I was preparing a talk for the National Fisheries Institute’s Leadership Summit the other week on AI and data-driven decision making in seafood. The same conversation kept coming up: what’s real, what’s hype, and where AI is actually delivering value. One report has been shaping that debate—MIT’s State of AI in Business 2025. You’ve likely seen the headline it sparked: “95% of AI pilots fail.”
It even moved markets.
Read past the headline and the story gets more useful. The study isn’t just about failure, it’s about why pilots stall, where AI is working, and what separates a few real wins from a long list of experiments.
Let’s dive in a little.
MIT analyzed 300 publicly disclosed AI initiatives and interviewed 153 senior leaders. It’s not a massive or perfectly structured dataset, so I was skeptical at first. After reading the full report, my view shifted: the value here is in the patterns.
Here’s the core idea:
Companies have invested roughly $30–40B in enterprise AI.
Adoption is sky-high—over 80% of organizations have tried tools like ChatGPT or Copilot.
Transformation is low. About 95% of efforts deliver no measurable return.
MIT calls this the GenAI Divide. Most organizations are on the wrong side: lots of testing, very little structural change. Why? Generic tools are easy to try, hard to integrate. Chatbots spread because anyone can use them. Custom enterprise tools stall out—only ~20% reach pilots and just ~5% make it to production. Seven of nine sectors showed little to no structural change; the exceptions were professional services and media, where AI fits cleanly into existing workflows.
The report points to a simple root cause: today’s AI doesn’t learn.
These tools are great for quick, tactical tasks. Seventy percent of workers prefer AI for drafting emails. Sixty-five percent use it for simple analysis. But for strategic or mission-critical work, humans still win by roughly nine to one.
Why? Every session starts cold. You can correct an answer, but the system won’t remember your correction tomorrow. It won’t adapt to your team’s norms. That memory gap keeps organizations stuck on the wrong side of the divide.
There’s another wrinkle: employees often trust consumer tools more than enterprise ones. Leaders reported that a $20 ChatGPT subscription felt more useful than a $50,000 specialized system. As one lawyer put it: “Our vendor tool gives rigid summaries. With ChatGPT, I can iterate until I get what I need.”
One finding seafood companies should note: build vs. buy outcomes weren’t close.
Internal builds reached deployment about 33% of the time.
External partnerships, especially when vendors customized to real workflows, hit 67%.
The difference is focus and follow-through. External partners bring domain experience and accountability; internal projects can drift into science experiments that never scale. In aquaculture—where budgets are tight and teams are lean—the lesson is clear: partner smartly. Don’t try to reinvent the wheel in-house.
And yes, I see the irony. I run a company that builds AI for aquaculture. But I believe this because we’ve watched it play out elsewhere: the wins come from partners who know the industry, tailor to workflows, and keep improving the system over time.
Another pattern: where money goes vs. where returns appear.
Most budgets flow to sales and marketing (often 50–70%) because results are easy to measure: faster lead times, better click rates, more demos. But the real ROI shows up in the back office: finance, operations, procurement. The companies crossing the divide aren’t bragging about chatbots; they’re saving millions by cutting outsourcing contracts, automating invoice processing, and reducing agency fees.
Technology adoption tends to rhyme.
Personal computers. In 1983, Steve Jobs predicted the computer would become the predominant medium of communication, overtaking television and even the book. Bold then, obvious now. But it still took 15–20 years for PCs to become everyday infrastructure at home and at work.
The internet. The web launched in 1989. E-commerce and cloud didn’t reshape business overnight. In 1995, Newsweek ran “The Internet? Bah!” arguing online shopping wouldn’t replace the mall. The vision was right; the timeline was wrong. It took years of standards, infrastructure, and habit-building.
AI is next. Believers say it changes everything; skeptics say it’s overhyped. History suggests both are partly right. AI will reshape work—but not instantly. Memory, adaptability, and integration still need to mature. We’re not even three years past ChatGPT’s launch. Pair that with the MIT finding that 95% of enterprise pilots stall before production, and the takeaway is simple: we’re early.
Be deliberate. Don’t chase every shiny tool.
Focus on fit. Pick systems that work inside real workflows and keep learning over time.
Look behind the curtain. Expect the biggest ROI in back-office operations before the flashy demos.
Partner where it counts. Favor external collaborations that customize to your processes and stay accountable to outcomes.
I’m confident AI is foundational. The technology is real and valuable. Like the computer and the web before it, it will take years before it feels second nature. That’s why seafood needs to stay clear-eyed: experiment, stay curious, and keep your focus on workflow fit and measurable ROI.
The real transformation will come soon enough.