A few weeks into my MSt in Entrepreneurship at the University of Cambridge, our Organisational Behaviour module set us a team exercise. The point of it was simple. The smartest person in the room, working alone, can't make an organisation better. Your best performer might well come up with the best answer, but a team will almost always beat its weakest members, and organisations run on teams.
My instinct was to use AI and get us to a strong answer quickly and reliably. Some of my teammates weren't comfortable with that, and a few felt I had cheated. That stung. It took me a while to see that they had a point, and so did I, and that the gap between us was the same one I see in almost every boardroom I walk into. AI makes individuals quicker. It doesn't, on its own, make organisations better.
The excitement is real. The value is not yet.
Nobody needs convincing that businesses are excited. McKinsey's latest State of AI survey found that 88 per cent of organisations now use AI regularly in at least one part of the business (McKinsey & Company, 2025). In India, NASSCOM puts 87 per cent of enterprises in the middle stages of AI maturity (NASSCOM, 2024).
Ask about results, though, and the mood changes. In the same McKinsey survey, nearly two-thirds of organisations hadn't started scaling AI across the business, and only 39 per cent could point to any effect on earnings, usually under 5 per cent. This year, WRITER found that 97 per cent of executives say they benefit from AI, but only 29 per cent see a significant return from generative AI across their organisation (WRITER, 2026). Read that again. People are getting value. Their companies mostly aren't. It is my Cambridge exercise, playing out at scale.
Part of the trouble is how we count. An employee who uses a chatbot to tidy up emails counts as "adoption". So does a bank that has built a model into how it underwrites loans. Those are wildly different things, yet they land in the same survey and produce the same confident slide in the board pack, right up until someone asks what has changed.
Why the excitement isn't turning into value
Skills: Most companies have plenty of people using AI tools. Far fewer have people who can build, run and govern AI systems properly. In India alone, Bain & Company expects 2.3 million AI job openings by 2027, against a talent pool of around 1.2 million (Bain & Company, cited in People Matters, n.d.).
Data: This is the unglamorous one. IDC expects nearly half of AI use cases to miss their return-on-investment targets in 2026, with weak data foundations among the usual culprits (IDC, n.d.). Most businesses still run on systems that were never meant to talk to each other, let alone feed a live model. The algorithm is rarely the hard part.
People and process: You can buy an excellent model and still get nothing from it if the people meant to use it carry on working exactly as before. If a team of entrepreneurs at Cambridge can disagree about whether using AI is even legitimate, you can be sure the same argument is happening, quietly, in finance teams and middle management. No licence fixes that.
Size: Big companies are more likely to be scaling AI (McKinsey & Company, 2025). Smaller firms often lack the budget, clean data or specialist staff to get past the pilot stage. That worries me, because from India's MSMEs to Britain's small businesses, these firms are the backbone of the economy. If they're left behind, the gains from AI will pile up in a handful of large companies.
Trust: the hidden barrier to value
My teammates' discomfort matters more than it might seem. People won't build important work on a tool they don't trust, or one they aren't sure they're allowed to use. But I think the "cheating" label misreads history.
We have always built new work on top of old work, first in libraries, then with Google, and now with language models. Business, in particular, has always run on help that goes uncredited. Secretaries drafted letters that went out under the boss's signature. Ghostwriters write speeches and memoirs for leaders. Songwriters write the hits that other artists make famous. Nobody calls that cheating, because the person whose name is on it stands behind it. Universities are right to be stricter, since coursework exists to show what you have learned yourself. In business, though, what counts is accountability.
AI raises the stakes on that rather than lowering them. A language model doesn't go and look things up in a library. It produces plausible text based on patterns it learned in training, and it can state an invented fact, or an invented source, with total confidence. Regulation is starting to help. Since August 2026, the EU AI Act has required providers to mark AI-generated content (European Union, 2024), and major providers now build invisible watermarks into generated text. But researchers have shown those marks can be weakened by paraphrasing or translation (Nemecek, Chaudhary and Ayday, 2026). So, the useful question for a business isn't whether AI was used. It's whether anyone checked the output.
Turning excitement into value
None of this is an argument against investing in AI. It's an argument for doing it properly. Most organisations find it far easier to run ten pilots than to scale one, because pilots are cheap and scaling means changing how the business actually works. The companies getting real value have accepted that. McKinsey's "high performers", roughly 6 per cent of the companies it surveyed, are nearly three times as likely as the rest to have fundamentally redesigned their workflows (McKinsey & Company, 2025).
I started my career as a software tester in Nepal, and I still look at technology through that lens. In software, nothing goes live until it has been tested against what it is supposed to do. A lot of enterprise AI skips that step. It gets rolled out on the strength of a good demo rather than proof that it moves a business result. Before signing off the next AI project, I'd encourage any board to ask three questions:
- Which business number will this move, and where does that number stand today?
- Whose job, incentives or decisions change because of it?
- How will we test that it works, and keep testing once it's live?
Looking back, my teammates and I weren't really arguing about technology. We were arguing about what good work looks like. And the exercise had already given us the answer: even the smartest individual, with the smartest tools, can't make an organisation better on their own. The businesses that take that lesson seriously, and build the measures, working practices and testing around it, will be the ones that finally turn their excitement about AI into real business value.
The author is founder and CEO of Qniverse.
The opinions expressed in this article are those of the author and do not purport to reflect the opinions or views of THE WEEK.
