Kirkpatrick Model

The four levels explained

Donald Kirkpatrick introduced the four levels in a series of articles in 1959, and they remain the most common way L&D teams talk about training effectiveness.

Level 1: Reaction. How participants felt about the training: relevance, engagement, quality of the trainer and materials. Measured with a short survey at the end of the session. Ask about relevance and intent to apply, not just whether people enjoyed it.

Level 2: Learning. Whether participants gained the intended knowledge, skills, attitudes, confidence and commitment. Measured with pre and post tests, practical assessments, simulations or observed roleplays.

Level 3: Behavior. Whether people apply what they learned once back at work. Measured with observation, QA scores, call monitoring, manager checklists or system data, usually 30 to 90 days after training.

Level 4: Results. Whether the business outcomes the training targeted improved: fewer errors, higher first contact resolution, lower early attrition, fewer safety incidents, more sales.

Worked example: evaluating a de-escalation program

A utility company's customer care center trains 120 agents on de-escalating calls about disconnection notices. Here is what evaluation looks like at each level.

• Level 1: Post-session survey, three questions. 88% rate the content relevant to their calls. Trainers note that agents asked for more practice time.
• Level 2: Agents complete a scored roleplay before and after training. Average score on the de-escalation rubric rises from 2.1 to 3.3 on a four-point scale.
• Level 3: QA analysts score 10 recorded disconnection calls per agent at 30 and 60 days on the same rubric. Average live-call score rises from 2.3 to 2.9. Agents on two teams whose leaders did not reinforce the techniques show almost no change.
• Level 4: Over the next quarter, supervisor escalations on disconnection calls drop 18% compared with the prior quarter, and complaints to the state utility commission about agent conduct fall.

Two lessons come out of this that a smile sheet would never show. The skill transferred less than it was learned, and manager reinforcement made the difference. That is exactly what Level 3 exists to catch.

The New World Kirkpatrick Model

Kirkpatrick Partners later updated the framework as the New World Kirkpatrick Model. The main shifts: start planning at Level 4 by agreeing on the business result first, then work backward to the behaviors and learning needed. Level 3 adds required drivers, meaning the processes and systems that reinforce, monitor, encourage and reward the new behaviors on the job. Level 2 now explicitly includes confidence and commitment alongside knowledge and skill.

The practical message: evaluation is not something you bolt on after delivery. If nobody agreed on the target result and the on-the-job reinforcement up front, Level 3 and 4 data will be weak and hard to explain.

Kirkpatrick vs ROI evaluation

Kirkpatrick stops at business results. The Phillips ROI Methodology adds a fifth level that converts those results into money and compares them with the full cost of the program to get a training ROI percentage. ROI calculations are useful for large, expensive programs where leaders want a financial case, but they need careful isolation of the training effect, for example by comparing trained and untrained groups. For most programs, solid Level 3 and Level 4 evidence is more credible than a precise-looking ROI number.

Common mistakes

• Stopping at Level 1. Most organizations measure reaction for nearly every program and behavior for only a few. High satisfaction tells you little about performance.
• Measuring Level 3 with no baseline. If you do not know how people behaved before, you cannot show change.
• Ignoring the work environment. Poor learning transfer is often a manager, process or incentive problem. Kirkpatrick data should point there, not just at the course.
• Claiming all the credit at Level 4. Results move for many reasons: seasonality, staffing, product changes. Use a comparison group or at least a clear timeline, and be modest.
• Evaluating everything to the same depth. Save Level 3 and 4 work for programs that are expensive, high-risk or strategically important.

How to apply it to high-volume training

In contact centers, retail and logistics, the data for the upper levels already exists. QA scores, handle time, first contact resolution, scan accuracy and early attrition are tracked daily. The work is mostly connecting them to training cohorts. Tag every new hire with their class, trainer and program version in your HRIS or learning and development reporting, then compare cohorts at 30, 60 and 90 days. That turns a hiring wave into a natural experiment: when you change the program for one class, you can see the effect within weeks.