Originally Published on December 10, 2025 on LinkedIn
A recent MIT GenAI study made waves by claiming “95% of AI projects fail to produce positive ROI.” Social media ran with the stat — but the real value isn’t in the 95% — it’s in the 5% that succeeded. I went directly to the source instead of relying on headlines and hot takes. MIT interviewed 300+ executives who had already deployed AI, and success was defined simply: did the project generate positive ROI?
Why most AI projects failed
MIT’s conclusion was blunt:
The core barrier isn’t infrastructure, regulation, or talent. It’s learning.
Most GenAI systems don’t retain feedback, don’t adapt to context, and don’t improve over time. Companies poured effort into sales and marketing pilots, built custom solutions internally, and deployed tools that couldn’t be meaningfully customized to real workflows. Most companies that built internal solutions failed (approx. 2 out of 3). Survey respondents expressed trust issues with selected AI vendors as flashy demos flooded their email inboxes. Another point of interest was that many GenAI adopters were impressed by external use of Microsoft CoPilot and ChatGPT but underwhelmed by internal GenAI tools.
What the successful 5% did differently
The winners picked narrow, workflow-specific problems, they bought instead of built, and they empowered managers and power users instead of running everything from a centralized AI lab. Moreover, GenAI software vendors’ tools were judged by business outcomes, not software performance. One of the most interesting insights: successful companies treated AI vendors less like software providers and more like business service partners, customizing solutions for their client’s specific needs.
The Catalyst for Success
It wasn’t the model. It was the Prosumer — the producer/consumer. Prosumers have been critical to innovation in the digital age from blogging to developing open-source software. Now, prosumers are leading the charge in successfully adopting GenAI. Front-line managers and power users who knew exactly where the system was broken and were empowered to fix it proved to be the most adept at successfully deploying AI solutions. They understood their processes end-to-end, prioritized what mattered, utilized high-quality, production-ready data, and chose tools that could be customized to fit their organization — not the other way around.
The secret sauce: Empower front-line managers and leverage power users to fix what’s broken.
More to come in Part II - I’ll dig into how Ordinal Prime enabled a US pharmaceutical company to successfully adopt AI.