A few weeks ago, I spoke to a leader at a business that had spent the best part of two years, and a budget I won’t embarrass them by quoting, building an AI capability that has so far produced almost nothing they can point to. The models were not the problem, and the technology worked exactly as advertised. But when they finally pointed it at their business, it was reasoning over a version of their customers that barely resembled the real thing.
I think about that conversation a lot, because it is not unusual. MIT’s Project NANDA recently found that 95% of enterprise AI pilots fail to deliver measurable return; only one in twenty produces real value. For all the noise around copilots, agents and automation, that may be the most important number in technology right now, and it is not a number about AI at all. It is a number about context.
Innovation at pace, or unnecessary AI noise
AI models continue to get better. Microsoft is moving fast, embedding Copilot, Work IQ and agents across everything it makes. The pace of development and rollout is never in doubt. But what is, is whether AI has anything worth reasoning over.
Most organisations are running AI on top of fragmented information scattered across disconnected systems and the usual organisational silos. And the single richest source of intelligence any business has, what customers actually say to it, on calls, in meetings, over messages and email, is precisely the part that tends never to reach the systems the AI depends on. So we are asking these remarkable models to make sense of a company from a record that is missing the most important thing in it: the conversation.
That is why I have become convinced that customer engagement, of all places, is where AI is most likely to pay back first.
Think about what is actually in those conversations. Every day customers tell you what they want, what is annoying them, where they are about to spend money and where they are quietly getting ready to leave. For years that was almost impossible to capture properly and harder still to use at any scale, so most of it evaporated. That is the part that is changing.
The businesses I see getting real returns are not the ones that started with AI. They started with the conversation. Rather than launching another sprawling transformation programme, they have extended Microsoft Teams, where their people already work, into the place customer engagement actually happens. Voice, AI voice agents, messaging and digital channels all sit in the same flow of work, and every interaction is captured as structured intelligence and written back into Dataverse, where Copilot and the agents can finally see it.
The effect is not subtle. A service team starts spotting the same issue recurring across thirty accounts before it becomes a systemic failure. A salesperson finds a buying signal that was sitting in a support call nobody thought to pass on. A leader sees a pattern of customer frustration that no dashboard was ever going to show, because the dashboard only ever had the structured fields, never the words.
The reality of the situation
None of this is magic, and anyone telling you otherwise should be treated with caution. You still have to capture the conversations properly, govern the data and be disciplined about what you are actually trying to find.
What has genuinely changed is the starting point. The businesses moving fastest are not modernising their entire data estate first. They are building on the Microsoft investments they already own and extending them, often through the Marketplace, so that what used to be a months-long integration becomes a matter of switching something on and configuring it well.
I keep seeing the same pattern: fix the operation first, build a real understanding of the customer second and let AI accelerate it third. The companies pulling ahead are not the ones spending the most on AI. They are the ones giving AI the best context to work with.
Which brings me back to that 95%.
If most AI initiatives are failing because they never solve the context problem, then the question is not whether AI will create value. It plainly will. The question is who ends up in the 5% that captures it.
And the businesses in that group will not have found a stronger model than everyone else, because everyone is going to have the same models. They will have done the unglamorous work of making sure their customers’ conversations actually reach the place where the AI can use them.
That is the part I would not underestimate. In the AI era, access to AI will be universal. Access to your own customer context will not. That is where the advantage is going to come from.
