Why is GenAI failing?

95% of corporate GenAI projects are failing. That’s what MIT has said in its recently published report ‘The state of AI in Business 2025’. Think about it, that’s almost all projects. So, why do they fail, and what does it mean for projects in the smart city and utility industry? 300 publicly disclosed AI initiatives were studied for the MIT report, which found that, despite billions of enterprise investment going into GenAI, only 5% of custom company AI tools are reaching production and extracting millions in value. So, what’s going wrong? It’s not because workers don’t want to use AI. In fact, 40% of enterprises said they purchased an official LLM subscription, yet 90% of workers used personal subscriptions to tools like ChatGPT to increase individual productivity. What is it, then? According to the report, one of the reasons is investment. From the surveys and interviews, the MIT team discovered 50 to 70% of AI funding will go towards marketing and sales because leaders assume that's where money is. But the real ROI is in back office automation, like reducing outsourcing or streamlining workflows. Yet this automation remains underfunded. There’s also the fact that on many occasions, GenAI projects are too wide, and don’t focus on one issue to solve. However, the main reason is the tools’ inability to learn and memorize information it has been given. Many don’t remember context or improve with feedback. So employees repeat the same corrections, and systems repeat the same mistakes. This is why, according to the MIT report, only 5% of enterprise GenAI projects get past the pilot phase and turn a profit. And you may be wondering. Which 5%? It’s the tools that learn, so the ones adjusted to retain feedback, adapt to context and improve over time. So, properly customized tools with a specific pain to solve. The vendors doing this type of GenAI are securing deployments and contracts, and the buyers are securing savings, seeing customer retention and sales conversion. And interestingly, external customized tools succeed about twice as often as in-house builds. Now, if enterprises with massive budgets struggle this much, what about cities and utilities? Cities and utilities are curious about the potential of GenAI. We’re seeing them develop AI internally, as well as acquiring it externally. And they too are experiencing the same challenges but others too. Cities and utilities play in another league. They have completely different rules, times and budgets. Their work is specific, technical, and highly regulated. That’s where big AI players are falling short. Big AI players are Sillicon Valley, Big Five people who are used to B2C, and have built a conversation around that client. It’s hard for them to disrupt a very specific industry, where workloads, workflows and just general day-to-day are very different. A lot of the time, they don’t speak their language (as in vocab and issues, not literally), so their GenAI tools may fall short and not live up to expectations because they lack the knowledge and don’t focus on specific industry issues. And in Europe, the challenge is even greater with stricter EU regulations with AI coming from the US. AI in smart cities and utilities is like any other solution in terms of problems. Often we’ve seen projects fail because vendor and customer have focused on the tech and how new and different it is rather than what it solves. But there’s a change and what we’re starting to see is promising: cities not asking “How do we use AI?” but rather, “What problem do we need to solve—and if AI is the right tool, go for it. It’s a change in mentality, we need to give it more time, as changes in work culture take time. So all that’s left to say to cities and utilities is: Focus on the use case before plunging into GenAI tools, learn from the mistakes of others.