What is a good training completion rate?

Somebody will ask for a number

Somebody in your organisation is going to ask what a good completion rate is. They will ask it in a steering meeting, and they will want a number they can put next to yours.

The honest answer is that there is no credible published benchmark for frontline training completion in Belgium or the Netherlands, and the figures in circulation are almost entirely vendor marketing. "We do not know" is not an answer you can take back to the meeting on its own, so this guide is about what to do instead.

Where the circulating figures come from

Follow any completion percentage you have seen in this market back to its source and you land in one of three places.

• A vendor case study, where the customer was chosen because the number was good, the denominator is undefined and the period was the launch quarter.
• A self-reported survey of L&D professionals, from organisations that mostly have desks.
• A slide with no attribution at all, a number invented once and passed along ever since because nobody wants to be the person who asks.

No national statistics office collects this, and no sector body publishes it. Neither the Arbowet nor the Codex asks anyone to report a completion percentage to anybody, so the data does not exist as a regulatory by-product either. Where compliance figures are collected in this market, they are about accidents and inspections, not training throughput. Nor is this a gap somebody will fill next year: completion is defined differently by every organisation that measures it, which would make an aggregate meaningless even if somebody did collect one.

A borrowed number answers a question about somebody else

Suppose a credible figure did exist. It would still be close to useless to you.

A completion rate is a ratio, and both halves are decided locally. The numerator depends on what counts as complete: opened, finished, or finished with the knowledge check passed. The denominator depends on who was assigned, which depends on how clean your employee list is, which depends on how quickly leavers are removed. Two networks can run identical operations, instruct their people equally well and report figures twenty points apart purely on those two decisions. Comparing across that boundary compares accounting policies, not performance.

Then there is what surrounds the number: whether the training gates access to the floor, whether the workforce turns over three times a year, whether people have reliable device access, whether managers push it or tolerate it. Change any one and the achievable ceiling moves.

So a borrowed benchmark is worse than none, because it invites the wrong conclusion in both directions. Sitting above it, you stop looking. Sitting below it, you go looking for a problem in the operation when the difference is in the definition.

Building your own zero point

Stop trying to find a number and start producing one. A baseline is a measurement of your own operation under conditions you wrote down, and it takes one cycle.

That result is your zero point. It is not good and it is not bad. It is the number your current process produces, and every later figure means something only against it.

Do this once per type of training, because the types behave differently. A mandatory instruction that decides who may go on the floor, a change rolled out across the network with a two-week window, an induction that follows a starter's first shifts and a recurring refresher are four different processes. One target across all four will be wrong for at least three of them.

Comparisons inside your own network

Once you have a baseline, the comparisons that mean something are all internal.

Each of these has an intervention attached. That is the test of whether a comparison is worth running.

Four things that quietly move the number

Before you conclude anything from a figure, rule out the four ways it misleads.

• Assignment hygiene. People assigned content that does not apply to their work. It is the largest distortion in most networks and it is invisible, because nothing on the screen says the assignment was wrong. The fix is a corrected assignment, not a chase.
• Leavers in the denominator. With real turnover, a stale employee list holds a permanent block of people who will never complete anything because they no longer work there. The number is wrong in the direction that makes you chase ghosts.
• Structural incompletes. Somebody on long-term leave, somebody who left in week two of their induction. Their record is incomplete for reasons unrelated to training. Report them separately or you are reading a retention figure as a training figure.
• Definition drift. Somebody changes what counts as complete, or how the denominator is built, and the trend line breaks without anyone noticing. Write the definition down and date it.

The bottom of the distribution is the job

The network average matters less than people think, for a practical reason: nobody is ever audited on an average. An inspector or an auditor asks about a specific site, a specific person, a specific date. If your network sits comfortably high and three sites sit far below, the network figure has told you nothing about your exposure, which is concentrated in those three. Under article 8 lid 4 of the Arbowet the duty to supervise compliance does not average out across locations, and an auditor on one site is not interested in the other forty.

So sort ascending and work from the bottom. The bottom three sites this week are the intervention list; the number at the top of the screen is context. When a site sits at the bottom repeatedly, treat it as a diagnosis rather than a performance issue. The causes are mundane: nobody was assigned, the assignment was wrong, the rota leaves no window, the shared device is broken, or the manager was never told it mattered. Four of those five are fixed at head office. Tracking completion across 50 locations sets out the weekly loop.

If a number has to go on the slide

Sometimes you need a target to plan against. Set one, and label it for what it is: an internal target chosen by you, not a benchmark, not data and not evidence. Derive it from your own baseline plus the improvement you think your changes are worth, and review it after one cycle.

If a figure appears on this site or in a conversation with us without a source, read it the same way: as an illustration. And remember that completion is only the first layer: whether the training worked is a separate question, covered in measuring training effectiveness beyond completion.

Measuring your own baseline in Aristotl

Aristotl records each completion per person, per site and per version of the course as the training happens, and shows completion per site, region and person in one view. The internal comparisons above become filters rather than projects: the same assignment, site against site, over the period you choose. Assignment is by site or team from one central library, which is where most assignment-hygiene distortion disappears, and a baseline defined in January can be read the same way in September.

Pick a rollout that is already finished and work out its real denominator, which sites sat at the bottom, and what the honest baseline for your network turns out to be.