What a training impact report should contain, line by line

The review is in a training room in Pune. Forty branch managers finished a five day programme on appraisal conversations three months ago. The business head asks the only question he cares about: did it work?

The L&D manager opens a twenty-two slide deck. Slide fourteen carries the number. Appraisal completion is up eleven points. The business head asks where the eleven came from. The answer is a spreadsheet that nobody else in the room has opened, built from a list that was assembled after the programme ended.

The number may well be true. It cannot be checked, so it buys nothing, and the next budget conversation starts from a weaker position than the last one. That is the cost of a report that describes work instead of proving it.

Why reports fail: they are written backwards

Almost every weak impact report is assembled in the wrong order. The result is discovered first, usually by asking the business what improved. The baseline is reconstructed second, from memory or from whatever data survives. The target is invented third, chosen so the result clears it.

The document then reads like a plan. It is a narrative with dates attached.

A reader's test is simpler than any methodology. Can I find every number on this page without the author in the room? If the answer is no for one number, the reader discounts all of them. A reviewer who cannot verify the baseline will not believe the result, and a reviewer who does not believe the result will not release the budget.

The second failure is structural. A report with no decision attached is a newsletter. It gets read, filed, and never referred to again, which is why so many L&D teams can produce twelve months of reports and still be asked to justify their existence each year.

The six lines

One page, six lines, plus a header block that names the programme, the audience and its size, the dates, the decision owner, and the date of the report. Everything else is an appendix.

Line 1: the baseline, with its address

Not "appraisal completion was low". A number, the system it came from, and the date it was pulled. If the baseline lives in a manager's head, it is not a baseline, it is an opinion with a decimal point.

The baseline is also the line that decides whether a report is possible at all. If it did not exist before the programme, the honest report says so on line one and names the date the baseline starts. That is a weaker report than the one you wanted. It is stronger than the one you would have invented.

Line 2: the target you set before you knew the answer

A target written after the result is a description of the result. The only thing that distinguishes a commitment from a rationalisation is its date. Put the date on the page. If the target moved mid programme, show both numbers and say why it moved.

The reason for the number matters as much as the number. "Eighty percent within thirty days, because the client contract penalises late appraisals" is a target. "Improve appraisal completion" is a wish.

Line 3: the result, with the counts that produced it

Same instrument as the baseline. Same definition of the population. Same length of window. Change any of the three and you are comparing two different things and calling it progress.

Then show the arithmetic. Forty trained. Thirty-eight in scope at the review date. Thirty-six measured. Two on long leave, excluded, named as excluded. A percentage without its denominator is a rumour.

The scale of the measurement problem is not small. ATD's evaluation studies found 79 percent of respondents considered level 3 evaluation valuable in 2019 while 54 percent reported using it, and the same gap appeared in 2016 at 73 and 60 percent (Saunderson, 2026). Most reports circulating in Indian organisations are level 1 and level 2 data wearing an impact heading, which is why the audience has learned to discount them.

Line 4: the attribution caveat, written like an adult

This is the line that makes the other five believable, and the one most often left out because it feels like admitting weakness.

The pattern is a sentence, not a paragraph: "If we had not run this, completion would probably still have moved, because the workflow change on 1 September removed a manual approval step. The programme's share is between four and thirteen points, and the data we hold cannot narrow it further."

Note what that sentence does. It names the confound before the reader finds it, gives a range instead of a point estimate, and states the limit of the data. It also respects a finding that evaluation research keeps repeating: the four levels are not a causal chain, and a good reaction does not prove learning, just as learning does not prove changed behaviour (Saunderson, 2026).

Line 5: cost, including the hours nobody invoices

Fees are the easy part. Add participant hours at a loaded rate, facilitator preparation, the manager time spent on pre-work and observation, and the cost of the reporting itself.

Then divide. Cost per participant is a procurement number. Cost per unit of the outcome is a business number: cost per additional appraisal closed on time, cost per additional certification passed, cost per point of the metric you chose on line 2. That single figure is what a finance reviewer will repeat to someone else, so make it easy to repeat correctly.

Line 6: the decision

A report that ends with "in summary, the programme was well received" has no line six. Line six says: continue with cohort two, change the two lowest scoring modules, stop the coaching component, review on 15 January, decision with the HR head.

If nothing is being decided, do not write the report. Write a two paragraph note and spend the saved week on the next baseline.

A worked example, on one page

Numbers below are illustrative, from a services company that ran an appraisal skills programme for forty branch managers.

• Programme: appraisal conversations, 40 managers, 5 days, August to September.
• Baseline: 61 percent of appraisals closed within 30 days of cycle end. Source: HRIS closure report, pulled 8 August, before session one.
• Target: 80 percent by the end of Q3. Set 12 August, because late appraisals trigger a client contract penalty.
• Result: 74 percent. 38 in scope, 36 measured, 2 excluded on long leave. Same HRIS field, same 30 day rule.
• Caveat: the HRIS workflow change on 1 September removed one approval step and probably moved closure on its own. Programme share estimated at 4 to 13 points. Cannot separate further.
• Cost: Rs 6.4 lakh in fees plus 152 manager hours at Rs 1,100, total Rs 8.1 lakh, or Rs 20,250 per manager. Cost per additional on-time appraisal: about Rs 9,600.
• Decision: continue for cohort two, rewrite the two modules with the lowest observation scores, review 15 January with the HR head.

Ten lines of a text file. It survives a sceptical reader, because every number has an address and the weakest claim is stated by the author, not discovered by the audience.

When there is no business metric to point at

Some programmes have no clean number behind them: a first time manager programme, a safety culture workshop, a communication skills course for a plant. Inventing a metric here is worse than admitting the gap.

The success case method handles this honestly. Brinkerhoff's version has five steps: plan the study, build an impact model of what success would look like, use a short survey to find the best and worst cases, interview those cases in depth, then report findings and recommendations (Medina et al., 2015). The method exists precisely because it produces a defensible conclusion quickly without a research team (Brinkerhoff Evaluation Institute).

In the six line format, the result line becomes two named cases: one person who changed something specific and one who did not, each with a date, a behaviour and a consequence. Counts replace percentages when the group is under about thirty people. "Four of nine managers now open with the appraisal purpose" is a fact. "44 percent improvement" is noise dressed as precision.

The honest limit

This format needs a baseline that existed before the programme. Without one, the first report you write will be thin, and the correct response is to set the baseline now rather than to manufacture a comparison.

The caveat line can also be abused. A programme with no effect can hide behind "many factors were at play" for years. The discipline is that the caveat sits under a result, not instead of one. If you cannot state a result, say that plainly on line three and make the decision about whether to keep measuring.

One page is a constraint, not a virtue. A programme with three outcomes gets three pages, one per outcome, and a single page that summarises the decision. What it must not become is twenty-two slides where the number is on slide fourteen.

What to do before your next review

Pick one programme that is running right now. Write the six lines on a single page today, while the programme is still live. Any line you cannot fill is the finding, and it is worth more than the report you were going to write at the end, because you can still fix it. If line one is empty, the baseline starts this week. If line six is empty, you already know the report will be filed and forgotten, and that is the thing to change first.

The mechanics of setting that baseline before day one are in why your training worked but you cannot prove it, and if the programme is behavioural rather than technical, the observation form in running Kirkpatrick level 3 without a research team is what feeds line three. For the metrics that sit above a single programme, the six-line L&D scorecard is the next layer up. BolderBrain's assessment and reporting keeps the baseline, the instrument and the result in one place, which is the only reason any of this is repeatable rather than reconstructed.