The 1975 Assumption

Florida State built the model for the Army in 1975. The original document is Branson's Interservice Procedures for Instructional Systems Development. Analyze, Design, Develop, Implement, and a fifth phase called Control. Evaluation lived inside Control, split into internal and external blocks, run after instruction was delivered.

Look at what the Army had to work with. Paper tests. Instructor observation. A questionnaire at the end of the course. If you wanted to know whether training worked, you asked afterward.

Evaluation Last

The sequence hardened into doctrine. Kirkpatrick's levels gave Evaluation a vocabulary: reaction, learning, behavior, results. Every one of them measured after the fact.

So the E phase became a survey and a quiz. Did learners like it. Did they pass. No one measured whether they could do the job six weeks later, when their manager has other problems.

Over fifty years the tools changed completely. The sequence hasn't.

Four Phases, No Feedback

When E is a survey, every phase upstream runs blind.

Analysis works from SME interviews and job descriptions, not from the challenges people actually face. Design commits to one sequence for everyone because there is no indication where to branch. Developers build the lesson and ship it.

Then a survey. Then the next project.

The loop never closes. You build something, you deliver it, but you have no idea what changed. Our own completion data from tens of thousands of corporate videos shows how little the post-course survey catches: most people are gone long before the end, and they never get the survey. That is the state of most training in operation right now, and the model is the problem.

Adaptive Learning

Adaptive learning already broke the sequence.

Adaptive systems serve a different next step for every individual based on their activity and responses. Every branch decision is a measurement. The system evaluates continuously.

Watch what that does to the phases. Analysis moves into delivery: the system finds out what each person knows while they're in it. Design stops being a sequence and becomes a decision space, closer to systems work than course work. Development either multiplies to cover every branch or generates content on the fly. Implementation and Evaluation become the same event.

Four phases collapse into two. The capabilities changed and the model is following, quietly, without a name. Researchers have started proposing one: ADGIE swaps Development and Implementation for Generation and Individualisation. Same diagnosis, different label.

There's a cost. An adaptive system produces paths no reviewer signed off. ADDIE's quality gates assumed a finished lesson you can read from end to end. When the system chooses the route, governance has to sit on the rules and the data. Most organizations haven't figured it out yet. Some are running adaptive content with no one accountable for what it is delivered.

Evaluation as a Property

Per-second data during a session tells you where a person hesitated, replayed, guessed, quit. Those are struggle signals, and they're the analysis nobody could run in the seventies, running on every learner, every time. Design gets a feedback loop. Development gets told which branch nobody takes. Implementation stops being a date.

The SCORM measurement gap is the same problem from the other side: the LMS reports 94% completed because completion is the only thing the 26 year old format was designed to count.

The model doesn't need a sixth letter. It needs the E moved inside the I, and the rest reordered around what the data now makes possible.

We built REACHUM for exactly this: role-play practice where the behavioral signal streams while the conversation is happening, not on a form after. And conversational video provides video interactive video without time consuming development time.

Performance is the delivery. When we shift evaluation into the middle, outcomes are optimized.

If your training process still ends with a survey and completion notification, you're running the 1975 model on 2026 instruments.

Shift the E.

FAQ

Is the ADDIE model still relevant?

The objectives are. The sequence isn't. Analysis, design and development still matter. Evaluation as a final phase doesn't fit modern systems or needs.

What are the weaknesses of the ADDIE model?

Evaluation is sequenced last, so every other phase runs without feedback. Design assumes one path for everyone. The model has no mechanism for learning from delivery while delivery is happening.

How do you evaluate in ADDIE?

Traditionally with post-course surveys and tests, mapped to Kirkpatrick's four levels. Modern delivery captures behavioral data during the session, which makes post-course evaluation the weakest available signal.

How does adaptive learning change ADDIE?

Adaptive learning requires continuous evaluation to route each learner. That collapses Implementation and Evaluation into one phase and moves Analysis so that it overlaps delivery.