The 60–90 day on-the-job follow-up that shows whether the training actually transferred — run on the same participant ID that already holds their learning score.
In short: Kirkpatrick Level 3 measures whether people actually use the trained behavior back on the job. You measure it with a 60–90 day follow-up that fires per participant on the same ID they had at enrollment — a self-report behavior scale, a short manager or peer rating, and one open "what got in the way" question. Because the follow-up sits on the same record as the Level 2 learning score, you get an application rate, a ranked list of transfer barriers, and a per-person view of learning that did — or didn't — turn into behavior.
Almost no training program gets to Level 3, for one reason: the data has to follow one person 60–90 days after the room empties. If the follow-up is an anonymous batch survey, you can never join it back to who learned what. Sopact Sense solves that by carrying one persistent participant ID from intake through every wave.
Create a follow-up instrument that fires 60–90 days after training completion, addressed to each participant on their original ID — never a re-typed name or a fresh anonymous form. That single field is what makes Level 3 possible: it lets the follow-up join back to the person's Level 1 reaction and Level 2 learning gain already on the record. Schedule it as a timed wave off the completion date so it goes out automatically per participant rather than as one blast.
Keep the instrument short and mixed-method. Include (a) a self-report behavior scale — how often the participant now uses the trained behavior on the job, on the same scale you will reuse every wave; (b) a short manager or peer rating of the same behavior, so you have a second view that isn't just self-perception; and (c) one open-ended "what got in the way of applying this?" question that captures the transfer barriers a number never will. All three land on the participant's ID, next to their learning score.
The headline Level 3 number is the application rate: the share of participants applying the trained behavior at a meaningful level, ideally confirmed by the manager rating rather than self-report alone. Because every response is on a persistent ID, the rate is a real count of people, not an average of an anonymous batch — and you can segment it by cohort, role, or manager.
Code the open-ended "what got in the way" responses into themes — no time, no manager support, no tools, forgot the method, no opportunity to practice. Rank the barriers by frequency and pull a representative quote for each. These barriers are the most actionable output of the whole level: they tell you why learning didn't transfer and what to fix, and each one can be routed to a named owner.
Because learning and behavior sit on the same record, you can correlate them: who learned the material (Level 2 gain) and then actually changed behavior (Level 3), versus who learned it and didn't. The gap between them — learning that didn't transfer — is exactly what Level 3 exists to expose. Treat this as correlation, not proof of causation; it points you at the transfer problem without claiming the training alone caused the change.
Paste this into the Sopact Assistant, swapping in your cohort:
Analyze Level 3 (Behavior) for [COHORT]: from the 60–90 day follow-up, measure whether each participant applies the trained behavior on the job (self-report plus manager rating on the same ID), report the application rate and the top barriers to transfer from the open-ended responses, and correlate behavior change with the Level 2 learning gain.
Expected output: an application-rate figure for the cohort, a ranked list of transfer barriers with a representative quote for each, and a per-participant behavior-change record correlated to that person's Level 2 learning gain — so you can see, by name, where learning did and didn't turn into behavior.
Input → output: in — the 60–90 day follow-up (self-report behavior scale, manager/peer rating, open barrier question) joined on the participant ID to their Level 2 score; out — application rate, ranked transfer barriers, and a per-person learning-to-behavior record.
GRADE: green | Manager-confirmed application on the same ID | traced to learning gain; amber | Self-report only | no manager view; red | Anonymous batch survey | cannot join to the person
Measure behavior, not satisfaction. Level 3 is not "did they like it" or even "did they learn it" — it is whether they do it on the job. Ask about observable behavior and frequency of use, and lean on the manager rating so the number isn't only self-perception.
Use the same scale every wave. If you change the behavior scale or the wording between cohorts, you can't compare application rates over time. Lock the scale and the codebook once and reuse them each wave so the trend is real, not an artifact of new questions.
Correlate, don't claim causation — and route barriers to a named owner. Correlating behavior change to the Level 2 gain shows where learning failed to transfer, but it isn't proof the training caused the change. Report it as correlation, then take the ranked transfer barriers and assign each unresolved one to a named owner so the follow-up actually fixes something.
Kirkpatrick Level 3 (Behavior) measures whether participants apply what they learned back on the job — the transfer of training into changed behavior — as opposed to Level 1 (their reaction to the training) or Level 2 (what they learned). It is the level most programs skip because it requires following each person 60–90 days after the training on a persistent ID.
Send a 60–90 day follow-up to each participant on their original ID that combines a self-report behavior scale, a short manager or peer rating of the same behavior, and one open "what got in the way" question. Compute the application rate (ideally manager-confirmed), code the barriers to transfer, and correlate the result to each person's Level 2 learning gain — all on the same record in Sopact Sense.
Send it 60–90 days after training completion — long enough that participants have had real chances to apply the behavior on the job, but soon enough that they still remember the training. Fire it as a timed wave per participant off their completion date, then link forward to Level 4 to connect the behavior change to an organizational result.
ChatGPT, Claude, and Gemini are fine for a quick test — but not for an answer you'll put in front of a funder or board. When it has to hold up, run it in Sopact Sense.
Try it in Sopact →
