Did the training work? Measuring effectiveness beyond completion

Completion answers the wrong question

A completion rate tells you the training was delivered. Everyone assigned finished the current version by the deadline, and the record shows it. That is worth having, and tracking it across sites is its own discipline. But it answers a delivery question. Nobody who asks whether training worked is asking about delivery.

The question comes from three directions. An auditor on a food site asks it because the standard does: IFS clause 3.3.3 wants a documented procedure that demonstrates training was effective, and BRCGS clause 7.1.2 wants documented competence for anyone on critical control points. An inspector asks it because the law is written as a result: the Dutch Arbowet asks for workers to be effectively informed and instructed, and the Belgian Codex (art. I.2-21) for training that is sufficient and appropriate to the workpost. And the operation asks it after an incident, when the file shows a completed course and the floor showed something else.

Clause numbers in this guide are as of September 2026: IFS Food version 8 and BRCGS Food Safety Issue 9.

Three layers: delivered, understood, done

Effectiveness is not one number. It is three layers of evidence, each answering a different question, each with its own owner.

Readers who know the Kirkpatrick model will recognise the shape: reaction, learning, behaviour, results. For frontline operations the first level adds little and the fourth is usually an incident or audit trend you cannot attribute to one course, so the working layers are the middle ones plus the record underneath.

Understood: make the knowledge check measure something

A knowledge check is only evidence if its questions test the decision the person has to make on the job. "What is the critical limit for the chill-down step?" tests recall. "The probe reads 9 degrees after the chill-down step. What do you do?" tests the decision, and that is the one the auditor will ask about.

The last item is the one most programmes skip. When most of a site gets the same question wrong, the person is rarely the issue. Either the course explains that step badly, or the procedure itself is unclear, and both are fixed at the source, not by retraining. How to build the questions is covered in knowledge checks that actually stick.

Done: evidence from the floor

The third layer cannot come from the training system alone. It comes from watching the work.

For critical tasks, observation is the most defensible method: the operator performs the check, reads the instrument, reacts to a deviation and records it, while a named colleague watches and dates it. That is what BRCGS 7.1.2 asks for on CCPs, and it is a good standard for any task where a mistake hurts someone. Write down who may observe, what they look for and what happens when the task is not done right.

For everything else, site-level evidence such as checklist results and corrective actions shows whether the training shows up in the work; the delivery record is not the floor covers how to read it.

You cannot observe everyone on everything. Sample: every CCP operator after training and at the interval you set, a few people per site per quarter for general tasks, and always after an incident or a procedure change.

Reading the gaps between the layers

The value is in the comparison. Each gap points at a different cause and a different owner.

The second row is the one to watch. Retraining people who already know the answer does not move the floor, and it burns the credibility of the next course. The pattern behind it, sites drifting from a procedure everyone was trained on, is the subject of operational drift.

Write the method down

IFS 3.3.3 asks for a documented procedure, so the method itself is part of the evidence. Keep it short and specific.

Then keep to it. As with refresher intervals, the auditor measures you against the method you wrote, not against an ideal one.

How Aristotl handles it

Aristotl covers the first two layers and feeds the third. Every course of 3-5 minutes carries a knowledge check, built from the same procedure and translated per person from one master, so what you hold per person is a result tied to a version, not a signature. The platform shows completion per site, region and person, and drop-off per module shows where people stop, which is often where the content is unclear.

On the floor, checklists run per site and per day with their proof, exceptions and corrective actions, and the day log exports as CSV or PDF for the store visit or the auditor. A certification can combine a course, a checklist completed on the floor and an external certificate into one qualification, so the observed task sits next to the knowledge check rather than in a separate folder.

What it does not replace: the colleague who watches the task. Observation stays with the line, and the method stays yours.

Take one procedure that matters and put its three numbers side by side for one site: who completed it, who passed the check, and how often the task was seen done right last month.