← HR strategy

HRM practice

Learning evaluation: from a good session to changed work

Use clear evidence of learning and application, while recognising what an evaluation can and cannot attribute to training.

ADM HR; reading: Armstrong & Taylor · 2023 · Original editorial explanation
Learning evaluation: from a good session to changed work — simplified model sketchBaseline · Learning evidence · Workplace use · Outcome review. Editorial interpretation, after ADM HR; reading: Armstrong & Taylor (2023).Baseline01Learningevidence02Workplaceuse03Outcomereview04ADM HR · ORIGINAL INTERPRETATIONADM HR; reading: Armstrong & Taylor · 2023
Original ADM HR application sketch. Read the explanation for assumptions and limitations. Download SVG ↗

The core idea

Learning transfer means applying learning in the workplace. Evaluation examines whether a learning intervention was useful and what changed. Armstrong and Taylor discuss evaluation within the wider learning process. Donald Kirkpatrick’s familiar categories—reaction, learning, behaviour and results—help distinguish different questions. They are not automatic causal steps: an enjoyable course need not change behaviour, and improved business results may have other causes.

Source and attribution [1]

Using it in practice

Before an intervention, specify what people should be able to do and the evidence that would demonstrate it. Capture a proportionate baseline, provide realistic practice and arrange a follow-up task. Ask about barriers to using the learning. Where possible compare changes with a credible reference group or earlier pattern, while recording other changes that could explain the result.

An example, not a reported case

Worked example · illustrative

A team attends incident-reporting training. Instead of relying on satisfaction scores, the evaluator compares anonymised sample reports against a rubric before and after practice, then reviews reports from subsequent shifts. Better reports coincide with a redesigned form, so the team reports a combined improvement rather than claiming training alone caused it.

What to watch for

Measurement can be expensive and intrusive. Match the evaluation to the decision and risk. A financial ROI calculation depends on defensible attribution and cost assumptions; not every worthwhile learning activity can or should be converted into a precise monetary return.

Decide what the evaluation will help you choose

Start with an actual decision: whether to repeat a programme, change its format, extend it to another team or stop it. Write down the uncertainty behind that decision. A pilot may need to establish that employees can use a new method safely; a mature programme may need evidence that it remains worthwhile. This prevents the evaluation becoming a large survey with no consequence. Name the decision owner and when the evidence is needed. Match the effort to the decision’s importance: a short team exercise does not usually justify an elaborate financial model, while an expensive organisation-wide programme needs more than participant satisfaction.

Separate learning, opportunity and work outcomes

Keep three questions visible. Can the person demonstrate the capability? Have they had a reasonable opportunity and support to use it? What changed in the work? A supervisor may perform well in a practice conversation yet return to a workplace where their manager cancels every one-to-one. That is a transfer barrier, not proof that the supervisor failed to learn. Conversely, a good operational result may reflect a new system rather than the course. The existing Kirkpatrick overview is a useful way to organise evidence, but moving through its categories does not by itself establish a causal chain.

Write a baseline that another person can repeat

Specify the measure, denominator, period and inclusion rules before collecting the follow-up result. For handovers, count incomplete records out of all eligible handovers, using the same definition at both points. Record who checks the sample and how disputed cases are resolved. Include a balancing measure, such as time taken or missed urgent escalations, so that a better completion rate does not conceal slower or less useful work. Record concurrent changes in staffing, demand and tools. Without these notes, an apparently precise percentage can compare two different processes.

Worked example: an improvement that is not yet proof

Invented example: a team has 24 incomplete handovers out of 200 before a combined coaching and form-redesign trial, and 12 out of 200 afterwards. The rate fell from 12% to 6%: six percentage points, or a 50% relative reduction. Both descriptions are correct, but neither establishes which part of the trial caused the change. A similar team falls from 10% to 8% over the same period. That comparison is useful context, not a controlled experiment: workloads, staffing and selection may differ. Report the counts, definitions and concurrent changes alongside the result. Do not claim that training alone halved errors.

Plan support as carefully as the session

Before delivery, agree a first application and a source of feedback with each manager. Afterward, ask about attempted use and obstacles rather than simply asking whether the material was useful. For a complex skill, observe a safe practice attempt, agree one improvement and allow another attempt. Keep the support proportionate; constant surveillance can undermine trust and consume the time needed for practice. Look for barriers shared by several learners. If all report the same contradictory target or inaccessible tool, assign an organisational action instead of issuing repeated reminders to individuals.

Make a cautious financial assessment

Estimate a financial benefit only where there is a defensible link to costs or revenue. Saved minutes are not automatically cash savings: explain whether time was redeployed, overtime avoided or paid capacity actually reduced. Include design, delivery, cover, participant time and follow-up in the cost picture. Show a range when assumptions are uncertain and separate observed amounts from estimates. A safety, inclusion or quality initiative can have a sound rationale without an invented return-on-investment percentage. Use the logic-model and pilot-evaluation guides to explain the proposed pathway and its limitations before choosing a financial headline.

Report mixed results and agree the next decision

The close-the-loop tool in Laura Overton’s CIPD member business-case collection encourages follow-through on the evidence promised to sponsors. An original reporting example is: practice assessments improved, only half of learners had an opportunity to apply the method, and the service outcome is still inconclusive. Recommend removing the opportunity barrier before purchasing another cohort. State the owner, a review point and what will be checked next. Retain disappointing findings alongside successful ones. A report is more credible when it distinguishes a delivery success from an untested claim about business impact.

Check whether improvement lasts and travels

A result immediately after supported practice may not survive normal workloads. Choose a later check that fits how often the task occurs, rather than applying a universal thirty-day rule. Ask whether the behaviour continues without the trainer, and whether the method works on different shifts or more difficult cases. Review who benefited and who lacked access to support. Before scaling, document which conditions appear necessary and what another manager must provide. Treat replication as a new test, not an automatic entitlement to the same result. Record adaptations so future comparisons remain interpretable.

Take it into your next conversation

Three useful questions.

  1. Which decision will change because of this evaluation?
  2. Are the baseline and follow-up measuring the same thing?
  3. Could a process or staffing change explain the result?

Related terms

Go to the evidence

Sources & attribution

Further reading: Michael Armstrong with Stephen Taylor (2023), Armstrong’s Handbook of Human Resource Management Practice, 16th edition, chapters 37–39. Edition & reading notes ↗

Further reading: Nadeem Khan and Dave Millner (2023), Introduction to People Analytics: A practical guide to data-driven HR, 2nd edition, PDF pp. 227, 244 and 294–295. Edition & reading notes ↗

[1] Armstrong & Taylor (2023), 16th edition, chapters 37–39 — reading notes ↗

[2] Laura Overton / CIPD, L&D tools: Building your business case — Close the loop member tool; consulted 28 September 2026 ↗

[3] CIPD, Learning needs analysis — linking diagnosis and evaluation; consulted 28 September 2026 ↗

The core idea is an original summary of the cited work. Application notes, examples and sketches are our interpretations, not quotations or reproductions of the authors’ figures. Publisher records may require access to read the full original work.

Original ADM HR teaching guide with attributed background reading. Practical sections expanded on 28 September 2026 following the CIPD and AIHR reading listed in the sources. Worked examples, suggested review periods and decision aids are original editorial illustrations, not validated models or universal standards. Member resources are linked for further reading; their documents and templates are not reproduced here. The original overview review date is retained; this expansion is not a fresh legal review of every linked article.

Published 2026-09-25 · Reviewed 2026-09-25. Editorial approach

Keep exploring

A little more context.

People analytics · models & decisions01

Logic models for HR interventions

Separate resources, activities and deliverables from the changes an initiative is meant to produce.

Programme-evaluation tradition6 min
People analytics · technique02

A/B tests and pilot evaluation in HR

A pilot explores feasibility; an impact evaluation asks what changed because of an intervention.

Statistical and evaluation practice5 min