OpenAI recently released the largest study of ChatGPT usage to date – 1.5 million conversations analysed. The headline? Less than 30% of usage is work-related.
If you’re in HR and you’ve been tasked with “fixing AI adoption” after your company spent six figures on Copilot licenses, that stat should concern you. Because it means your employees already know AI works. They’re just not using it at work.
I’ve trained thousands of professionals and I can tell you exactly what’s happening. Your employees tried AI at work a few times. It didn’t deliver. They went back to using it at home where the stakes are lower – helping kids with homework, planning holidays, writing awkward emails to their landlord. At work, when it takes more time to fix the AI’s output than to do it yourself, you just stop.
Then the question comes down from the finance department next door: “we spent £150,000 on licenses and adoption is flat. What’s the plan?”.
What’s actually broken
The utility threshold is the exact point where AI either saves someone time or costs them time. Most of your employees never cross it. They’re stuck at 98% – putting in 100% effort, getting 98% back.
Say you have 500 employees, spending £150,000 annually on Copilot licenses. Without training, you might see a slight increase in net productivity value. Your employees might use it for 15 minutes a day and get occasionally useful outputs, but it isn’t anything transformative.
Add £100,000 for proper enablement training and that number jumps to millions in net value. These are the same tools, and the same people. The difference is they’re each gaining back 5-6 hours per week because they know how to make it work for their specific workflows.
You see, only 10% of effective AI training is about using the tools. The other 90% is mindset shift. How do you think differently about starting a task? Do you get AI to ask you questions instead of asking it questions? That’s what separates the people gaining 30 minutes a week from the people gaining 5 hours.
There could be a lot of productivity (and bottom line) gain left on the table. All because finance wouldn’t approve more spend for training. When you go back to that CFO meeting, show them that slide.
What it looks like when people actually cross the threshold
I worked with an HR team that was asking AI to “write our employee communications” and getting generic corporate jargon that, let’s face it, nobody wanted to read. We changed the approach. Instead we fed in 5 high-engagement internal updates from the past year, adding the company’s actual tone of voice, along with the specific message they needed to land, then asking AI to critique those drafts against what actually gets people to open and act on internal emails.
HR trusted it because they were still writing the communication – AI was just pressure-testing whether it would actually land. Time to draft and approve company updates dropped from 3 days of stakeholder back-and-forth to same-day turnaround.
I saw the same pattern with a sales team asking AI to “write me a cold email” and getting a terrible output that took another ten minutes to redraft. We fed 10 top-performing emails plus ICP notes into a prompt library. Then switched from generation to critique: “Score this draft against our 5-point framework, then rewrite with 2 alt versions.”
Reply rates went up. Time-to-first-draft went down. Within a month, the same sales team that was complaining started asking when they could get AI into their proposal process.
The pattern is always the same. People don’t need AI to do their job. They need AI to help them do their job better.
When you’re stuck between legal, ops, and leadership
Here’s what I hear in our community of 25,000+ professionals every week: “We had a big launch and nothing since.” “Legal says no, but won’t tell us what yes looks like.” “We bought three tools and I don’t know which one to use for what.”
If you’re “using AI” on paper, it will have zero impact in reality. I often see legal teams blocking anything touching customer data. And as HR, you are stuck in the middle – operations want results, legal want control, leadership want ROI numbers.
Here’s what works: create a red-amber-green risk framework. Low-risk workflows (internal comms, meeting notes) get immediate approval. High-risk ones (customer data, contracts) go through legal review. Legal can finally say yes to something.
Then build context packs for each team: their style guides, best templates, FAQs, past examples. With that context loaded in, AI stops producing generic outputs. Tasks switch from “generate from scratch” to “critique this draft” or “extract key points with citations.”
The moment legal has a framework instead of blanket resistance, and teams have context instead of generic ChatGPT, everything moves. That’s the unlock.
What to actually do this quarter
You don’t need a massive transformation programme. You just need momentum. And you need your management team to lead by example. The top-down effect is critical is you want widespread adoption.
Pick one high-pain workflow. It should be something people complain about every week. Get three people to fix it with AI using proper context and guardrails. Run it for four weeks. Track the business metric, not the usage dashboard. Then showcase it. One of those people could do a 30-minute show-and-tell. Real work, real results, real prompts people can copy.
This works at every level. I’ve implemented these processes from enterprise clients like Hyatt and Lufthansa down to two-person startups. The pattern holds. Stop measuring whether people logged into Copilot. Start measuring whether they can’t imagine working without it.
The OpenAI study shows both work and non-work usage growing every month. Your employees know this technology works because they use it at home where the risks are lower. That’s on enablement, and enablement is your job.
Look, you can be the one to lead that transformation, or spend the next year in meetings explaining why adoption is still flat.
Joshua is the Co-Founder and CEO of Mindstone, an AI training platform helping non-technical professionals build practical AI skills. Mindstone works with some of the world’s largest organisations, including Hyatt, Pearson and Lufthansa, to make AI accessible, useful and impactful across entire workforces. Previously, Joshua co-founded SuperAwesome, which became the world’s largest kids’ online safety technology platform and was acquired by Epic Games for a nine-figure sum.
With a BSc in Computer Science from King’s College London and an MBA from the Open University, Joshua brings deep technical foundations together with commercial and leadership experience. An EdTech enthusiast, advisor and investor, he is known for finding simple, innovative solutions to complex problems and for building teams capable of scaling ambitious ideas.
- Joshua Whole
