Attach it to a habit that already has a time and a place, give it to one crew for one task for one month, and make sure it removes more steps than it adds.
Key takeaways
Attach the tool to a habit that already has a time and a place, give it to one crew for one task, and make sure it removes more steps than it adds. Adoption on a job site fails for practical reasons far more often than cultural ones — a second login, a phone with no signal in a basement, a form that takes four minutes at the end of a ten-hour shift.
The crews that adopt new tools are not the ones that were persuaded. They are the ones for whom the new way was genuinely shorter than the old way on the first day.
Every crew already has fixed points in its week. The end-of-shift log. The Monday toolbox talk. The Friday timesheet. The delivery signature. A tool introduced at one of those moments inherits a slot in the day; a tool introduced on its own has to win one, and it will not.
This is why “take a photo of the wall before you close it” adopts easily and “open the app and write a summary” does not. The first is a small change to something already happening.
Before and after. If the old way was three taps and the new way is five, no amount of explanation about downstream benefit will hold, because the downstream benefit accrues to the office and the five taps accrue to the foreman. Where the new way genuinely is longer at the point of use, say so and pay for it in some other currency — fewer phone calls from the office, an earlier finish on Fridays, a shorter timesheet.
The one-crew, one-task, one-month rule
One crew, chosen because their foreman is respected rather than because they are the most technical.
One task, narrow enough to describe in a sentence: the daily log, the delivery record, the photograph set for closed walls.
One month, because two weeks measures novelty and a quarter buries the result in everything else that changed.
One decision at the end, made by the foreman, not the office. If the foreman would not choose it, the rollout is dead and the only question is how much it costs to find out.
Construction already knows how to teach a procedure to people who did not ask for it. The Canadian Centre for Occupational Health and Safety publishes an orientation checklist built around areas to be covered, a description, a completion tick, and space for the worker’s and supervisor’s signatures for record-keeping — and notes that not all training can or should be done on the first day, and that the checklist should be customized to the workplace. That shape works for a tool rollout: a short list of what the person must be able to do, demonstrated rather than described, signed by both.
In British Columbia, employers are responsible for providing worker instruction and training under section 21 of the Workers Compensation Act, with the OHS Regulation requiring mandatory education, training or certification for specific topics. AI tools are not on that list, but the training culture that exists to satisfy it is the one your rollout should borrow.
Crews are quicker than head office to spot the risk here, and they are right to be. Say out loud that the tool drafts, records, transcribes and reminds — and that it does not decide whether work is safe, whether a hazard has been controlled, whether a condition is acceptable, or who is qualified to do a task. The CCOHS material on job hazard analysis describes that work as something people do by breaking a job into steps and identifying hazards in each; software can hold the record of it.
Qualification in particular is not transferable. Skilled Trades Ontario sets out the compulsory trades and the certification required to work in them, and the Electrical Safety Authority is explicit that only a Licensed Electrical Contractor with a seven-digit ECRA/ESA licence number may be hired to perform electrical work in an Ontario home, with certified electricians and Master Electricians working under that licence rather than being hired directly.
Skilled Trades Ontario describes apprenticeship as spending about 80% of the time learning directly from skilled tradespeople across more than 140 trades. That has a blunt implication for any tool rollout: whatever the journeyperson does, the apprentice will do. If the senior tradesperson works around the new tool, the next generation of your crew learns the workaround, not the tool. Train the journeypersons first and visibly.
Timesheets, location data, site photographs and voice recordings are personal information about workers, and the Office of the Privacy Commissioner’s fair information principles include identifying purposes, limiting collection and limiting use, disclosure and retention. Crews assume the worst about monitoring unless told otherwise, and an unstated purpose is the fastest route to a tool being quietly defeated. Our sister firm covers employee privacy and personal information for Ontario employers.
A framing contractor in Surrey introduces photo-based site records. The first attempt fails: a new app, a separate login, and a request for a written summary at the end of the day. Uptake collapses in two weeks and the office concludes the crews are resistant.
The second attempt changes three things. The photograph step is attached to an existing habit — the pre-close walk that already happened before drywall. The summary is generated from the photographs and the day’s log entries and appears in the foreman’s queue for a yes or a correction, so the writing burden moves from the site to a thirty-second check. And the rollout starts with one crew whose lead had complained loudest about paperwork.
Six weeks later the same crews are using it, and the reason is not enthusiasm. The pre-close photographs settled a dispute about a service penetration, and the foreman told the story at a toolbox talk. That is what adoption looks like on a site: a tool that visibly saved somebody an argument.
Mandatory works for a step inside an existing process and fails for a separate activity. “The daily log is now completed in this app” is enforceable. “Everyone should be using AI” is not a rule, it is a slogan.
Design for voice, photographs and taps in that order, and treat typing on a site as an accessibility problem rather than an attitude problem. Most of the useful site inputs are not text anyway.
Have a stated correction path before launch, and make correcting easier than ignoring. A record that cannot be fixed quickly will be routed around, and then you have neither the tool nor the record.
One month of one crew tells you about adoption. A quarter tells you about value. Do not confuse the two, and do not extend a pilot because you dislike the answer it has already given.
A 30-minute call is enough to tell you whether AI pays for itself here.
