Making Copilot Stick.
An AI adoption programme for a major construction contractor: behaviour, not licences.
The Problem
The licences were bought. The behaviour hadn't changed.
Around 300 people, spread across sites, most of them nowhere near a desk, and every level of scepticism from resistant to already-sold. Enterprise Copilot was switched on. Whether anyone used it well was the real design problem.
What Shipped
What three months delivered.
6
Modules and 60+ hours of training collateral, all held to one enterprise standard.
2
Working Copilot Studio agent prototypes, aimed at the organisation's highest-friction tasks.
300
Staff in scope across a distributed, site-based workforce, most of them nowhere near a desk.
The Goal
Move everyone one step. Not everyone to expert.
The workforce started clustered at the sceptical end. The goal was never to push everyone to advocate, only to move each person one honest step along.
Toggle it and watch the whole cohort shift right by one.
The Approach
Sequenced like a curriculum, not a compliance course.
Phase one sat first because nothing else works until Copilot is a habit. People met it inside chat, Teams and email (the tools they already open every day) before being asked to do anything harder. Win here, and the rest of the sequence has somewhere to stand.
Modules included Chat & Teams, OneDrive & SharePoint, and Copilot Agents, each with trainer run packs, participant handouts and live demo scripts.
The Build
Two agents, built for the highest-friction tasks.
Purpose-built in Copilot Studio, with conversational-UI principles applied to how they talk.
Agent One
Decoding project jargon
Every project runs on its own shorthand: acronyms, codes, inherited names. The agent untangles them against the organisation's own material, so new starters stop guessing.
Copilot Studio prototypeAgent Two
Reliable answers from risk material
Risk documents are long, dense and consequential. The agent answers from the source and shows its working, designed so a confident wrong answer can't pass as a right one.
Copilot Studio prototypeWhy It Matters
The difference a design choice makes.
Same synthetic risk question, two different levels of trustworthiness. Toggle between them.
For confined-space entry you'll need a hard hat, safety glasses, steel-capped boots, high-visibility clothing, gloves and a full-body harness. Make sure gas monitoring is in place and a standby person is present at all times.
This should cover the standard requirements for most confined-space work.
Why It Matters
New starters stop guessing.
Site message
The agent untangles shorthand against the org's own material, so the answer is theirs, not a guess from the open web.
The moment someone watched an agent untangle their own project's shorthand, AI stopped being an initiative and started being useful.
Still here? You're my kind of reader. There's more below.