Promptathons: Kier’s key to making the most of Copilot

Sometimes more senior people can be more reticent to do something that they’re less aware of, and so it’s all about finding new ways of getting people to try something once or even twice. Certainly with Copilot, you don’t need to try many times to to to start engaging.

This was actually done by Unisys, who support us in our IT: they ran a promptathon, and they made it quite real. When you’re bidding for work, you can get files and files of information that you have to run through. We had all our senior leadership teams and our executive committee in a room, and half of them were given AI to search through these documents and come up with an answer, and half had to do it manually.

[The exercise] was practical, it was real, it was quick. We only spent an hour doing it, so it was nowhere near the time you normally get. But you just started and then started to see the art of possible, and then we swapped the teams over.

We’ve always said it’s human at the helm: it has to be. The human has to check, and that was reinforced [in the promptathon] because one or two of the people with AI were getting some answers that weren’t totally factually correct. So the exercise was embedding that need to check your answers. You use it to make yourself more productive, but you must be in charge and be responsible.

The promptathons are run regularly. We’ve got a full-day one coming up with our pre-construction teams from both infrastructure and construction coming together to address how we use AI to make us even better at delivering what our clients want and need.

We have ongoing literacy and upskilling training, which can involve promptathons. It can also involve communities getting together and sharing best practice.

None of us learn in the same way, so you you need to tailor training to suit many different personas. You need to keep rolling out training, with short snippets of “Do you know this? Do you know that? Try this prompt.”

Earlier this year, I was lucky enough to go on a strategy and leadership course at Oxford University, and they were talking about human in the loop, which I get. You hear about AI hallucinations – my wonderful Copilot has told me I built something I haven’t. The great thing with Copilot is that I’ve got that corporate memory, and it gives you the reference so you can check it. And part of our training is giving people the understanding of what a hallucination is and the importance of checking their work.

You hear a lot of jobs are going to go because of AI, but our intention is to make our people more productive and more effective, which means we can deliver more work and not not lose jobs. We want people in charge of what is brilliant technology.

Pulling the AI policy together took a while. It’s not easy. It has to continually evolve. Our people are asked to use Copilot. That doesn’t mean they can’t use other tools, but it has to be done with permission, with authority, and generally, people are finding that Copilot does what they need.

We keep an eye on things, so if people move away from that, we can pull them back into the fold, but generally, I think people now understand the ‘why’. If people know why you’re doing something, they’ll generally support it. If it’s just something that’s thrown at them – “you have to do this” – particularly with engineers, they’ll find a workaround. So we really concentrate on the why.

One of our work streams is just keeping that governance live and working out how we move as fast as technology is moving.

We are developing enterprise agents. One is specifically for our legal team to help them with NDAs. Again, it came out of one of the hackathons. And then safety is our license to operate. We said “let’s use technology to support our people to be as safe as they can be”, so the first enterprise agent allowed people to find out how to comply easily and quickly, and that’s got brilliant feedback. And we’re continually improving that.

We are doing something with our risk assessments and method statements, so we can take the next step and help people successfully produce succinct, compliant ways to set up people to work safely. We anticipate launching that in the next quarter.

We’re also doing some exciting stuff with prediction: we’ve got really strong safety data, and we’re using it to see if we can predict hotspots. What is it that happens ahead of an incident? Can we ensure we put a bit of extra effort into a hotspot to ensure it doesn’t happen? So using data to predict and thus avoid: that’s quite exciting as well.

Kier staff are empowered to produce their own agents. We’ve got just under 100 personal agents and team agents being used. What’s important is more than that have been produced and are not being used, but that’s fine: we want people to try, and if it doesn’t work, move on.

It’s really important that people are not scared to give something a go and realise it wasn’t quite what they wanted. Take our recruitment team: they’ve got an agent to help them produce job descriptions. Our comms team have got an agent to check documents for brand compliance. It’s just taking away the toil and enabling people to be really productive.

We want our people to feel comfortable using data and digital tools, and with the the rollout of some of our literacy programmes, we’ve improved confidence in Copilot up to 86%.

Keep up to date with DC+: sign up for the midweek newsletter.