Seven lessons in artificial intelligence

While most businesses now have artificial intelligence (AI) policies in place – largely to police quality, enforce ethics or address data protection issues – relatively few are pushing AI’s capabilities in formal project management processes beyond the basics. That’s changing fast.

A proliferation of AI tools and embedded AI functionality in existing systems is forcing teams to consider how to optimise it, scale its usage and redesign their operating model as new capabilities, use cases and hygiene factors come into play.

We caught up with three experts who are using AI to optimise operations and ensure they’re ready for what’s next, to give you seven takeaways you can use in your job.

1. Champion it

The rapid development of AI technology, allied to inconsistent usage by people in project teams, has sustained the hype/scepticism duality that’s defined AI since ChatGPT first emerged more than two years ago. Convincing people it can work – and ensuring others use it sensibly – is job number one.

“There are new AI tools cropping up every other day that do just about everything you could ever want,” says James Doherty, a project controls expert at BMT, which undertakes projects for a variety of organisations, including the Ministry of Defence.

“The challenge is implementing these safely at scale. This isn’t just rolling out Monday.com; it takes effort behind the scenes."

2. Experiment

Optimisation is about finding the appropriate application of AI to specific tasks. Often that means the best results won’t come from packaged solutions.

“I was looking for a way of speeding up our project controls process so that we could audit the resourced schedule we’d produced against the basis of estimate,” says Doherty.

“It would’ve taken a project controller at least a week or two manually, so I wrote a macro using ChatGPT to extract the schedule data from Project Online, then used the same AI to write the sophisticated Excel formulas needed to analyse the data. It turned the whole exercise into something that you could do in an afternoon.”

Remember, if it doesn’t quite work, try a slightly different approach or come back to it in six months’ time to see whether the models have evolved.

3. Control

Experimentation has to happen inside risk management protocols.

“To overcome those security challenges, we would typically set quite generic tasks,” says Doherty.

“Until recently, we’ve needed to be descriptive about the objective, but without mentioning any specific project information. The licensed Copilot that we’re using now has enterprise-grade data protection, but there are still limits. We can provide some more project-specific information, which significantly helps with context and the usefulness of outputs.”

4. Evaluate

“It’s very important for a project management team to have people to filter through all the nonsense and cherry-pick the tools that will really work for them,” says James Garner, Head of AI and Data at Gleeds. This means forming a project team or committee with both technical and delivery expertise to vet, trial and select AI solutions.

“There’s lots of rubbish out there, but there is some amazing stuff, too,” he adds.

5. Refine

Gleeds rolls out AI to selected project teams – chosen to ensure that it’s not just techies who get to experiment – so that it can refine approaches all the time.

“We’re constantly looking to create feedback loops and learn from mistakes,” says Garner. “The challenge is to do that when you’re trying to run your projects and your department: it happens in parallel with business as usual.”

6. Scale

At Gleeds, the team are mindful that scaling is as much about acceptance as it is about getting the tech right. That’s why the pathfinder teams and the feedback they offer are so important. But they’re also aware that if project teams start to use ‘shadow AI’ (on their own devices) to get work done, all the good work on controls and risk could be undone. Scale too fast, and you accumulate risk; too slow and you risk losing momentum or control.

7. Upskill

Optimisation demands AI know-how.

“You should feel suitably qualified and experienced before using AI to generate any work for you,” says Doherty. “It’s about assurance. You have to sign your name at the bottom of anything you’re delivering.”
Training is therefore key.

“We are making sure that all of our people are being given the appropriate course for them,” says Garner. “It might be very basic, but we’re also putting people on courses to develop higher-level knowledge, too, right up to master’s level. Then there’s more informal training to coach people through the use cases in the organisation.”

Read more in the autumn 2025 issue of APM’s Project journal