Most learning evaluation fails in one of two directions. Either it measures nothing beyond whether people enjoyed the day, or it attempts a full return-on-investment calculation with an attribution assumption that collapses under the first serious question from finance.
There is a workable position between the two, and it starts by accepting that you will not isolate the effect of training from everything else happening in an operating business. You do not need to. You need evidence strong enough to justify continuing, changing or stopping.
Four questions, in order
Did they learn it? Assessed, not asked. A short competency assessment against the specific things the programme claimed to build. Ten well-constructed questions or one practical task, marked. If a cohort cannot demonstrate the capability at the end of the programme, nothing further needs measuring — the design is wrong.
Are they doing it? Observed in the workplace, six to twelve weeks later. Usually a short structured check by the supervisor against three to five specific behaviours. This is the step almost everybody skips and it is where most programmes are revealed to have failed, generally for environmental reasons rather than learning ones.
Did the operational measure move? Chosen before the programme runs, not after. Rework rate, first-time-right on inspections, time to competence, unplanned downtime on a defined asset group, CoC rejection rate. One measure per programme.
Was it worth it? A defensible comparison, not a precise ROI. Programme cost including participant time against the value of the movement in the operational measure, stated with the assumption visible.
Why the pre-commitment matters
The measure has to be agreed with the business owner before the first session. Two reasons. It forces the conversation about what the programme is actually for, which frequently changes the design. And it prevents the retrospective search for a metric that happened to improve, which is the most common form of self-deception in this field.
Write it down as a one-line statement: "This programme exists so that X moves from A to B by date C, as measured by the report the operations team already produces." If the business owner will not commit to that sentence, you have learned something valuable before spending any money.
Attribution, handled honestly
You will be asked whether the improvement was caused by the training. The honest answer is usually "partly, and here is what else changed." State the other factors. A comparison group is better where one exists naturally — two shifts, two plants, a staggered rollout across regions. Staggered rollouts are the single most useful measurement gift available and they are usually a scheduling accident rather than a design choice. Make them a design choice.
What to stop measuring
Satisfaction scores as a headline. They are useful for detecting a genuinely poor facilitator and useless for everything else.
Training hours and headcount as achievement measures. They are activity, and reporting activity is precisely what keeps the function in the cost centre category.
Completion percentages on self-paced content. A completed module is not a capability.
A realistic standard
You will not get a clean number, and you should stop trying. What you can get is a consistent record: this capability gap existed, this intervention addressed it, the people demonstrably acquired the capability, their supervisors confirmed they are applying it, and this operational measure improved over this period while these other factors were also in play.
That paragraph, produced consistently across a portfolio of programmes, is more persuasive in a budget discussion than any ROI multiple, because it can be interrogated line by line and it holds. Which is the actual test.