The Kirkpatrick four-level training evaluation model — reaction, learning, behavior, results. Definitions, examples, and sample questions per level for 2026.
The Kirkpatrick model is a four-level framework for evaluating training: Level 1 Reaction asks whether participants found the training engaging and relevant; Level 2 Learning asks whether knowledge or skill increased; Level 3 Behavior asks whether people apply it on the job; Level 4 Results asks whether an organizational outcome moved. Donald Kirkpatrick introduced the levels in 1959, and they remain the most widely used vocabulary in training evaluation.
The model is easy to recite and hard to run. The practitioner version of the problem: “We can quote our satisfaction scores from every session for three years, and I still can’t tell the board whether anyone works differently.” The levels are not the hard part — the record-keeping underneath them is, and that is what this page works through, level by level.
Key takeaways
Sopact calls the practice Continuous Kirkpatrick: the four Kirkpatrick levels run as one continuous loop on one trainee record — a persistent participant ID, assigned at enrollment, that carries the reaction pulse, the learning gain, the 90-day behavior evidence, and the tied result metric. Kirkpatrick supplies the level map; the Loop supplies the engine. The model stops being a year-end reporting framework and becomes the running structure of the program’s own data.
The differentiation is the data model, not the questionnaire. Form-centric survey tools treat every send as a fresh anonymous batch, so Level 1 and Level 3 can never land on the same person, and the ladder breaks between rungs however well the questions are written. A record-centric system never breaks the thread: the four levels become four views of one dataset. This page works the model itself; the full training lifecycle — needs assessment through delivery to ROI — is worked on training evaluation.
Donald Kirkpatrick published the four levels as journal articles in 1959 and consolidated them in Evaluating Training Programs (1994). In 2016, James and Wendy Kirkpatrick of Kirkpatrick Partners published the New World Kirkpatrick Model: plan backwards from Level 4, define leading indicators, and sustain Level 3 with Required Drivers — the reinforcement, monitoring, and accountability processes that keep new behavior alive after the course ends. The theory kept improving.
Tooling went the other way. The smile-sheet era (paper, then SurveyMonkey and Google Forms) made Level 1 free and made anonymous, disconnected responses the default. The LMS era (Cornerstone, Docebo, Moodle, TalentLMS) added completion rates and quiz scores — a Level 2 proxy that measures exposure and recall, not transfer. No mainstream tool owned Levels 3 and 4, because both require longitudinal, person-level records that neither a form tool nor an LMS keeps. The one test that separates the eras: ask to see one trainee’s enrollment baseline and their 90-day behavior evidence on the same screen, scored on the same rubric. A tool built on disconnected forms cannot show it.
Each card below takes one Kirkpatrick level and shows three things: how it is commonly measured today, the exact point where that practice breaks, and the same level run on Sopact’s Loop — collected clean at the source on a persistent trainee ID, read on arrival, acted on in time. Question wording for every level lives in the banks on training evaluation survey questions; this page stays on what each level means and what it takes to measure it.
Level 1 is the easiest level to measure and the easiest to over-trust: decades of research find that satisfaction barely predicts learning or transfer.
Level 2 is a subtraction, not a score — and a subtraction needs two measurements of the same person on the same instrument.
Level 3 is the level the whole model exists for, and the one the tooling era quietly abandoned. The full treatment of the transfer problem is on behavior change after training.
Level 4 closes the ladder. Which metrics belong at each level is covered in training metrics; the arithmetic that turns a Level 4 result into a percentage is training ROI.
A worked Kirkpatrick model example, from a twelve-week communication-skills cohort run on Sopact Sense: at intake, participant P-1247’s baseline read “I freeze in meetings” (Level 1–2); his exit assessment showed a 34-point gain scored on the same rubric as his baseline (Level 2); a mid-cohort mentor interview and peer ratings supplied on-the-job evidence (Level 3); and by cohort end he had presented at an all-hands (Level 3 evidence a funder can quote). At the program level, the same records supported a Level 4 read: a regression across the cohort found mentor minutes predicted skill gains more strongly than LMS module completions — a finding that redirected budget, and one that is only computable when every level sits on one persistent ID.
The pattern compresses to any vertical. Customer-service onboarding: pulse per module (L1), scenario scored pre/post on one rubric (L2), QA-observed call behavior at 60 days (L3), escalation rate for the trained cohort against baseline (L4). Safety training: relevance pulse (L1), hazard-identification test (L2), supervisor spot-checks (L3), incident rate (L4). The instruments change; the ladder and the ID do not.
Most organizations that adopt the Kirkpatrick model measure Level 1, sample Level 2, and never reach Levels 3 and 4 — not because upper-level data matters less, but because it demands longitudinal infrastructure: the same person, followed for months, on one record. Level 1 can be measured in the room. Level 3 lives a quarter later, in the workplace, after the survey tool has forgotten everyone’s name. When the follow-up finally goes out on a fresh anonymous link, response rates collapse and nothing can be matched back — so the evaluation quietly retreats to the levels the tooling can hold.
The academic criticisms point at the same weakness from the other side. The standard criticisms of the Kirkpatrick model are that it implies a causal chain the levels do not guarantee — good reactions predict neither learning nor transfer; that it under-specifies how to measure behavior and results; that it evaluates after the fact instead of shaping design; and that it ignores context and inputs. All four criticisms are fair, and all four sharpen to the same operational point: the model names what to measure and stays silent on the record architecture that makes measuring it possible. Fix the architecture and the model works as intended; leave it broken and no refinement of the levels will save the evaluation.
Apply the Kirkpatrick model in five steps: define the Level 4 result first and pick one operational metric for it; assign every trainee a persistent participant ID at enrollment; design one instrument per level, with the Level 2 baseline and exit scored on the same rubric; schedule the Level 3 follow-up at 60–90 days on the same ID; and read each level against the one below it, per person, not as cohort averages. Working backwards from Level 4 is the New World Kirkpatrick discipline; the persistent ID is what makes it executable by a program team instead of an evaluation department.
Step one usually exposes the real gap: if no operational metric can be named, the program has a design problem no survey will fix — start at training needs assessment instead. The per-level instruments themselves are a solved problem: the level-keyed banks on employee training survey questions supply the wording.
The Kirkpatrick model differs from its rivals by scope: Phillips ROI keeps Kirkpatrick’s four levels and adds a fifth that converts results to a return percentage; CIRO evaluates context and inputs before the training as well as outcomes after; and Brinkerhoff’s Success Case Method studies the most and least successful participants in depth instead of averaging everyone. The honest comparison is below — including where each one breaks.
| Model | What it adds | Where it breaks in practice |
|---|---|---|
| Kirkpatrick four levels (1959) | The shared vocabulary: Reaction, Learning, Behavior, Results — simple enough for every stakeholder | Assumes the upper levels get measured; most implementations stop at Levels 1–2 |
| New World Kirkpatrick (2016) | Plan backwards from Level 4; Required Drivers and leading indicators keep Level 3 alive | A planning discipline — still needs per-person data connected across months to execute |
| Phillips ROI | Level 5: net benefits over cost as a percentage, for the budget conversation | Inherits every join problem beneath it; isolating training’s effect is the hard part |
| CIRO | Evaluates context, inputs, and design — not just the aftermath | Thin on behavior; little guidance for measuring on-the-job transfer |
| Brinkerhoff Success Case Method | The why behind extreme outcomes, in participants’ own words | Success cases must be findable first — which requires per-person outcome data |
Model choice matters less than record architecture: every model’s upper levels depend on one person’s data connected across time. Which platforms can actually hold that thread is the job of training evaluation software.
Continuous Kirkpatrick is the Loop run on the level map: collect clean at the source so every wave lands on the same trainee record, analyze on arrival so reactions are coded and gains computed while the cohort still runs, improve in time so the weak module is fixed and the refresher aimed before the moment passes. Levels 3 and 4 stop being aspirations in a slide and become two more reads of a record that never died.
The same thread is what makes the claim defensible: a Level 4 number traces back through 90-day behavior, learning gain, and reaction to a trainee’s own words — the standard described in Loop traceability. When a funder asks “how do you know?”, the answer is a chain, not an assertion.
One method, three moves that never stop
Then the cycle runs again, a little sharper each cohort. Read the method: the Loop methodology →
The fastest test of the model is one level of your real program run through it. Each prompt below pastes into Sopact Sense’s Assistant — one per Kirkpatrick level; the arrow above each links the Academy walkthrough with the expected output and tips.
Academy walkthrough → Apply the Kirkpatrick model to a survey
Map the four Kirkpatrick levels onto [PROGRAM]: for each level - Reaction, Learning, Behavior, Results - specify the instrument, the timing, the question set, and exactly what lands on the persistent participant ID. Then flag which level our current data cannot support, and why.
Academy walkthrough → Analyze pre, mid, and post survey data
Here are the pre and post responses for [COHORT]: [PASTE OR ATTACH]. Score both waves on the same rubric, compute each participant's Level 2 learning gain, list everyone below [THRESHOLD], and quote the baseline and exit answers behind each flagged gain.
Academy walkthrough → Measure behavior change after training
From our 90-day follow-ups for [COHORT]: [ATTACH]. Extract Level 3 evidence per graduate - is the skill applied on the job, how often, and with what result - then rank the transfer barriers by frequency and quote each barrier in the graduate's own words.
Academy walkthrough → Connect training to results
Join [OPERATIONAL METRIC] to the trained cohort's records: compare trained vs not-yet-trained staff against baseline, show whether the metric moved for the people with 90-day behavior evidence behind them, and state the attribution limits a board would probe.
Each walkthrough is short and practical: what to do, the prompt to run, the output to expect, and the tips that keep it reliable.
Level 1 Reaction measures whether participants found the training engaging and relevant. Level 2 Learning measures whether knowledge or skill increased, pre to post. Level 3 Behavior measures whether people apply the skill on the job, typically read 60 to 90 days later. Level 4 Results measures whether an organizational metric moved. Sopact carries all four levels on one persistent participant ID, so each level reads against the last.
From a communication-skills cohort run on Sopact Sense: a participant’s intake baseline read “I freeze in meetings”; his exit assessment showed a 34-point gain on the same rubric (Level 2); mentor and peer evidence confirmed on-the-job change (Level 3), ending in an all-hands presentation; and across the cohort, mentor minutes predicted gains better than LMS completions — a Level 4 finding that redirected budget.
Four recur: the levels imply a causal chain they do not guarantee (good reactions predict neither learning nor transfer); the model under-specifies how to measure Levels 3 and 4; it evaluates after the fact rather than shaping design; and it ignores context and inputs. Sopact’s reading: every criticism sharpens to one operational gap — the upper levels demand longitudinal, person-level records — which is what Continuous Kirkpatrick fixes.
The New World Kirkpatrick Model (James and Wendy Kirkpatrick, 2016) updates the original: plan backwards from Level 4 results, define leading indicators, and sustain Level 3 through Required Drivers — the reinforcement, monitoring, and accountability processes that keep new behavior alive. It is a planning discipline, and it still requires per-person data connected across months to execute; that record layer is the part Sopact Sense supplies.
Phillips keeps Kirkpatrick’s four levels and adds Level 5: net program benefits divided by program cost, expressed as a return percentage. The practical difference is effort — ROI requires isolating training’s effect and converting results to money. Choose Phillips when the budget conversation demands a percentage; either way, the Level 3–4 join problem underneath is identical, and Sopact’s persistent trainee ID is what makes both computable.
Because Levels 1 and 2 can be measured in the room, while Levels 3 and 4 require following the same person for months. Form-centric survey tools treat every send as a fresh anonymous batch, so the follow-up can never be matched back to the cohort. Sopact calls the structural fix Continuous Kirkpatrick: a persistent participant ID assigned at enrollment turns the four levels into four views of one dataset.
Collect behavior evidence at 60 to 90 days on the same participant ID: a short follow-up asking for concrete application examples, a manager or peer observation, and one open-ended barrier question. Sopact Sense reads each follow-up on arrival, extracts the application evidence, and surfaces non-appliers by name with their stated barrier — so Level 3 triggers a refresher instead of a shrug.
Pick one operational metric the organization already records — error rate, retention, sales, time to productivity — and join it to the trained group against baseline, ideally with a not-yet-trained comparison. A credible Level 4 claim keeps the Level 3 behavior evidence behind it and states its attribution limits plainly; on Sopact’s one-record thread, that chain is a query rather than a year-end reconstruction.
Yes — as the level map. Sixty-five years on, it remains training evaluation’s shared vocabulary, and no successor has replaced it. What changed is feasibility: persistent-ID records and AI that reads open-ended evidence on arrival make Levels 3 and 4 measurable by an ordinary program team. The model was never the barrier; the infrastructure was, and Continuous Kirkpatrick is Sopact’s name for closing that gap.
Continuous Kirkpatrick is Sopact’s name for running the four Kirkpatrick levels as one loop on one trainee record: a persistent participant ID carrying the reaction pulse, the learning gain scored on one rubric, the 90-day behavior evidence, and the tied result metric. Kirkpatrick supplies the level map; the Loop supplies the cadence, so evaluation happens while the program runs — not after it ends.
Next: run the model across the whole program on training evaluation, or pick up the per-level wording on training evaluation survey questions.
