The State of AI in Business 2025 Report finds that 95% of enterprise AI pilots are failing. Well OK, MIT isn't Cambridge, but we should hear them out. Here are my thoughts.
1. Because “using AI” isn’t the same as designing for it.
Consumers interact with AI at the surface: a chat window, a recommendation, a helpful auto-complete. The product team behind it designed the whole loop — how data flows in, how feedback shapes behaviour, how human judgement steps in when needed.
Most businesses, by contrast, bolt AI on to broken processes. They throw a chatbot at bad customer service, or a language model at disjointed data. The result feels clever for a week, then collapses under the weight of its own improvisation. If you don't have a joined up data strategy, or martech approach, AI doesn’t fix chaos. It amplifies it.
2. Because ownership got lost in the hype.
In the race to automate, many teams handed their logic and data to black-box platforms. The vendors keep improving; the clients keep renting. When systems stop behaving, no one can explain why. What started as a time-saver becomes a new dependency — clever, opaque, and quietly expensive. Owning the system means owning the reasoning. If you can’t see how your AI makes decisions, you don’t own the value it creates.
3. Because “human in the loop” became a slogan, not a design principle.
We love to say “humans stay in control,” but few companies actually map where that control lives. Who reviews model outputs? Who decides when the machine is confident enough to act? How do we learn from its mistakes? Without that architecture, “the human in the loop” is just a marketing promise. The real trick isn’t removing people — it’s keeping them in the right places, with better feedback and less grunt work.
4. Because culture still trumps capability.
The hardest part of AI isn’t the model — it’s the mindset. A culture that rewards speed over understanding will keep buying faster hammers for the same bent nails. A culture that treats automation as an experiment, not a replacement, will actually learn from it. Smarter tools need wiser teams. Otherwise, you’re just giving a toddler a power drill.
So what works?
Start by redesigning the loop. AI should be a partner in process, not a passenger in chaos. Own your data, your logic, and your feedback systems. Make it clear where humans add judgement and where machines add stamina. Build less magic, more clarity. When that happens, AI stops being a stunt — and starts being part of the craft.
At The BotForge, we help companies do exactly that: agentic workflows that keep people in the design loop, not out of it. Because the real goal isn’t a business run by AI — it’s a business that keeps getting smarter because you are.