Kier Group builds schools, prisons, and major water and nuclear infrastructure projects across the UK. It employs around 11,000 people. Two years ago, the company started pushing Microsoft Copilot across its workforce, and today about 40 percent of staff use it daily, with the rest running a lighter version built into Microsoft Teams.
What stands out is not the tool itself but how Kier got people to actually use it. Rather than mandatory training modules, the company ran small group sessions called prompt-a-thons, where colleagues tested their own answers against Copilot's. That peer-to-peer format, done in small groups instead of solo webinars, is what got adoption to actually stick.
This matters because most companies do not struggle to buy AI tools. They struggle to get employees to open them. Recent UK data shows AI use across businesses is still shallow even where the tools are already installed, and staff hesitate for the same reason Kier's own leaders did at first: the fear of asking a question that makes them look behind.
The bigger story is where Kier is heading next. It is building AI agents that go beyond writing emails or summarizing meetings. One agent organizes safety data collected across job sites so managers can spot risk patterns before someone gets hurt. Another is trained on Kier's safety standards and paired with drone footage and site photos, cutting the number of times a worker needs to climb scaffolding or stand near heavy machinery just to check on progress.
That focus on safety is not cosmetic. Construction remains one of Britain's most dangerous industries, recording 35 fatal injuries and around 50,000 non-fatal injuries in the most recent reporting year, more than any other UK sector. Yet construction also lags badly on AI: official figures put AI use in the sector at just 13 percent of businesses, well behind finance and retail.
Kier is not alone in moving early. Skanska and Balfour Beatty have both been building their own AI systems for safety and project work since 2024, with Balfour Beatty running a Microsoft-based tool inside its own network to test similar use cases. But Kier's approach, pairing everyday Copilot use with purpose-built safety agents, gives a clearer picture of what a full rollout looks like in an industry that has been slow to move.
There is a labor angle too. The sector needs roughly 240,000 new apprentices over the next decade just to keep pace with demand, and Kier's leadership is clear that AI is meant to support new workers learning the trade, not replace the need for them. For any business in a physical, safety-heavy industry, from manufacturing to logistics, the lesson is less about the software and more about the method: informal, low-pressure training beats a mandate, and the biggest early wins come from pointing AI at the most dangerous parts of the job first.
Microsoft's push into physical, hands-on industries is worth watching closely. Large companies in finance and technology have adopted Copilot far faster than construction has. Kier's experiment suggests that gap is starting to close, and the industries next in line are the ones where a mistake costs more than a missed email.