Last year, Taught by Humans was awarded an Innovate UK grant to build AI competency training for key sectors - specifically the AI Solutions for AI Competency in Key Sectors programme (a bit of a mouthful). We decided to focus on construction, which doesn't always get a lot of attention in the AI conversation but probably should.
We spent the summer doing user research interviews with people actually working in construction, and then the end of the year building a curriculum. By January 2026, we had 10 companies and over 200 learners starting a five-week programme. We're just wrapping up now, and I want to share what we found - including the things that surprised us and the things we definitely got wrong.
How It Worked
We went with a microlearning approach: five weeks, five short lessons per week, with a daily email nudge to keep people moving. Each lesson gave learners a choice - read the content, or chat with Dotly and get something built around their role, their company and their industry. That personalisation ran at three levels: industry context (construction-specific use cases and language), company context (the tools and workflows each business actually uses), and individual context (what each learner already knew and how they were engaging).
The companies ranged from a two-person firm to teams of over a hundred. From CEOs to architects to site managers. That range turned out to be one of the most interesting parts of the whole thing.
What We Found
Structure Matters More Than Flexibility (This One Surprised Me)
We gave learners the option to switch to self-paced - no daily emails, do it whenever. Some people did switch. And those learners were significantly less likely to finish the programme.
The daily email nudge wasn't annoying. It was the thing keeping people going. I'll be honest, my instinct in learning design is often to give people more flexibility and autonomy - but sometimes what people actually need is a prompt. Something we'll be incorporating going forward.
How People Used Dotly Varied Enormously - And That's Fine
Some learners didn't use Dotly at all - opting to just read the content. In our early data collection, those learners are feeling less confident in AI overall, and don't have examples of where AI is actually useful in their jobs.
For this pilot, we let people make that choice. But going forward, some interaction with Dotly is going to be a requirement to completing learning - because if you haven't had a conversation built around your role and your context, it isn't really personalised learning.
That said, we want to do this in a nuanced way, because learners have genuinely different styles. Some want to read the content first, then have a quick chat with Dotly to consolidate. Some save Dotly for a Friday session and review the whole week in one go. Some skip the reading entirely and just talk. All of those work. Clicking complete without any real engagement doesn't.
The Finding We Really Didn't Expect
One of the larger companies had no approved AI subscriptions. No ChatGPT, no Copilot, no Gemini. Most of their peers in the cohort were learning about tools they were already using - they could bring real examples, test things they'd tried, ask about actual situations they'd run into.
This company's learners were doing something different. They were using Dotly as their AI, not just learning about AI in the abstract. And by the end of the programme, they had practical use cases they'd developed themselves, through the conversations. We didn't design for that. But it might be the most interesting thing that came out of the whole cohort.
What We Learnt
The most practical lesson: people do not read emails the way you think they will. All our instructions about how the programme worked were in the welcome email. Early learners completed all five lessons in a single day rather than one a day.
Which is technically impressive. But not really what we were going for.
Instructions need to be at the point of need, on the platform, at the moment someone needs them - not buried in something they read once and forgot. We've taken that on board.
Echoing What We Already Knew
None of this is that surprising when you look at the learning science (or honestly just at how people actually behave).
- Meeting people in their context matters
- People learn differently, and that's not a problem to solve - it's a reality to design for
- Practical learning sticks better than passive reading
- People genuinely appreciate being recognised for their effort, even in a short programme
The construction cohort reminded us why we built Taught by Humans the way we did. And gave us a few things to go back and fix.
We're still collecting the final data and will share more when we have it. If you want to know more about the programme or what AI training built around your team's actual context looks like,