New webinar on 3rd March 2026 | 9:00 am PT
In this webinar, discover how Sopact Sense revolutionizes data collection and analysis.
The Kirkpatrick Model explained — all four levels, why most program stop at Level 2, and the data architecture that finally makes Level 3–4
Last updated: April 2026
The funder email arrives on Tuesday. Not about satisfaction scores — those have been answered at 4.3 out of 5. The funder is asking whether participants applied the skills on the job, whether behavior changed at 90 days, whether the cohort produced business outcomes worth renewing. The question is pure Kirkpatrick Level 3 and Level 4, which is exactly where the answer gets stuck. Level 1 data lives in the post-training survey tool. Level 2 lives in the LMS. Level 3 data was never collected because the 90-day follow-up was "planned for next cycle." Level 4 data lives in the HRIS with no connection back to training. This is The Cascade Break — the structural failure where Donald Kirkpatrick's four-level cascade of Reaction → Learning → Behavior → Results is measured as four disconnected events because no persistent learner identity links the levels across the tools that produce them.
The Kirkpatrick Model is not broken. The model has been the global standard for training evaluation since 1959 and remains the framework every serious funder, accreditor, and L&D leader references. What breaks is the data architecture beneath it. The four levels are explicitly designed as a cascade — each level's evidence depends on the previous level's measurement. Break the cascade anywhere, and Levels 3 and 4 become narrative claims rather than measurements. This guide covers all four levels in practical terms, explains why 65% of programs stall at Level 2, and shows what the architecture looks like when the cascade actually holds.
The Kirkpatrick Model is a four-level framework for evaluating training programs that measures participant Reaction (Level 1), Learning (Level 2), Behavior change on the job (Level 3), and organizational Results (Level 4). Developed by Donald Kirkpatrick in 1959 and refined by James and Wendy Kirkpatrick in the New World Kirkpatrick Model, it remains the most widely used training evaluation framework in workforce development, corporate learning, healthcare training, military instruction, and leadership development. The model's power is not its complexity — the four levels are conceptually simple — but its causal logic: each level's outcome depends on the previous level's success.
The Kirkpatrick Model works when the four levels are measured against the same participants over time. It fails when Level 1 satisfaction averages get calculated on Monday, Level 2 pre-post deltas get calculated three weeks later on a different participant list, and Level 3 follow-up surveys go out to whoever opens a bulk email 90 days after program end. Averages across different participant populations are not a cascade — they are four independent snapshots with no causal link between them.
Sopact Sense treats Kirkpatrick implementation as an architecture problem first and an instrument design problem second. Every learner receives a persistent unique ID at enrollment that carries through every Level 1 post-survey, Level 2 rubric, Level 3 follow-up, and Level 4 outcome record — so the four-level cascade is a default output, not a heroic reconciliation effort.
The four levels of the Kirkpatrick Model are Reaction, Learning, Behavior, and Results — measured in that order and answering progressively higher-stakes questions about training effectiveness. Each level requires specific instruments, specific timing, and a persistent participant identifier to link the level back to the learner who generated the previous level's data.
Level 1 — Reaction measures whether participants found the training engaging, relevant, and well-delivered. It is typically captured through a post-training satisfaction survey within 24 hours of program end. Response rates are high (often above 85%) and the data collection is simple. Level 1 tells you almost nothing about whether the program worked, but it tells you whether participants will recommend the program to peers — which affects enrollment for the next cohort.
Level 2 — Learning measures whether participants acquired the intended knowledge, skills, or attitudes. Level 2 requires a paired pre-training and post-training assessment using identical items scored against identical rubrics, with the delta calculated per individual participant — not as an average across two separate groups. Without a persistent learner ID linking pre to post, the Level 2 delta is mathematically impossible to compute for any specific person.
Level 3 — Behavior measures whether participants applied the skills on the job, typically at 30, 60, or 90 days post-program. Level 3 is where 65% of training evaluations stall because the ID required to link the 90-day follow-up to the participant's original pre-training baseline lives in a tool that does not talk to the follow-up delivery tool. Level 3 is also where funders base renewal decisions — so the gap between what the model promises and what most programs actually deliver is expensive.
Level 4 — Results measures whether the training produced targeted organizational outcomes — productivity, retention, safety incidents, revenue, promotion velocity, customer satisfaction. Level 4 requires linking training records to business outcome data over 6 to 12 months post-program. Fewer than 35% of organizations consistently measure Level 4, and most of those measurements are narrative case studies rather than statistical attribution.
Kirkpatrick Level 2 evaluation methods measure knowledge, skill, and attitude change through paired pre-training and post-training assessment against identical rubrics. The most common Level 2 methods are multiple-choice knowledge tests, performance observation scored against a competency rubric, self-reported confidence ratings on named skills, structured case-study analysis, and role-play evaluation against behavioral criteria. The method matters less than the structural requirement: pre and post must share identical items, identical scoring criteria, and — most critically — identical participant identifiers so the delta can be calculated per person.
Level 2 methods fail when pre-training scores live in an LMS export and post-training scores live in a separate survey tool with no shared ID. The typical retrofit is matching responses by email address, which breaks when participants use work email for pre and personal email for post, or when participants change names mid-program. Sopact Sense assigns the persistent ID at enrollment and inherits it into every Level 2 instrument automatically — so every paired pre-post delta is computable without reconciliation.
Kirkpatrick Level 3 evaluation measures whether participants are applying the trained behaviors in their actual work environment, typically through a combination of learner self-report, manager or mentor observation, and direct performance data against named behavioral criteria. Level 3 is the level that matters most to funders and senior sponsors because it is the first level that demonstrates the training changed something beyond the classroom. It is also the level where most programs quietly fail.
The Level 3 measurement sequence has four components. First, define two to four specific observable behaviors the training is intended to produce — written as actions ("conducts structured 1:1s with direct reports weekly") not qualities ("is a better leader"). Second, establish a baseline self-report score against those exact behaviors at enrollment. Third, deliver matched post-program observation at 30, 60, or 90 days — from the learner, and ideally from a manager or mentor against the same behaviors. Fourth, generate the delta disaggregated by cohort, role, and context — paired with open-ended reflection on what enabled or blocked application.
See the workforce cohort example below for what Level 2 and Level 3 look like when pre, post, and mentor observation share the same persistent learner ID. The correlation example shows the deeper pattern: which Level 2 knowledge gains actually predicted Level 3 behavior change, and which did not.
The New World Kirkpatrick Model, developed by James D. Kirkpatrick and Wendy Kayser Kirkpatrick, explicitly reverses the order of program design while keeping the order of measurement. Design starts at Level 4 by naming the organizational result the training must produce, then identifies the Level 3 critical behaviors that would produce that result, then defines the Level 2 knowledge and skills participants need to perform those behaviors, then designs the Level 1 learning experience to deliver that knowledge. Measurement still flows Level 1 → 2 → 3 → 4, but design flows in reverse — which is why the New World model ships Level 3 and 4 evidence instead of treating them as optional.
The reverse-design logic is not controversial; it is structurally correct. Level 4 results exist only if Level 3 behaviors happen. Level 3 behaviors exist only if Level 2 capability exists. Level 2 capability exists only if the Level 1 experience was well enough designed to transfer knowledge. If you design Level 1 first without knowing the Level 4 target, every downstream level is hope rather than plan. See the training evaluation sibling page for how the reverse-design logic integrates with the broader Kirkpatrick Ceiling framing.
Kirkpatrick is the baseline. Phillips ROI extends Kirkpatrick with a fifth level that monetizes Level 4 outcomes for CFO-facing financial justification. CIRO evaluates Context, Input, Reaction, and Outcome — front-loading design quality before measuring outcomes. Brinkerhoff's Success Case Method studies the top and bottom performers through qualitative interviews to isolate what enabled success and what blocked it.
The choice between them is rarely binary. Most mature programs run Kirkpatrick as the default framework with Brinkerhoff layered in for narrative depth, or Phillips ROI for enterprise compliance and leadership development where financial justification is required. CIRO and CIPP are most common in multi-phase public sector and international development programs. See the training evaluation methods page for the full seven-method comparison.
Apply the Kirkpatrick Model by starting with the funder's or sponsor's actual question, naming which level answers that question, then designing backward from there. A funder asking "did the program produce outcomes?" needs Level 4. A funder asking "are participants applying the skills?" needs Level 3. A program director asking "is our instructional design working?" needs Level 2. A conference organizer asking "did attendees find it valuable?" needs Level 1.
The application sequence has five steps. Name the target level. Define the two to four specific observable behaviors that level measures. Design the instruments that will capture those behaviors at the required intervals. Assign a persistent participant ID before the first instrument runs. Build the report template against the ID chain before the cohort ends.
Most Level 3 implementations fail because the LMS is treated as the only measurement surface — and the LMS participant ID doesn't travel to managers, mentors, or 90-day follow-up. This walkthrough shows mentor-based behavior evaluation running on a persistent learner ID spine, no LMS integration required.
Two to four observable behaviors defined before the cohort — scored identically by learner and mentor against the same rubric.
Mentor observations inherit the learner's ID automatically — no CSV matching, no LMS integration.
Self-report, mentor observation, and 90-day follow-up — agreement and dissent surface automatically.
Watch the walkthrough — then see the workforce and correlation examples above for what Level 2 pre-post deltas and Level 3 behavior evidence look like on the same spine.
See Training Intelligence
The single highest-leverage decision in Kirkpatrick implementation is made before the first intake form is built — not after the cohort graduates and the funder question arrives. Programs that design the persistent ID architecture upfront produce Level 3 and Level 4 evidence as default outputs. Programs that retrofit after the fact produce narrative claims dressed as measurements.
The Kirkpatrick Model is a four-level framework for evaluating training programs, measuring Reaction (Level 1), Learning (Level 2), Behavior on the job (Level 3), and organizational Results (Level 4). Developed by Donald Kirkpatrick in 1959 and extended in the New World Kirkpatrick Model, it remains the most widely used training evaluation framework globally. The four levels form a causal cascade — each level's outcome depends on the previous level's success.
The four levels of the Kirkpatrick Model are Reaction, Learning, Behavior, and Results. Level 1 Reaction measures participant satisfaction with the training experience. Level 2 Learning measures knowledge and skill acquisition through pre-post assessment. Level 3 Behavior measures on-the-job application at 30-to-90 days post-program. Level 4 Results measures organizational outcomes — productivity, retention, revenue, safety incidents — over 6 to 12 months.
The Cascade Break is the structural failure where Kirkpatrick's four-level cascade (Reaction → Learning → Behavior → Results) gets measured as four disconnected events because no persistent learner identity links the levels across the tools that produce them. The model is not broken — the data architecture beneath it is. Sopact Sense closes the cascade by assigning persistent learner IDs at enrollment that carry through every subsequent instrument.
Measure Kirkpatrick Level 2 learning through paired pre-training and post-training assessment using identical items and identical scoring rubrics, with the delta calculated per individual participant. Common methods include multiple-choice knowledge tests, performance observation scored against a competency rubric, self-reported confidence ratings on named skills, structured case-study analysis, and role-play evaluation. The pre-post pair must share the same participant identifier or the delta is statistically meaningless.
Measure Kirkpatrick Level 3 behavior change by defining two to four specific observable behaviors at intake, capturing a baseline self-report score at enrollment, collecting matched scores from the learner and ideally a manager or mentor at 30, 60, or 90 days post-program, and pairing the statistical delta with open-ended reflection on what enabled or blocked application. All four measurements must share the same persistent participant ID.
Measure Kirkpatrick Level 4 results by linking training records to business outcome data — productivity, retention, safety incidents, revenue, promotion velocity — over 6 to 12 months post-program through the persistent participant ID assigned at enrollment. Fewer than 35% of organizations consistently measure Level 4 because the participant ID in training records rarely matches the ID in HRIS or operational systems without manual reconciliation.
The New World Kirkpatrick Model, developed by James D. Kirkpatrick and Wendy Kayser Kirkpatrick, reverses the order of program design while keeping the order of measurement. Design starts at Level 4 (target organizational result) and works backward through Level 3 behaviors, Level 2 capabilities, and Level 1 experience. Measurement still flows Level 1 through 4. Reverse-design ensures Level 3 and 4 evidence is built in from the start rather than retrofitted.
The Phillips ROI Model extends the Kirkpatrick Model with a fifth level that monetizes Level 4 organizational outcomes using the formula ROI% equals Net Program Benefits divided by Program Costs, multiplied by 100. Phillips is required when CFO-facing financial justification is needed for enterprise leadership development or compliance training investments. Kirkpatrick's four levels remain the underlying structure — Phillips adds the financial translation layer on top.
The Kirkpatrick Model is used for leadership development evaluation by naming Level 4 outcomes (promotion velocity, team retention under the leader, 360-degree feedback scores, business unit performance) and working backward to Level 3 critical leadership behaviors and Level 2 capability development. Level 3 for leadership programs typically includes structured manager observation, direct-report 360 feedback, and self-report against named leadership behaviors — all tied to the same participant ID.
The Kirkpatrick Model is used for sales training evaluation by defining Level 4 outcomes (quota attainment, pipeline velocity, deal size, win rate) and working backward to Level 3 sales behaviors (discovery call structure, objection handling, closing technique), Level 2 product and methodology knowledge, and Level 1 training experience quality. Sales evaluation benefits from Level 4 availability — CRM systems already track outcomes — but requires persistent participant ID to link training to CRM records.
Yes — the Kirkpatrick Model applies to coaching evaluation with Level 1 measuring coach-coachee fit and session quality, Level 2 measuring insight or skill acquisition through structured self-report, Level 3 measuring behavior change in the workplace at 60 to 90 days post-engagement, and Level 4 measuring outcomes aligned with the coaching goal. Coaching evaluation at Level 3 depends heavily on manager observation and 360-degree feedback tied to the coachee's persistent ID.
Most programs stop at Kirkpatrick Level 2 because the data architecture required to link Level 3 and Level 4 evidence back to individual participants does not exist in standard training stacks. LMS platforms issue their own participant IDs that do not follow learners outside the LMS. Survey tools issue separate response IDs. HRIS systems use yet another identifier. Without a persistent learner ID chain, Level 3 and Level 4 become manual reconciliation projects that rarely finish before the reporting deadline.
Implementing the Kirkpatrick Model takes between 2 weeks and 3 months depending on scope. A single-cohort Level 1-2 implementation can be stood up in 2 weeks with existing instruments. A full Level 1-4 implementation with persistent ID architecture, Level 3 follow-up cadence, and Level 4 outcome linkage typically takes 4 to 8 weeks of design work before the first cohort enrolls. The architecture is built once and reused across every subsequent cohort.
