Usage is visible…capability, not so much
You’ve seen the slide. Active usage (Copilot, Claude, ChatGPT, pick one) is sitting in the high 90s. Everyone’s excited, and looking forward to the return on the investment. But this figure tells you almost nothing about the level of capability of the people using the tool. Microsoft’s own definition of active usage is someone who has used Copilot at least once during the selected period.
However, I hear the same frustration on repeat: the usage number is high, but the ROI still isn’t landing.
Usage without capability doesn’t compound.
Two organisations can have the same headline number but be in a completely different place. One provides training, and a handful of champions drag the average number up. The second pairs training with genuine change management, and a third of their workforce has started to innovate on what work might look like. Same headline, different trajectory.
Design the right approach
ROI leaks when you aren’t focusing on human behaviour. I come across people who are excited about using their genAI tool all the time. But when you dig deeper, the excitement is all about the basics. A single four-hour session covering features and functions leaves people reaching for the next capability up from where they started. Everything else in the session barely makes an impact. These sessions rarely cover the different motivations needed to create a spark, how much leadership behaviour shapes tool use, or the social learning that turns a skill into a habit.
A well-designed approach, one built on change management principles vs a single training session, builds lasting capability. A single session with personalised examples isn’t that, no matter what it’s marketed as.
Is efficiency leading to cognitive offloading?
Thinking is slow. And when an organisation is focused purely on efficiency, AI gets celebrated for removing that slowness. I think that’s exactly why we are seeing more and more of what is being called “cognitive offloading”.
A feature-function training approach reinforces the “efficiency only” view held by leaders across the organisation. It has been built around an efficiency operating model: get the training done, get the people using the tool. If the culture around Gen AI use is “get more done, faster”, then you are building habits where people stop checking their work. More prompts, less human oversight. That is not capability. That is usage without oversight - and it definitely doesn’t compound.
If you’re worried AI is dulling your people, check what your operating model has been rewarding for the last decade.
A different approach to building capability
A change management approach can also be slow, but it is what protects the organisation from cognitive offloading. Giving people time and space to review current work practices - what matters, what AI can help with, how their own behaviour needs to shift, is what makes adoption stick.
Coming back to measurement: rethink using averages. If you have a group of enthusiastic champions who’ve genuinely started to change how they work, they could be inflating your figures. The median often sits well below the mean when this small group of heavy users holds up the numbers. Find ways to encourage (and measure) social learning across the organisation. This might be including an “AI moment” at the start of every meeting, something Rio Tinto used effectively for their “Safety Shares”. (Believe me, I no longer include my superglue next to my eyedrops in the fridge after one of those shares!)
The tool isn’t shaping how your people think. Your rollout is.