This week Microsoft rebuilt Copilot from three separate products into one app with three tabs: Home for chat and its new Cowork agent mode, Code for building small tools by describing them in plain language, and Autopilot for agents that run in the background, one of them named Dot, watching stock levels without being asked. The market liked it. Microsoft's shares rose more than 4% on the day, the best performer among the "Magnificent Seven" that Thursday.

Buried inside the announcement was a more interesting number than any of the new features. Microsoft has more than 450 million commercial Microsoft 365 seats. Only 30 million of them, 7%, are paying for Copilot. That's despite $100.3 billion in Microsoft 365 cloud revenue this year, and despite Copilot having been bundled into most Microsoft 365 plans and pushed hard for the best part of two years.

Read that number the way an agency owner should read it: not as a verdict on Copilot, but as a data point about adoption itself, from the company with the widest reach and the most reason to have solved it by now. If Microsoft, with a captive audience of nearly half a billion licences, can only convert 7% of them into active paying use, the honest conclusion isn't that Copilot is weak. It's that giving people access to an AI tool and getting them to actually use it well are two different jobs. Most organisations, of every size, are only doing the first one.

That's the exact pattern behind most of the "are we behind on AI" conversations Go Wisely has with agency owners. The worry is almost never about access. Most agencies already have some AI tool live somewhere: a ChatGPT subscription one strategist pays for personally, a Copilot licence bundled into the Microsoft 365 renewal nobody's looked at since, a Claude account one account director swears by. The gap isn't access. It's that the licence sits there, technically live, practically unused, while the team keeps doing the work the old way because nobody built the bridge between "we have this" and "we actually use this, properly, on real client work."

Go Wisely worked through exactly this with a 36-person global events agency. Before the training started, a full AI readiness diagnostic across six dimensions showed the same shape Microsoft's own numbers now confirm at scale: pockets of confident use sitting next to genuine access that nobody had turned into a habit. The response wasn't a longer tool list or a better recommendation. It was six structured sessions built to run across the whole team, not just the people already comfortable, so a junior coordinator and a senior producer left with the same working fluency, rather than one more confident colleague and twenty-five people still waiting for someone to show them how. The programme scored 4.6 out of 5 across 131 survey responses, with not one dissatisfied response, and the number that matters most isn't the score. It's that the use spread across the team rather than staying with the two or three people who'd have picked it up anyway.

None of this requires an agency to distrust the tools. It requires being honest about which half of the adoption problem is actually solved. Buying the licence, subscribing to the plan, switching everyone onto Claude or Copilot or whichever tool fits the brief: that's the easy half. Microsoft's own numbers this week prove that even the company selling the tool hasn't cracked the other half yet. The harder half is building the habit, briefing the use case properly, and checking, honestly, months later, who's actually opening the tool during a real deadline week rather than during the training session where someone was watching.

An agency that gets that second half right isn't behind the market. It's ahead of Microsoft's own customer base, and right now, that's a genuinely low bar to clear.

Go wisely.