Training Feedback Survey in India: Questions, Timing and a Training Effectiveness Score

The four-level model was created by Donald Kirkpatrick in the 1950s and is maintained today by Kirkpatrick Partners, whose definitions are used below (Kirkpatrick Partners, The Kirkpatrick Model).

Two points from the same source change the design. First, relevance is the Level 1 item that matters most: the page says that if a programme is not relevant to a participant, application will not occur. Second, Level 3 measurement cannot wait until 90 days after learning, because too long a gap removes the chance to adjust. So ask about relevance on day one, and collect the first behaviour read at 30 days.

Start the design from Level 4. Write down the business outcome the programme must move, then the behaviours that drive it, then what people must learn, and only then the session.

Give this at the end of the session, on a phone, in under three minutes. The scale is five-point agree to disagree unless the table says otherwise. Every item asks about the session that just ended, not about training in general.

Questions 1, 2 and 5 are the relevance items. Report them separately from the trainer items, because a liked trainer can mask an irrelevant programme. Question 7 matters when a programme written in English is delivered to a sales team that works in Tamil or Marathi.

Kirkpatrick Partners describes Level 2 as traditionally measured with pre and post tests, and adds confidence and commitment as early predictors of what happens back at work (Kirkpatrick Partners). So Level 2 has two parts: a short knowledge or skill test, and a few self-report items.

For the test, write five to ten questions tied to the learning objectives, with one correct answer each. The same test goes out a few days before the session and again immediately after. Score each person as correct answers out of the total, then report the share who pass a threshold set in advance, such as seven out of ten, before and after.

Ask items 11 and 12 both before and after. A rise in confidence with no rise in test score is a warning sign, not a success. Item 13 predicts what the 30-day survey will find.

Send the 30-day survey to the participant and a matching short survey to the participant's manager. Both go out again at 90 days with identical wording. Name the behaviour in the question.

Items 18 and 23 measure the same event from both sides. When they disagree, believe the participant. Kirkpatrick Partners lists performance support and the performance environment as components of Level 3 (Kirkpatrick Partners), and items 18, 19 and 23 are how you measure them.

Level 4 is rarely a survey question. Kirkpatrick Partners recommends identifying leading and lagging indicators that are usually already tracked somewhere in the organisation (Kirkpatrick Partners). For a sales programme, that is conversion rate by trained versus not-yet-trained teams. For a compliance programme, it is audit findings. Agree the indicator and the comparison group with the sponsor before the session runs.

What we would do. We would send the manager survey on the same day as the participant survey, not a week later. A manager answers when the question lands beside the one their team member is answering. If you can afford only one follow-up wave, make it the earlier one. The later wave tells you what stuck; the earlier one tells you what to fix while there is still time.

Boards and CHROs want one number per programme. Build it from shares, not averages, so it stays readable. Use four components, each a percentage of respondents:

• Relevance: share who agree or strongly agree with items 1 and 5, averaged.
• Learning gain: share passing the post-test threshold minus the share passing the pre-test.
• Application at 30 days: share answering "always" or "usually" to item 17 among those who said a situation arose (item 16 = yes).
• Manager-observed change at 90 days: share of managers answering "somewhat better" or "much better" to item 22, excluding "cannot say".

Score = (Relevance + Learning gain + Application + Manager-observed change) / 4. Illustrative numbers for one programme: relevance 80, learning gain 35, application 50, manager-observed change 40 gives a score of 51. The components matter more than the total. In that example, half applied it and fewer managers saw a change, which points to the performance environment, not the content.

Report each component with its base. A 30-day survey answered by a third of participants is weaker evidence than one answered by nearly all. Compare programmes only when thresholds, wording and timing are identical. The employee survey question bank explains why small wave-to-wave movements are usually noise, and the same caution applies here.

• Asking only at the end of the session. You learn whether the room was happy and nothing about use. Add the 30 and 90 day reads for every programme that costs more than a day of staff time.
• Collecting names on the reaction form. Participants rate a trainer they will meet again. Keep Level 1 anonymous, link Levels 2 and 3 with a self-generated code, and tell participants who sees what.
• Averaging the scale. A mean hides whether the result is many fours or a split between fives and twos. Report the share who agree, with the base.
• Skipping the manager. Without items 21 to 24 you have only self-report, and self-reported application runs high.
• Surveying in English only. Deliver the survey in the language the training was delivered in, and back-translate it.
• No comparison group for Level 4. Compare trained teams with teams not yet trained, or the same team before and after, and say which you did.

If you fix only one of these, add the 30-day participant survey. It is six questions, takes two minutes on a phone, and turns a satisfaction form into an evaluation.

Internal training feedback surveys usually go to the organisation's own participant list. Our panel suits the market version of this job, such as a survey of working professionals about training they received or want, found with a screener on occupation and sector, and this is what a pilot and main study cost at list price with respondents included. Screened-out respondents do not count towards a bundle.

Pro is sold in bundles of 400 to 1,000 responses and costs 20% less per response than Plus. Enterprise plans come with a dedicated research team and are priced on request.

The SuperJ network has 2 crore Indian respondents. The survey-eligible panel is a subset, and availability depends on your targeting. Members include working professionals, so name the role and sector you need and ask for a feasibility count. Members complete Aadhaar-based KYC once, and each study receives a zero-knowledge proof of verification. Surveys run in as many Indian languages as the study needs. Quality checks flag speeders, straight-lining and inconsistent answers. Read /panel-methodology and pricing.

The limit: the app reaches smartphone users only, so a factory floor without personal phones needs a kiosk or paper version.