The 95% Problem: What MIT's 2025 AI Study Means for Service Businesses
Businesses have poured $30–40 billion into generative AI — and roughly 95% of those initiatives returned nothing measurable. The gap isn't the technology. It's how the AI gets implemented. Here's what that means for a local service business deciding whether to take the leap.
In 2025, MIT's Media Lab published one of the most sobering looks at corporate AI to date. After reviewing more than 300 publicly disclosed initiatives, interviewing dozens of organizations, and surveying scores of executives, the researchers landed on a number that should stop every business owner in their tracks: roughly 95% of enterprise AI initiatives delivered zero measurable return.
Not 95% were slow to pay off. Ninety-five percent returned nothing. And here's the part that matters most for a local HVAC, plumbing, or roofing business: the failures had almost nothing to do with the quality of the AI models. They came down to approach.
of corporate AI initiatives delivered no measurable return, per MIT's 2025 State of AI in Business study. Only about 5% reached production with real value.
Why so many AI projects return nothing
The barrier to "doing AI" is now zero. Anyone can sign up for a tool and have something running in an afternoon. That ease is exactly the trap. The MIT research, and decades of software history before it, point to the same failure patterns:
- Trend-chasing over problem-solving. Projects get green-lit because the business feels it needs "an AI initiative," not because they solve a defined, measurable problem.
- No integration. Generic tools get piloted widely, but the ones that actually reach production are embedded into the real workflows that run the business — not bolted on the side.
- No measurement. Winners measure outcomes. Losers measure demos. A system that looks impressive in a meeting is not the same as one that books jobs.
- Automating a broken process. Technology amplifies whatever you point it at. Automate a flawed process and you just do the wrong thing faster.
What the successful 5% do differently
The businesses that crossed what MIT calls the "GenAI divide" weren't using fancier models. They demanded process-specific customization, integrated AI into the systems that actually run the business, and measured results in dollars and booked work — not activity. In other words, they treated AI as part of the operating system of the business, not a novelty layer sprinkled on top.
That's the same principle behind every system I build: it has to integrate with how your business already runs, it has to be measured against real outcomes, and it has to be governed so it can't go off-script.
The finding that matters most for small businesses
One MIT result jumps off the page for any owner without a full-time tech team: AI built with an outside implementation partner reached production roughly twice as often as internal-only efforts — about 67% versus 33%. It isn't about intelligence. It's about mileage: the applied knowledge that comes from running implementation after implementation across different businesses. We unpack that finding in detail in In-House vs. an AI Implementation Partner.
What this means if you run an HVAC, plumbing, or roofing business
You don't have a $30 million AI budget to waste, which is exactly why the MIT findings are good news. The barrier isn't money or model quality — it's discipline. Start with a real problem (slow lead response, missed calls, estimates going cold), integrate the fix into the tools you already use, measure it against booked jobs, and put guardrails around it before it ever talks to a customer. Do that, and you land in the 5% that works.
That's the entire reason Bravo Automation Labs exists — to put local service businesses on the right side of that divide, safely.
Book a Free Systems Audit →Source: Forbes, reporting on MIT Media Lab's State of AI in Business 2025.