How to Build a Training Course That Builds Capability

The output of a training course is not the course. It is what people can do when it is over.

That sounds obvious, and almost no course is built that way, because the natural place to start is the material. This guide sets out the sequence I use instead.

Where this comes from.

The design direction is backward design, set out by Grant Wiggins and Jay McTighe in Understanding by Design (ASCD, 1998): establish the desired result, then the evidence that would demonstrate it, then the learning experience. Nothing in section 1 below is original to me. What follows is that framework applied to professional training, with twenty years of my own practice in how it survives contact with real organisations.

You build the course in one direction and the learner meets it in the other. Start where the learner starts and you will design a syllabus.

Capability, then evidence, then experience, then content when you design it. Content, then experience, then evidence, then capability when they take it.

Section 1Decide four things before you write anything

These are not four steps. You settle them together and each one changes the others. Content comes after all four.

Expect the traffic between them. Working out what evidence would convince you will usually send you back to rewrite the capability, because the first version turns out not to be observable. Meeting the learners will change what you decided must not deteriorate.

The capability.

Write what a competent learner will be able to do differently, in observable terms.

Not: understand AI, learn prompt engineering, become familiar with Copilot. Those describe exposure.

Instead: a learner can identify an appropriate task for AI, supply the right context and evidence, assess what comes back, and decide whether to accept, correct, reject or escalate it.

What must not deteriorate.

Ask what gets worse as learners get faster. In professional training this is usually technical judgement, source discipline, accountability, confidentiality, or the ability to do the work unassisted. Write these down. They constrain every decision that follows, and nobody asks the question.

What you believe causes learning.

Be explicit about it. Many models work. The one I use is attention, then active engagement, then feedback, then mastery. It is not a new model. It is a working compression of Gagné's nine events of instruction, first set out in The Conditions of Learning (1965): attention is his first event, active engagement is eliciting performance, feedback is his seventh, and mastery covers assessing performance and enhancing retention and transfer. I use four rather than nine because four can be held in your head while designing.

• Attention. Learners need a reason to care before they need a definition.
• Active engagement. They have to do something with the material, not hear about it.
• Feedback. They need to know where their reasoning was strong and where it has to change.
• Mastery. Capability comes from repeated application in new contexts, not one successful attempt.

What evidence would convince you.

For each capability, ask what you would need to observe to believe somebody can do it. Useful evidence usually involves learners framing a problem, selecting evidence, completing a realistic task, finding a deliberately introduced error, explaining a decision, or defending why they accepted or rejected an answer.

Attendance tells you somebody attended. Completion tells you somebody completed. Neither tells you what they can now do.

Check. Can you say what a competent learner will do differently on Monday, and how you would know?

Section 2Assemble the expertise

You can design the educational architecture yourself. You cannot invent the professional standard.

Do the domain research, then find the people who actually do the work. Not to review a finished course, which is too late to change anything, but to set the difficulty before you design it.

Without them every scenario acquires one artificially clean answer. Experienced learners recognise that immediately, and once they do they stop treating the course as real.

Use them to establish:

• what actually goes wrong in this work, and how often
• which decisions have to stay with a person
• where two competent professionals could reasonably disagree
• whether the evidence in your scenario would genuinely exist in a real organisation
• whether your intended answer is the only defensible one

The last question is the important one. Professional education gets weak very quickly when every case has one clean answer. Where competent people could disagree, the course should teach learners to justify a defensible judgement rather than guess which answer the designer wanted.

Expertise should not decorate a finished course. It should change the decisions the course asks learners to make.

Check. Did anything in the design change after talking to the experts? If not, you were seeking approval rather than expertise.

Section 3Shape the experience

Ask where the learning needs another person.

Some things are learned perfectly well alone. Others only work when a learner hears how a colleague reasoned, or when a trainer can challenge them. That question decides the medium, not the other way round.

Then choose the medium.

A webinar is good for awareness. A workshop is better when people need to practise and compare. Tool skills need the tool. Judgement needs cases, discussion and feedback. A skill that needs repetition is usually better spread over weeks than delivered in one block.

Learning that transfers is practical, relevant and social.

The first two are Malcolm Knowles's ground. His andragogy holds that adults need to know why they are learning something, bring existing experience that has to be engaged rather than bypassed, and are oriented towards problems from their own work.

Social means three relationships, not one.

People learn from people, and a designer who thinks only about the first of these is designing a delivery mechanism rather than a learning experience.

The trainer and the learner. A skilled trainer notices uncertainty, draws in the quiet participant, challenges overconfidence, adapts an explanation and gives feedback that static content cannot. Whether you call that person a teacher, coach, facilitator or guide, the relationship is doing real work. Technology can extend it. It should not make you forget why it matters.

The learners and each other. They compare reasoning, watch how somebody more experienced approaches a problem, explain their own thinking and discover its weaknesses while doing so, and ask each other questions they would never ask a piece of content.

The learner and everybody else. Colleagues, managers, the people they will do the work with afterwards. This is where transfer happens or fails, and it is mostly outside your control, which is not a reason to ignore it. Jean Lave and Etienne Wenger's Situated Learning (Cambridge, 1991) treats learning as participation in a community of practice rather than transfer into an individual head. A learner returning to a team that works the old way will revert, whatever the course achieved.

Design for all three. Most courses design for the first and call it delivery.

Build in variety.

A course that repeats one pattern becomes predictable and learners start working out the format. Change how they meet ideas: short explanation then discussion, demonstration, paired work, small group problem solving, live critique, a deliberately flawed example, teaching it back, an open question with no single right answer.

Vary the order too. Sometimes demonstrate first. Sometimes let them try before showing them, which is Manu Kapur's productive failure: an unsuccessful attempt before instruction prepares the learner to take more from the instruction than they would have taken cold. Sometimes give groups different evidence and make them reconcile.

Difficulty is a design material, not an accident. Robert and Elizabeth Bjork's work on desirable difficulties shows that conditions which slow performance during learning often improve retention and transfer afterwards. The word doing the work is desirable. Difficulty that only confuses is just difficulty.

Check. Could a learner predict the next twenty minutes? If so, they are working out the format rather than the substance.

Section 4Make it, test it, run it

Content comes late.

Only once the architecture is set do you build explanations, cases, exercises, slides, facilitator notes and handouts. Build what the experience needs and no more.

Where AI helps here.

Generating alternative scenarios, varying difficulty, producing deliberately flawed outputs, drafting facilitator notes, simulating the questions learners will ask. A plausible syllabus generated in thirty seconds is still only a plausible syllabus. Cheap generation makes the judgement about which content deserves to exist more valuable, not less. I argue that at length in Cognitive Rusting and The Scaffolding of Intellect.

Rehearse it, in whatever form you can afford.

A full dry run with real people is best and is rarely available. Cheaper forms work and most of the value is in them: read the material as a learner rather than as its author, walk the session through in your head, visualise where a group would stall or go quiet, complete every exercise yourself, ask somebody unfamiliar with the subject to follow your instructions.

Check the timing. Look for ambiguous questions, exercises that give away the answer, and difficult exercises that turn out not to be difficult. Ask whether the activity separates somebody who understands the subject from somebody who can follow instructions.

For AI training, include realistic failure. Omissions, unsupported claims, misleading framing, source problems, plausible wrong answers. If learners only meet well behaved output, you are preparing them for an environment that does not exist.

Deliver the learning, not the plan.

Watch the learners rather than the slides. Notice where they engage and where they get lost. Pay attention to disagreement between them, because it usually exposes something worth opening up. Create space for people to explain their thinking rather than give an answer. Sometimes the most useful thing is to let a discussion run past the next planned activity.

Check. Did the rehearsal change anything? If nothing changed, you rehearsed the delivery rather than the design.

Section 5Find out whether it worked

Go back to the capability statements. What can learners now do? Where are they still struggling? Which mistakes persist? What transferred into actual work? What should change next time?

Satisfaction answers a different question. Donald Kirkpatrick separated four levels of evaluation in the 1950s: reaction, learning, behaviour and results. Most training is still evaluated at level one and reported as though it were level three. People can enjoy a course and learn very little, and find one hard and gain a great deal. Measure what the training was meant to change.

Check. Are you measuring the capability you defined in section 1, or are you measuring completion and calling it evidence?

Section 6Where this is hard

Three conditions make the sequence difficult, and it is better to name them than to pretend the method is robust to everything.

There is no access to real learner work.

Every exercise becomes synthetic, the difficulty is guesswork, and section 2 can only get you part of the way. Get hold of anonymised real artefacts if you possibly can. They are worth more than any amount of invention.

The commercial and the educational pull in different directions.

The sponsor is buying something specific: a headcount trained, a deadline met, a certification, an audit answer, a budget spent before year end. The learning needs something else: time, small groups, rehearsal, real work, people who can be brought together. Those are not always the same purchase. Name the tension early, decide what you are willing to trade, and be clear with yourself about which of the two you are protecting when they conflict.

It assumes a trainer who can read a room.

The sequence produces good scaffolding. What happens inside it depends on somebody noticing confusion, following a productive disagreement and knowing when to stop teaching. Handed to a facilitator reading a deck, this method degrades into content delivery with better documentation.

Section 7Three things worth protecting

Make it practical. Make it relevant. Make it social.

Then build enough structure around those conditions for learning to happen, and no more. The sequence above is a practice, not a process. Understanding the learners will change the capability statement. Talking to an expert will change the exercise. What happens in the first session will make you redesign the second.