Your training worked. You just cannot prove it

A vice president of L&D walks into a review with three slides. Attendance: 94%. Satisfaction: 4.3 out of 5. Feedback: "the trainer was excellent." The CFO listens, thanks her, and asks the only question that matters: what changed in the business?

Nobody in the room has an answer, because nobody measured anything before the training started.

This is the most common failure in corporate learning, and it is not a training failure. It is a measurement failure, committed weeks earlier, usually by a well-meaning person who was told to "just get the programme running." By the time the review arrives, the evidence that would have settled the argument does not exist and cannot be reconstructed.

The uncomfortable reason impact stays unprovable

Training is judged like an event because it is bought like an event. A vendor quotes a day rate. A calendar slot is booked. People attend. The invoice is paid on delivery, not on outcome. So the paperwork that accumulates around training describes logistics: dates, rooms, headcount, hours, satisfaction scores. All of it answers "did this happen?"

None of it answers "did this work?"

There are three specific things missing in almost every programme we review:

1. A baseline. If you never measured the capability before the programme, you have no starting point. A post-training score of 68 means nothing on its own. It could be an improvement on 40, or a decline from 75.
2. A comparison point. Even with before and after numbers, something else may explain the change: a new manager, a process fix, a seasonal shift, a kinder client mix.
3. A business metric chosen in advance. If you decide after the fact that this programme was about "customer satisfaction," you will find a number that rose and call it impact. That is not measurement, it is selection.

Fix those three and the CFO conversation changes completely. Not because you become cleverer at statistics, but because you stopped leaving the evidence until the end.

Measure in the order the work happens

The order matters more than the sophistication. Most teams try to start at the top of this list, with the outcome, and work backwards into a fog.

Two rules keep this honest.

Rule one: nothing gets measured after the fact if it could have been measured before. If the programme is about delegation, ask managers and their reports to rate delegation behaviour before the first session. It takes ten minutes and is the difference between a claim and a finding.

Rule two: pick the business metric at design time, in writing. One metric, not five. "Escalations per branch per month." "Rework hours on the onboarding queue." "First-response resolution rate." Write it into the programme brief next to the learning objectives. If you cannot name the metric before you buy the training, the programme is not ready to run.

The method that works without a control group

Randomised control groups are the gold standard and are almost never available inside a company running a real programme. You cannot tell half the sales team they are not getting the training and hold the other half back for a year.

What you can do is compare expected with actual.

Take a cohort of twelve branch managers. Before training, measure each one on the competencies the programme targets, using a 360 and a work-sample assessment rather than self-report. That gives you a starting profile per person, not a single average. Then forecast, in writing, what a meaningful change would look like. For example: "average customer-escalation handling score should move from 2.8 to 3.5 by the end of the quarter; the four lowest scorers should gain at least one full point."

Now run the programme. Measure again at 60 and 90 days, with the same instrument, the same raters where possible, and the same scoring rules.

The comparison is no longer "was the training good?" but "did this cohort move the way we said it would?" That is a defensible claim, and it survives a finance review because the prediction was made before the result was known.

If the cohort moves and a comparable group that was not trained does not move, you have something close to causal evidence. If they both move, you have a business-wide trend and an honest finding: the training may have contributed, but it did not cause the shift alone. Saying that out loud earns more credibility than a slide of unverifiable wins.

Kirkpatrick without a research team

The four-level model is well known and badly used. Level one, reaction, is easy and nearly worthless on its own. Level four, results, is what executives want and what most teams cannot attribute. The useful middle is level three: behaviour on the job.

Three practical instruments get you to level three without a research department:

• Multi-rater feedback, repeated. The same eight questions, asked of the same manager, peers and reports, before and 90 days after. Blind spots show up as disagreement between self and others, which is more actionable than any single score.
• A work sample. A short scenario scored against a rubric. Cheap to run, hard to game, and it does not depend on the learner's self-awareness.
• Manager observation, structured. Five behaviours, a four-point scale, once a month. This is where most programmes die, because nobody chases the form. Automate the reminder and make it a two-minute task.

Notice what is missing: the smile sheet. Keep collecting reaction data, because a hostile room tells you something about design. Just stop presenting it as evidence of impact.

What to stop doing this quarter

• Stop reporting attendance as a success metric. It is a logistics number. Report it as coverage if you must, never as outcome.
• Stop running programmes with no pre-measurement. If the baseline is missing, run the assessment at the start of the first session and label the result honestly as a starting point, not a before-and-after.
• Stop accepting "we will evaluate at the end" in a vendor proposal. The evaluation design is part of the proposal, and a vendor who cannot describe the baseline instrument has not thought about impact.
• Stop choosing the success metric after the results arrive. It reads as spin, and experienced reviewers can tell.

A six-week sequence you can run now

1. Week 1. Pick one live programme, not the whole portfolio. Name the business metric and the target movement in one sentence.
2. Week 2. Choose the instrument for the baseline. Multi-rater for behavioural programmes, work sample for technical ones, and a cognitive or EQ assessment where the programme is about capability rather than knowledge.
3. Week 3. Measure the baseline. Publish the starting profile to the sponsor before training begins. This single act changes the tone of the final review.
4. Week 4. Run the training as planned. Capture reaction data, but do not report it as impact.
5. Week 5 to 6. Re-measure at 30 days. Send managers the two-minute observation form.
6. Week 12. Re-measure at 90 days. Present expected versus actual, including the parts that did not move.

The last point is the one most teams skip and the one that builds the most trust. A report that shows two competencies improving, one flat, and one declining is far more persuasive than a report where everything improved. Reviewers believe the second one less, not more.

The honest limit

Not everything can be attributed. Culture programmes, leadership offsites and long-horizon development often produce effects that show up years later, tangled with promotions, reorganisations and market conditions. Pretending otherwise is how L&D loses credibility.

The professional answer is to say so at the start: "we can measure capability change and behaviour change with confidence; business result will be directional and we will label it as such." A sponsor who accepts that framing will trust the numbers you do produce.

Measurement is not a report you write at the end. It is a decision you make before the programme starts, and it costs about ten minutes per cohort. The only reason it does not happen is that nobody put it in the brief.

Put it in the brief.