise is hiding.
Two numbers ran the AI headlines this week, and they tell opposite stories. MIT says 95% of enterprise GenAI pilots return zero measurable profit. Vendors say their agents return 171%. Both can't be the read, and the gap between them is the whole game.
What the AI noise is hiding. Every stat sourced. The hype named as hype. One number worth acting on. Read the full brief below, free, no email required.
The big firms all diagnosed the same disease this week. None of them named the cure. The full brief shows the gap they left open, the one honest case that proves both the win and the limit, and the single number worth acting on.
MIT’s “GenAI Divide” study found 95% of enterprise GenAI pilots deliver zero measurable impact on profit and loss. It’s still the most-cited number in the field, and nobody has knocked it down.
At the same time, vendors keep selling a “171% ROI” and claims that most buyers hit ROI in year one. Those figures are unaudited and aggregator-level. Name them for what they are: the hype the room already distrusts.
So which is true? Both, in a way. The dispersion is the signal. When results scatter from zero to 171%, it means nobody has a clean, agreed ROI number yet. IBM’s 2025 CEO study puts it plainly: only 25% of AI initiatives delivered the ROI companies expected.
The reason isn’t the model. It’s the coordination around it. Here’s the case.
BCG’s AI Radar 2026 splits success into 70% people and change, 20% process and data, 10% algorithm. Bain puts roughly two-thirds of the work in data, process, and change, one-third in tech. The model is the easy part. The humans around it are the hard part.
Adoption is near-universal. Scaling is rare. McKinsey’s State of AI reports about 88% of companies use AI and only about a third scale it. Gartner counts just 17% with agents actually deployed. BCG says only about 5% are truly agent-first.
The mood shifted too. Bain says boards are losing patience. Deloitte calls 2026 the move “from ambition to activation.” Per BCG, CEOs now lead about 72% of AI decisions and are pushing more than 30% of AI budgets toward agents. The patience is gone. The spending isn’t.
And the guardrails lag the spend. Deloitte’s State of AI in the Enterprise 2026 found only 21% of enterprises have mature governance for their agents. Agents are scaling faster than the controls around them.
Camunda’s State of Agentic Orchestration 2026 found 73% see a large gap between agentic vision and reality, only 11% of agent use cases reached production, and 80% say their agents are still glorified chatbots. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, and estimates only about 130 of the thousands of “agent” vendors are real. The rest is agent-washing.
Then there’s the one case worth studying, because it’s honest on both sides. Klarna put AI on the front line of customer service. The assistant handled about two-thirds of chats, did the work of roughly 700 full-time employees, cut resolution time from 11 minutes to under 2, and saved around $40M. Real win. Then in May 2025 Klarna walked part of it back and rehired humans for the complex cases. The honest version is both sides. It shows what agents do well and where they still need people.
Here’s what every firm did this week, and what none of them did.
They diagnosed the disease loudly. Adoption without value. Coordination and people as the real bottleneck. Governance running behind. Good diagnosis, all of it.
Then they prescribed more of the disease. The cure on offer is another coordination layer. Retrain hundreds of thousands of people. Rebuild the data architecture. Buy the platform. Run the transformation program. They tell you coordination is the problem, then sell you more coordination to manage it.
Nobody named the coordination tax as the thing to remove. They treat change management as the 70% cost bucket you manage, not a structural cost you eliminate. And it’s always a program, with a start and an end, not an operating model that keeps the org aligned after the consultants leave. The dependency is the business model.
One more gap. Every dataset is Fortune-scale. The prescribed fix doesn’t exist below a nine-figure budget. The $10M to $250M operator, the one who feels the coordination drag most, is standing on unclaimed ground.
That’s the wedge. You don’t manage a tax. You remove it. You install an operating model, not a program. And you do it for the mid-market operator nobody is serving.
95%.
Ninety-five percent of enterprise GenAI pilots return zero measurable profit, per MIT. Not because the AI is weak. Because the coordination around it never changed. The pilot sat on top of the same meetings, handoffs, and chase-work that slowed the company down before.
Want to know whether your organization is in the 95% or the 5%? Start by measuring the coordination tax you’re paying right now. That’s what the Agentic Readiness Score does. It’s free, it takes a few minutes, and it shows you where the drag is.
One weekly read on what the AI noise is hiding. Every stat sourced, the hype named as hype, one number worth acting on. No spam, unsubscribe anytime.
Start by measuring the coordination tax you’re paying right now. Free, a few minutes.
Next Tuesday: another scan, same rule. Cut the hype. Show you what the noise is hiding.