By Justin Rose, Founder, JDR Metal Art
Ask most organisations whether their safety training actually works and somebody will go and find a spreadsheet. Names down the left hand side and a long wall of green cells showing that everyone is in date.
It’s quite reassuring to look at. The trouble is it answers a different question to the one that really matters.
All a completion record tells you is that somebody sat through the course. It doesn’t tell you if the forklift driver now sounds the horn at the blind corner by goods-in, or whether the new lad on the press line properly isolates the machine before he clears a jam when the shift is running 40 minutes behind. Those are the things the training was paid for in the first place and they’re exactly the things most of us find hard to measure.
The step everyone agrees with and nobody plans for
A brief guide to health and safety training lays out five steps and the last of them is simply to check the training has worked. Hardly anyone would argue with that. Far fewer put the same effort into it as they put into getting the course booked.
Some of that is because it’s genuinely hard to do. Research the Association for Talent Development put out last year found 90% of the people it surveyed saw isolating a programme’s impact on results as a challenge, and only 15% felt ready to deal with it. These are people whose whole job is training. Safety teams, who tend to be commissioning courses on top of everything else they’re juggling, are starting from a fair bit further back.
The other reason isn’t quite so flattering. Attendance is easy to count and the number only ever goes up. Behaviour is a lot messier, it shifts slowly and every now and then it tells you a course didn’t really land at all. Nobody particularly wants to put that in a quarterly report.
Decide what “different” looks like before anyone books a room
Most of the useful measurement work actually happens before the course has even started.
If you can’t say what people ought to be doing differently on a normal Tuesday afternoon three months from now then you won’t be able to measure it (and to be fair, the trainer probably won’t be able to teach it either).
- “Improved awareness of manual handling” is far too vague to observe.
- “Two person lifts for the long pipe sections on the delivery bay” is something you can go and look for, and so is “permits signed back before an area gets handed over”.
Two or three of those per course is plenty.
Then comes the bit people tend to skip, which is finding out how often those things happen now. Without a baseline you end up comparing behaviour after the training against a hazy memory of how bad things used to be, and that memory has a habit of being kind to everyone involved.
In the Kirkpatrick model this is Level 3 (behaviour on the job). It’s the level organisations struggle most to put numbers on, which is why so many of them stop at Level 1: the feedback form asking whether the trainer was engaging that gets handed round just before everyone heads off to the car park.
Watching the work, and the trouble with watching
Observation is the obvious way to find out whether the training has carried over into the job. The catch is that it’s very easy to fool.
People work differently when the safety manager is standing at the end of the aisle holding a clipboard. Most of us would. So one planned observation the week after the course mostly proves that people know what the safety manager looks like.
A few things help with this:
- Spread your observations out across shifts (nights and weekends included) because that’s when supervision is thinner and people’s real habits come through.
- Get trusted peers to do some of the observing, since a colleague from the next team along gets a lot less of a performance than somebody down from head office.
- Bring in some evidence that doesn’t rely on anyone standing there at all, like the quality of the permits that come back or defect reports showing somebody picked up on a problem early.
The best method is often the simplest one. Ask someone to talk you through the task and listen to the reasons they give you. If they can explain why the isolation sequence goes in that order then they’ve learned something. If they can reel off the steps but go quiet when you ask why, there’s a good chance they’ll skip one when the pressure’s on.
Listen for it in ordinary conversation
A lot of the best evidence turns up when nobody’s talking about training at all. After a decent working at height course, you might hear a scaffolder query a ladder that’s been “fine for years”. Near-miss reports start getting more specific, the cracked bracket on line two or the valve that sticks after a cold start. Toolbox talks can get a bit more argumentative too, which is normally a good sign.
Keep an eye on the stop-work count
Stopping a job is about the clearest signal you’ll get. If people have been trained to spot a hazard and told they’ve got the authority to stop work, then how often they actually use that authority tells you whether they believed the training or just sat through it.
If the stop-work count on a busy site is sitting at zero I’d want to know why.
Check in on supervisors
Supervisors make a huge difference here. A course can teach the right way of doing something in a morning and a team leader saying “we haven’t got time for that today” can undo it all by lunch.
HSE puts its training guidance alongside advice on supervising workers for good reason. If you want an early idea of whether the training is going to stick, keep an eye on what supervisors do in the fortnight after it.
Check again later, because skills fade
Checking a week after the training mostly tells you about short-term memory. What really matters is what’s left six months down the line when nobody has mentioned the course since.
Skill fade shows up pretty much wherever anyone’s bothered to test for it. A 2025 review of school-based CPR training found compression quality dropped off markedly within 6 to 8 months. CPR just happens to get studied closely enough to track that sort of decline, and there’s not much reason to think rarely used workplace skills (spill response, say, or rescue from height) hold up any better.
Light check-ins at around 30, 60 and 90 days give you a feel for the drop-off. They also tell you something quite practical, which is when refresher training is actually needed, and that’s rarely the same date the certificate runs out.
When nothing changes
Sometimes you do all of this properly and the behaviour still doesn’t move. That’s frustrating, but it doesn’t automatically mean the course was poor.
No amount of training will fix a procedure that can’t be followed with the kit people have been given. It won’t get around a bonus scheme that rewards speed, or a rota that leaves one person doing what’s really a two-person job. When people clearly know what they should be doing and still aren’t doing it, the obstacle is usually somewhere in the work itself. Sending them on the same course again will mostly just teach them that nobody was listening.
This is where measuring properly earns its keep. Honest data on how people behave after training is one of the few reliable ways of telling a learning problem apart from a design problem, and those two need very different fixes.
So, did it work?
None of this needs a big platform or a team of analysts. Mostly it needs a clear picture of what should change and the patience to keep checking long after the certificates have been filed away.
Next time somebody asks whether the training worked, try answering with something you’ve actually seen people doing. It takes more effort than pulling up a completion rate, and now and then the answer won’t be a comfortable one. That’s usually a sign you’re measuring the right thing.
About the Author: Justin Rose
Justin Rose is a gate designer, custom metal fabricator, and the founder of JDR Metal Art; building high-end custom driveway gates and automated entry systems for over 500 homes, farms and ranches across the US since 2011. He combines a deep understanding of advanced automation technologies, precision welding, and custom powder coating finishes to create durable, heirloom-quality entrance solutions for estates, ranches, and residential properties across the United States.
References
- Health and Safety Executive (2012). Health and safety training: A brief guide. Available at: https://www.hse.gov.uk/pubns/indg345.htm
- Association for Talent Development (2025). ATD Research: Organizations Struggle With Measuring the Impact of Training. Available at: https://www.td.org/content/press-release/atd-research-organizations-struggle-with-measuring-the-impact-of-training
- Training Industry. The Kirkpatrick Model. Available at: https://trainingindustry.com/wiki/measurement-and-analytics/the-kirkpatrick-model/
- Health and Safety Executive (2012). Health and safety training: Provide information, training and supervision. Available at: https://www.hse.gov.uk/simple-health-safety/training/needs.htm
- Onbasilar, U., von Groote, T., Brülle, R. et al. Skill retention after school-based CPR training – a systematic review and meta-analysis. BMC Public Health 25, 3798 (2025). https://doi.org/10.1186/s12889-025-25024-w
