Scenario-Based Learning: Examples and How to Design It

Go from storyboard to an interactive course in minutes

Start Free

Picture a property manager in their first month in the role. At 8:10 on a Monday morning a tenant emails two photos: the hot water system is leaking onto the laundry floor, there has been no hot water since Saturday, a drip was reported three weeks ago and nothing happened, and the tenant will stop paying rent if it is not fixed today. The property manager passed the repairs module last month and can tell you that a hot water failure is an urgent repair. The module never asked them to decide what to write back in the next hour, with the owner overseas and a $500 approval limit.

I'm an instructional designer at Coursebox, where we make branching scenario software that turns a described situation into a scenario with decisions, so I'll mention our product where I tested it. Scenario-based learning is the design approach built for that gap between knowing a rule and using it. In this guide I set out Ruth Colvin Clark's definition and her six design components, what counts as a scenario and what does not, when to use one, how linear, branching and role-play formats differ, and what the evidence shows. Then I design one course around that leaking hot water system. Along the way there are three real cases, a product test and two free templates.

What is scenario-based learning?

Scenario-based learning is an instructional design approach in which learners take on a work role and respond to a realistic situation, such as a complaint or a fault, making decisions whose consequences play out, with guidance and feedback built in.

View the full-size diagram

The most careful definition comes from Ruth Colvin Clark, president of Clark Training and Consulting in Colorado. In the first chapter of her 2013 book, Scenario-Based e-Learning, she defines it as "a preplanned guided inductive learning environment designed to accelerate expertise in which the learner assumes the role of an actor responding to a work-realistic assignment or challenge, which in turn responds to reflect the learner's choices" (p. 5).

Each word in that definition does a job:

• Preplanned: the designer builds the situation, the choices and the responses in advance, so it is a designed course, not learning on the job.
• Guided: learners are not left to work it out alone. "One of the most important success factors in scenario-based e-learning is sufficient guidance to minimize the flounder factor" (p. 6).
• Inductive: learners meet the case first and work out the principles from it, rather than being told the rules and then applying them.
• The learner is an actor: they take the role and do the work, and the scenario responds to what they choose.
• Accelerate expertise: a run of job situations that might take months or years to meet at work is compressed into a course. Clark's phrase for it is "job experience in a box" (p. 6).

Three points set the scope:

• It is a design approach, not a technology. Clark writes that it "can be realized with a range of technologies as simple as a branching PowerPoint" (p. 12). Her book is about e-learning, but the same components describe live role-plays and drills, such as the police training in Finland and Louisville.
• It goes by other names. In Clark's words, it "has also been labeled problem-based learning, whole-task learning, guided discovery, immersive and goal-based learning" (p. 7). She did not coin the idea, and she says so.
• It rests on the idea that knowledge is tied to where it is used. John Seely Brown, Allan Collins and Paul Duguid argued in a 1989 paper in Educational Researcher that knowledge "is situated, being in part a product of the activity, context, and culture in which it is developed and used." Clark draws the training consequence: knowledge learned in a job context is "more likely to be retrieved when needed on the job" (p. 13). Our guides to situated learning theory and the cognitive apprenticeship model cover that background.

Clark's design model has six components, set out in chapters 3 to 8 of the book. Three are required. In her words, "some components, such as the task deliverable, trigger event, and feedback, are required elements, others may vary according to your learning domain" (p. 35). Here they are at a glance.

A small trial with Finnish police shows several of these components in a real programme: test scenarios designed by senior trainers, difficulty that rises in steps, feedback from instructors and a review between rounds.

Case study: Special response police officers in Finland

This was a small randomised trial with Finland's special response police teams, run with the Police University College of Finland. The researchers asked why "questionable use of force decisions continue to be made despite significant resources directed toward police training", when stress narrows attention and degrades decisions in a critical incident.

• What they did: A three-day programme combined classroom teaching on stress, visualisation and controlled breathing with practice in realistic critical incident scenarios. Scenarios started in slow motion and moved up to full speed, instructors gave feedback, and between rounds officers reviewed what had happened and rehearsed the best response. Senior Finnish National Police trainers, independent of the researchers, designed the test scenarios, which ran in real buildings with police actors, marking rounds, low light and noise. Twelve of the 16 officers invited took part, and a coin flip put six in the programme and six in usual training.
• Evidence and result: In Andersen and Gustafsberg's 2016 paper in SAGE Open, senior trainers who did not know which officers had been trained rated them in the test scenarios. The trained officers made an average of 3.83 correct shoot or no-shoot decisions out of 4, against 2.33, and noticed 5.67 of 6 threats, against 2.5. The control group reported more confidence before the scenarios.
• Limitation: Six officers per group, all men, in one elite unit; the authors call it "preliminary evidence". The programme bundled stress-control teaching with scenario practice, and the control group did no scenario practice at all, so the gain cannot be credited to scenarios alone. The outcomes were test scenarios, not real incidents.
• Lesson: Build scenarios with the people who do the job, raise the difficulty in steps and review between rounds. Then judge competence in a fresh scenario rated by experts, because the untrained group reported more confidence and did worse.

What counts as a scenario, and what does not

Clark is precise about what scenario-based learning is not (pp. 10 to 12).

• Competition does not define a scenario. A scenario can form part of a game, but adding competition is not evidence of better learning. Judge the decisions, guidance and feedback rather than the game label.
• It is not a story followed by questions. In that design "the learner does not become an actor in the scenario. Rather, he or she is more of an observer reflecting on and analyzing it", and such lessons "approximate a directive design more than scenario-based e-learning" (p. 11).
• It is not the same as a simulation. "Although scenario-based e-learning can be based on a simulation, simulation is not an essential element." She defines a simulation as "a dynamic model of a real-world phenomena or process often simplified to reflect core principles" (p. 11).
• It does not need expensive media. One of her figures shows a branched scenario based on links in presentation software (Figure 1.5, p. 14).

Where scenario-based learning comes from

The idea of learning inside a realistic mission is older than Clark's book. In the early 1990s, Roger Schank and colleagues at Northwestern University's Institute for the Learning Sciences designed "goal-based scenarios", in which learners pursue a mission, in a role, within a cover story. Their article, The design of goal-based scenarios, appeared in the Journal of the Learning Sciences volume dated 1993/1994. In a 1999 chapter, Schank, Berman and Macpherson wrote that the activities in a scenario "must have decision points with consequences that become evident" (p. 163), which is still the core of the approach.

Clark's 2013 book turned this family of ideas, which she notes has gone by names including goal-based learning, into a design model for workforce e-learning, grounded in research on guided discovery and expertise. Three facts about the book:

• There is one edition. Scenario-Based e-Learning: Evidence-Based Guidelines for Online Workforce Learning was published by Pfeiffer, an imprint of Wiley, released in December 2012 with a 2013 copyright. I found no second edition in any listing.
• Clark wrote it alone. Frank Nguyen wrote the foreword. Richard E. Mayer, often listed beside her, co-wrote her other book, e-Learning and the Science of Instruction, which our guide to Mayer's cognitive theory of multimedia learning draws on.
• Ruth Colvin Clark is not Richard E. Clark. Richard E. Clark of the University of Southern California co-wrote a 2006 critique of minimal guidance with Paul Kirschner and John Sweller. Both Clarks agree that learners need guidance.

Which source to read

• Read this one first: Clark's Scenario-Based e-Learning (Pfeiffer, 2013). Chapter 1 gives the definition and when to use it, chapters 3 to 8 cover the components one by one, and chapter 10 reviews the evidence honestly.
• If you have ten minutes: Clark's article "Accelerating expertise with scenario-based learning" (T+D, January 2009). Two pages, free on her site, with a short "when to consider" checklist.
• For the history: Schank, Fano, Bell and Jona, The design of goal-based scenarios (1993/1994), on mission-based learning by doing.

When to use scenario-based learning, and when not to

Use a scenario when the job calls for judgement that changes with each situation, when learning it on the job is unsafe, impractical or slow, and when learners already have some related experience. Use a directive lesson, which explains and then practises, for routine procedures and for people new to the knowledge.

That is Clark's own guidance. Her Table 1.1 (pp. 8 to 9) recommends scenarios for "Building critical thinking skills and experience in strategic tasks" and for "Accelerating expertise when on-the-job training is unsafe, impractical, or too slow", and directive designs for "Building procedural skills to learn relatively routine tasks". Her 2009 article puts it plainly: "routine tasks that involve straightforward procedures are best left to a more traditional training approach."

Clark treats experience as a condition, not a detail. She writes that learning while solving a job-realistic problem "can impose quite a bit of mental load", so "in general scenario-based e-learning is best suited to learners who already have some job experience" (p. 10). That does not rule scenarios out for new starters. It changes where they go: some designers "have found it more effective to use the scenario as a culminating learning experience. The best solution no doubt depends on the prior knowledge of your learners and the complexity of your instructional goals" (pp. 9 to 10).

The line between procedure and judgement also depends on the organisation. A refund can be a fixed procedure in one business and a judgement call in another, and Clark notes that the choice "may depend on the business model and constraints of the organization" (p. 19). Most real courses mix both, and she says so: "learning can be blends of different design approaches" (p. 15).

Scenario-based learning examples across Clark's eight domains

Clark groups scenarios into eight learning domains (chapter 2). The examples in this table are mine, one for each domain.

In a tradeoff scenario, Clark notes, "the goal is not so much the end point as the knowledge acquired on the journey" (p. 38), so the feedback has to explain what each path cost, not only where it ended. Many workplace scenarios mix domains. The hot water dispute from the opening is interpersonal, but it also turns on following a repairs procedure and an operational decision about what to authorise.

Linear, branching, simulation or role-play: which format?

Choose the format by the skill and the learners' experience, not by what looks most impressive. "Linear" and "branching" are practitioners' words rather than Clark's. She describes branched scenarios, menu-driven designs, whole-screen designs and virtual worlds. The working definitions in this table are mine, anchored on her points.

More branches do not mean more guidance. Clark compares a branched customer service lesson, which "offers more guidance because response options are limited to a single choice among three options and the on-screen coach provides immediate feedback", with a whole-screen virtual shop that "offers more choices and feedback is not available until after a number of decisions are made" (p. 86). The tighter design suits less experienced staff. Of virtual worlds, she warns about "the high cost of development due to expensive technology" and the extra mental load of navigating for learners who are new to 3D worlds (p. 31).

I found no study comparing linear and branching e-learning scenarios at work, so treat any claim that one format always teaches better with suspicion. The nearest evidence compares types of simulation in higher education: in Chernikova and colleagues' 2020 meta-analysis, role-plays had a moderate effect on learning (g = 0.63, 26 studies), while simulations based on documents, the closest kind to a text-based e-learning scenario, were "the least effective" (g = 0.31, 15 studies). These are comparisons across different studies and settings, not a randomised head-to-head test of formats. They do not establish that a branching course outperforms a linear one.

For building one, our guide to branching scenario examples in e-learning covers how to map and write the branches. For conversations, see the role-play simulation examples for learning and the customer service role-play scenarios, which come with scripts to adapt.

Branching does not need software at all. One police department practised in live scenarios that changed with each officer's decisions, and measured the result in its own records.

Case study: De-escalation training for Louisville police

This was a stepped-wedge trial in Louisville, Kentucky, where the police department's patrol divisions started de-escalation training in an order set at random by the researchers. The department wanted to know whether the training changed what officers did in encounters with people in crisis.

• What they did: In 2019 the department trained 1,049 officers in a two-day de-escalation programme with six modules. On the second day officers practised in two live scenarios with trainers playing people in crisis. "While scenarios began the same, the training staff took different actions based on the officers' decision-making", so each scenario "could play out in many ways, ending successfully or unsuccessfully."
• Evidence and result: The 2020 evaluation report by Engel and colleagues compared eight patrol divisions month by month, using official records. After training, "overall uses of force by the LMPD declined by 28.1%, citizen injuries during encounters with LMPD officers declined by 26.3%, and LMPD officer injuries during encounters with citizens declined by 36.0%."
• Limitation: The live scenarios were one part of a full programme, so the drop cannot be credited to them alone. The analysis rests on eight divisions in one city, with 8 to 12 months of records after training, and arrests also fell over the period.
• Lesson: A branching role-play can sit inside a programme whose effect on the job is measured. Stagger the rollout so that trained and untrained groups can be compared on the organisation's own records.

Scenario-based e-learning: delivering scenarios online

Clark's book focuses on online workforce learning. Online, a scenario can be as simple as linked slides in presentation software, which one of her figures shows (Figure 1.5, p. 14), or as complex as a whole-screen virtual workplace.

For a self-paced scenario, plan guidance that does not depend on a facilitator being present. It can appear as an on-screen coach, an explanation after a choice or support at selected points. That is why her branched customer service lesson limits each decision to three options with immediate feedback.

A worked example: a disputed hot water repair

Here is the hypothetical example I'll design with from here on; the dollar figures are illustrative. A mid-size real estate agency in an Australian city has a property management team that looks after about 600 rental homes for investor owners. The learners are newly promoted property managers in their first month in the role. They come from the agency's leasing or property management assistant jobs, so they know the software, the forms and the owners' names, but none has handled a disputed repair on their own. A senior property manager coaches them and signs off their work.

The problem is the old approach. New property managers read the repairs procedure and pass a quiz on it. Then a dispute arrives, and the quiz turns out to have tested whether they could recognise an urgent repair, not whether they could handle one. Disputes are rare for any one manager, and mistakes cost the agency complaints and owner relationships.

The scenario starts with a trigger event. At 8:10 am on a Monday, the tenant of a two-bedroom unit emails two photos. The hot water system is leaking onto the laundry floor and there has been no hot water since Saturday. The tenant reported a drip three weeks ago and nothing happened, and says they will stop paying rent and take the dispute to the tribunal if it is not fixed today.

So that decisions can be judged, the course supplies the agency's own documents as case data:

• The tenant's email and two photos: water on the floor, and the unit's label.
• The maintenance log: the drip was logged three weeks ago, and a quote request to the owner went unanswered.
• The agency's repairs procedure (extract): failure of the hot water service is an urgent repair. Contact the owner the same day. If the owner cannot be reached, the property manager may engage the agency's approved tradesperson for urgent repairs within the limit in the management agreement.
• The management agreement: the agency may approve repairs up to $500 without the owner. The owner, who lives overseas and replies slowly, has also written "get three quotes for anything over $800".
• The plumber's message: making it safe and replacing a failed valve costs $380. If the tank has failed, replacement is about $1,650.
• The rent ledger: the tenant is paid up to date.

The procedure is the agency's own. The course does not teach tenancy law, and anyone adapting it should check their own state's rules on urgent repairs and rent.

The intended outcome is that a new property manager can classify a disputed repair, decide what to authorise within the agency's limits, write the tenant reply, the owner message and a defensible file note, and know when to bring in the senior property manager. It does not certify competence or give anyone authority beyond the management agreement.

The example passes all four questions in the when-to-use table. The task is a judgement call: the right reply depends on the tenant, the owner, the limit and the history. Learning it on the job is slow and risky. The learners have related experience, so they can carry the mental load of the scenario. And the aim is to apply a procedure they already know to a messy case.

Clark's six components, worked through one scenario

Each component gets the same three parts: what goes in the hot water course, how to check it, and a common slip when the component is written.

1. The task deliverable

The task deliverable is what the learner must produce or decide by the end of the scenario. Clark's 2009 article starts the design here, before anything else: decide what learners will hand in, and the rest of the scenario exists to get them there.

• In the course: three pieces of writing: a reply to the tenant, a message to the owner and a file note recording each decision and why.
• How to check it: the senior property manager reads all three against the procedure, and the learner compares their file note with a model one.
• Common slip: turning the deliverable into a multiple-choice pick ("Which reply is best?"), so nobody sees what the learner would actually write.

2. The trigger event

The trigger event is "how the scenario kicks off. In many lessons it takes the form of an assignment from the manager" (p. 39). Clark also describes a "Murphy's Law trigger" that "starts with a situation in which incorrect decisions and actions were made" (p. 64), which is what an ignored drip three weeks earlier sets up here.

• In the course: the tenant's Monday 8:10 am email, with the two photos, as the first screen.
• How to check it: a learner who reads only the trigger should know what is at stake and that they are the one who has to act.
• Common slip: opening with a page about repairs policy, so the trigger arrives after the learner has been told what to think.

3. Case data

Case data "refers to specific information about the scenario needed to complete the assignment" (p. 66). In a real job, people decide from documents, histories and numbers, and the scenario has to supply them.

• In the course: the six documents: the email, the maintenance log, the procedure extract, the management agreement, the plumber's message and the rent ledger.
• How to check it: for each decision, could a learner reach a defensible answer from the documents alone? If a decision needs a fact the course never gives, either add the fact or cut the decision.
• Common slip: naming the documents without putting the facts in them, so learners are asked to decide blind.

4. Guidance and instruction

Guidance is the support available while learners work through the scenario. Clark lists nine ways to build it in (pp. 77 to 85), including fading support from high to low, moving from simple to complex, choosing open or closed response options, adding coaching and advisors, and embedding worksheets.

• In the course: the procedure extract and a one-page urgent repair checklist stay open throughout, and a coaching note from the senior property manager is available before each decision. For brand-new staff with no agency background, a worked scenario comes first: the senior property manager talks through a different repair, a broken oven, from start to finish.
• How to check it: ask whether someone who has never handled a dispute could make the second decision with what is on screen.
• Common slip: hiding the procedure to make the scenario feel like a test. Property managers have the procedure open at work, so the scenario should too.

5. Feedback

Feedback is what learners see after each choice. In a scenario it has two layers: what happens next in the story, and an explanation of why. Clark counts feedback as one of the three required components. Without it, the learner has made a choice and learned nothing from it.

• In the course: each option shows its consequence first, then a short note from the senior property manager naming the part of the procedure that applied. If the learner promises the tenant they will "chase the owner" with no time frame, the tenant lodges a complaint that afternoon, and the earlier unanswered report becomes the agency's problem.
• How to check it: every option needs useful feedback about the decision and its likely consequences; consequence-first is this example’s design.
• Common slip: marking choices "Correct" or "Incorrect". A label tells learners they were wrong, but not what their choice would have cost.

6. Reflection

Reflection reviews the learner’s decisions against the task and its standards. Clark allows its form to vary by domain. Chernikova and colleagues found differing associations across learner groups, but those subgroup comparisons do not establish that reflection causes worse learning for one group. Keep the debrief focused, give it a clear purpose and test whether it helps learners revise their decisions.

• In the course: three open-text questions: what you decided at each point and why, what someone reading the file note later would conclude, and one thing you would do differently. Then the learner compares their file note with the senior property manager's model.
• How to check it: each question should point back to a decision the learner actually made in the scenario.
• Common slip: a generic "What did you learn?" or a long essay that nobody reads.

How much guidance does a scenario need?

Enough that learners can make each decision with what is on screen, and less as they gain experience. That answer is common ground on both sides of a long argument about guidance.

In a 2006 paper in Educational Psychologist, Paul Kirschner, John Sweller and Richard E. Clark argued that minimally guided instruction "is less effective and less efficient than instructional approaches that place a strong emphasis on guidance of the student learning process", and that the advantage of guidance recedes "only when learners have sufficiently high prior knowledge to provide 'internal' guidance" (p. 75). They added that "For novices, studying worked examples seems invariably superior to discovering or constructing a solution to a problem" (p. 80). Cindy Hmelo-Silver and colleagues replied in 2007 that problem-based and inquiry learning "are highly scaffolded" (p. 99), so they are not minimal guidance at all.

None of that argues against a guided scenario. Ruth Colvin Clark builds guidance into her definition, and her evidence chapter names "guidance and structure" as one of the critical success factors for guided discovery, scenario-based e-learning included (p. 143). Alfieri and colleagues' 2011 meta-analyses, covering 164 studies, found the same pattern: "unassisted discovery does not benefit learners, whereas feedback, worked examples, scaffolding, and elicited explanations do" (p. 1). Our guide to discovery learning sets out those results in full.

What changes is how much guidance different learners need. Slava Kalyuga and colleagues named the expertise reversal effect: "Instructional techniques that are highly effective with inexperienced learners can lose their effectiveness and even have negative consequences when used with more experienced learners" (2003, p. 23). Chernikova and colleagues found the same split inside simulations: "learners with low prior knowledge learned better when supported by examples", while those with high prior knowledge gained more from reflection. Our guide to cognitive load theory shows how to fade support from a worked example to solo practice.

For the hot water course, that means two versions of the same scenario. Newly promoted property managers go straight into it, with the checklist and coaching notes available on request. Brand-new staff watch the senior property manager work through the broken oven first, and meet the hot water dispute at the end of the module, as Clark's culminating experience.

How to design a scenario-based course

Clark's components tell you what a scenario contains. This workflow is my own; it puts them in her design order, adds her criteria for when to use a scenario and builds in the lessons from the cases. It follows the hot water course.

1. Check the task should be a scenario. Run the four questions from Clark's criteria. The repair dispute passes on all of them. Logging a job in the property software does not, so it stays a short directive demonstration.
2. Collect real cases from people who do the job. Ask the senior property manager for three or four repair files that went well or badly, and what a good property manager noticed that a new one would miss. In the Finnish trial, senior police trainers designed the test scenarios.
3. Write the task deliverable and how it will be judged. The three pieces of writing, and the criteria the senior property manager will judge the file note against.
4. Write the trigger and gather the case data. The Monday email, then every document a learner needs, with the real figures in it.
5. Map the decisions and their consequences. Three or four options per decision is enough; Clark's branched designs offer "three or four alternative responses" (p. 27). Write what happens after every option, not only the right one.
6. Write the feedback and set the guidance. Consequence first, then the coaching note. Decide which learners get a worked scenario first.
7. Plan a short debrief. Questions that point back to the learner's own decisions, and a model answer to compare with.
8. Test on a fresh case, and plan the workplace check. Try the scenario with two or three people from the target group. At the end of the course, give learners a new repair dispute and have the senior property manager rate their file note against the criteria, rather than asking how confident they feel. If you can, stagger the rollout so that early and late groups can be compared on complaint records, as Louisville did with use of force.

The decision map for the hot water course

Here are the three decisions, with what happens after each option. After every consequence, the senior property manager's note names the part of the procedure that applied.

The second decision shows why the case data has to be complete. The $500 limit, the $800 quote rule and the $380 valve are what make authorising the repair defensible, and a learner without them can only guess.

What Coursebox course requests leave out

To see how people brief scenario courses, I looked at course requests typed into Coursebox in the 90 days to 7 October 2026: one per person, from English-language browsers, with no uploaded file. Of 266 prompts, 104 were course requests. Two independent AI readers each classified all 104 against written definitions built from Clark's components, and I re-read every positive call in full. Counts use the union of the two readers’ classifications after my review. They are classification estimates, not statistical upper bounds; borderline cases and missed requests remain possible.

Fifteen requests mentioned scenarios, case studies, role-play, simulations or real-world situations in some form. None of them named what starts the scenario, the facts learners need, what follows a choice or a debrief.

The combined classifications identified 4 of the 15 had learners act in the scenario at all. Several asked for "real-world examples" as illustrations, which is not scenario-based learning. Of the requests for workplace training, only two had learners act in a scenario, and one of those used scenarios only as the format of quiz questions. One asked for "scenario(s)" in every module, placed between the explanation and a short knowledge check, which puts the scenario inside a directive lesson.

The sample is small, it covers what people typed rather than the courses they finished, and some may have added scenarios later by editing. Still, a builder given "add scenarios", whether a person or an AI, has to invent the trigger, the case data and the consequences, and the easy default is a story followed by questions. If you want a scenario in Clark's sense, write those components into the brief yourself.

Does scenario-based learning work?

Scenario and simulation training shows large gains in the studies that exist, but most of those studies come from medicine, policing and aviation, and I found no controlled study of a company's office-based scenario e-learning. Clark's own verdict in her evidence chapter is measured: "Practitioners need much more research evidence than is often available. Hence, we need realistic expectations" (p. 139).

The largest review is Chernikova and colleagues' meta-analysis of simulation-based learning in higher education.

An effect size (g) of 0.85 means the average learner in a simulation scored most of a standard deviation higher than the comparison. Against a group taught another way, the figure was similar (g = 0.82, 53 studies), so this is not only simulation against nothing. The limits are as important as the number. The authors warn that the findings "do not straightforwardly generalize to other domains", and the closest kind to a text-based e-learning scenario, a simulation built from documents, had a smaller pooled effect.

Other reviews are more cautious:

• Virtual patients. Cook, Erwin and Triola's 2010 review found that computer patient cases "are associated with large positive effects compared with no intervention", but that "Effects in comparison with noncomputer instruction are on average small." The features that helped included "repetition until demonstration of mastery" and "enhanced feedback".
• Case-based learning. A 2012 review of 104 papers by Thistlethwaite and colleagues found that "students enjoy CBL and think that it enhances their learning", but that the data "are inconclusive as to the effects on learning compared with other types of activity."
• Simulation games. Sitzmann's 2011 meta-analysis of 65 studies with 6,476 trainees found gains over comparison groups, with "strong evidence of publication bias", and concluded that games work best as a supplement to other instruction. Games did worse than comparison methods that actively engaged learners. Games and scenarios can overlap, but the claim that scenarios are more motivating gets no support here either: Wouters and colleagues found serious games were not significantly more motivating than conventional instruction.
• Branching video with nurses. In a 2025 randomised trial with 95 practising nurses in Taiwan, nurses who learned from branching decision videos rather than a lecture improved their trauma knowledge, and the gain held for 12 weeks. Their clinical reasoning scores were higher than the lecture group's, "but not statistically significant". It is the closest controlled study I found of branching e-learning with working professionals, and I read only the abstract.

Clark reports two studies that show what scenarios can do at their best. In one, by Lim, Reiser and Olina (2009), learners taught Excel through a scenario did much better than learners taught step by step on a budget task neither group had practised. In the other, she quotes Gott and Lesgold's account of the Sherlock troubleshooting tutor for US Air Force technicians: "twenty to twenty-five hours of Sherlock practice time produced average improvements that were, in many respects, equivalent to the effects of ten years on the job" (p. 152). Her next sentence is the one to remember: "Note, however, that the learners already had some background experience."

The most useful warning comes from a programme that learners loved.

Case study: New child protection investigators in Illinois

From February 2016, every new child protection investigator for the Illinois Department of Children and Family Services trained at a simulation academy run by the University of Illinois Springfield. New investigators make child safety decisions in real homes and testify in court, and turnover in the job is high.

• What they did: Trainees practise "from the first knock on a family's door to testifying in family court, guided by expert trainers and working with actors playing the family in a mock house and mock courtroom." Actors stay in character and give feedback, and trainees get individual and group debriefs. In 2020 to 2021 the week ran online, with a trainer's head camera as the trainee's eyes in the mock home.
• Evidence and result: The 2021 evaluation by Cross and colleagues at the University of Illinois found that self-rated confidence rose across the week, with medium to large effects (d = .64 to 1.10) for the 84 trainees who completed all six surveys. Field results did not follow. Comparing 270 simulation-trained investigators with 134 earlier ones, child maltreatment re-reports were slightly higher (a hazard ratio of 1.32, below the threshold for clinical relevance the evaluators used), and in a same-period comparison with 88 investigators there was no difference. The odds of leaving the job within 12 months were 1.50 times higher for the simulation group, a reversal of an earlier year's result that the authors thought the pandemic most likely explained.
• Limitation: The comparisons are historical, and the authors write that the comparison "is confounded with era and we cannot fully resolve the ambiguity." Confidence was self-rated, with no observed skill rating, and the online year may not match the in-person programme.
• Lesson: A realistic, well-liked scenario programme can raise confidence sharply with no measurable change on the job. Rate what learners actually do in the scenario before claiming any effect at work, and treat an online version as a new programme to be evaluated on its own.

In a 2007 study of 46 recently instrument-rated pilots, all of them college students, 93% said before training that they were comfortable with GPS, despite poor baseline performance. Of three groups, the scenario-trained one then made the fewest errors on air traffic control clearances in a simulator flight.

Scenario-based learning vs problem-based learning, case-based learning and simulation

These approaches overlap, and Clark herself lists problem-based learning as another name for what she calls scenario-based e-learning. The table is my synthesis, not a ranking.

The practical difference is who acts. In a case study, learners analyse a decision that has already been made. In a scenario, they make it, and the case moves on with their choice. Problem-based learning shares the problem-first order but usually leaves learners to research and argue in a group, while Clark's scenarios supply the case data and coaching on screen.

Scenarios also sit close to two whole-course models. Merrill's principles of instruction put a real-world task at the centre of every lesson, and the 4C/ID model builds whole programmes from sequences of real tasks with fading support. For the wider theory behind learning from problems, see our guide to constructivism. In the hot water course, I would use a scenario for the dispute and a short directive demonstration for logging the job in the software.

Benefits of scenario-based learning

Here is what scenario-based learning offers, with the strength of the evidence for each.

• Job experience in a box. A run of situations that might take months to meet at work is compressed into a course, which is Clark's case for the approach (p. 6).
• Practice at deciding, not only knowing. The learner acts in the situation and sees what each choice leads to, which a story followed by questions does not give.
• Positive results in specific settings. Chernikova and colleagues found an overall effect of g = 0.85 across simulations in higher education, including medical, teacher and management education. This is not an effect estimate for workplace branching scenarios. The Taiwan nursing trial described earlier reported a gain lasting 12 weeks.
• Practice without exposing a real customer, patient or budget to the decision. A wrong choice in a scenario has no customer, patient or budget attached, so learners can try the option they would be tempted to take at work.

Motivation is not guaranteed. Wouters and colleagues did not find a statistically significant motivational advantage for serious games over conventional instruction. That is not proof of equal motivation, and serious games are not interchangeable with every scenario format.

Limitations of scenario-based learning

Scenarios are not the right answer for every course, and their own advocates say so.

• They cost more to build. Clark's 2009 article warns that "the design and development effort is likely to be a bigger investment of time and resources than more traditional forms of training." In her book she reports that one whole-screen design "required four to five hundred hours for the first simulation" (p. 184). A short text-based branched scenario is a smaller job, but every branch still needs writing and review.
• They can overload new starters. "For learners who are brand new to a work environment, immersion in a problem may impose too much mental load and bog them down" (Clark 2009). A worked scenario first, or the scenario at the end, is the fix.
• Some learners prefer to be told. Clark notes that "your learners may prefer a more traditional directive training environment" (2009).
• The workplace evidence is thin. Most controlled studies come from medicine, policing and aviation, and I found none comparing linear and branching scenarios, or scenarios and directive e-learning, in office training.
• Confidence and skill can point in opposite directions. In Finland the untrained group felt surer and did worse. In Illinois confidence rose while field results did not, and the pilots felt comfortable with GPS despite poor baseline performance.
• Reflection needs a purpose. The simulation meta-analysis found no overall added benefit associated with a reflection phase, with different results across learner groups. Those comparisons do not prove that removing a debrief will improve your course.
• Scenarios date quickly. When a procedure, price or limit changes, every branch that mentions it needs rework, so keep figures in the case data rather than scattered through the feedback.

Using scenario-based learning with an AI course builder

In about 30 Coursebox support threads about scenarios since May 2026, the two biggest groups asked what AI role-play scenarios are and who can use them (8), and how to edit or steer what the AI wrote inside a scenario (7). The second group is about control: people want the scenario to say what they designed. So I tested how much of a designed scenario survives a brief.

On 11 October 2026 I briefed Coursebox to build the hot water course. Coursebox is our product, so weigh my account with that in mind. The brief described the learners and the agency, gave every fact in the case data with the figures, set the three deliverables and asked for eight sections in order: a short "Before you start", the Monday email, the three decisions, each with three options and the consequence shown before any explanation, an open-text write-up with a model file note, a debrief and a wrap-up. It also said not to add multiple-choice recall questions or any tenancy law.

The first outline turned the three decisions into multiple-choice quizzes, which is the scenario with questions that Clark describes. I sent one revision naming the fix: put the learner in the role, show the consequence of each option first, then the senior property manager's note, and do not mark options correct or incorrect. The second outline had the eight sections in the order I asked for, with no quizzes, and each decision page used a Scenario block that asked "What do you do?".

Most of it then worked. The trigger page kept every detail of the tenant's email: the two-bedroom unit, 8:10 am, no hot water since Saturday, the drip three weeks earlier and the threats about rent and the tribunal. I ran the first decision in preview and chose to tell the tenant I would chase the owner. The scenario did not mark me wrong. It answered with a consequence, "The tenant still does not know what happens next, and there is no time frame", then gave me the tenant's next message and a new set of choices. The second decision page kept the $380 valve, the $500 limit and the $800 quote rule.

The issue was the case data. The trigger page showed six case file cards with the right titles, such as Procedure extract, Agreement notes and Plumber detail, but none of them held the facts I had supplied. They described what a file might contain instead: "The file may show the tenancy details and any prior maintenance communication." So on the page where learners classify the repair, there was no limit, no quote rule and no price. A scenario without its case data asks learners to guess, not to decide.

I checked the outline and four rendered pages and ran one branch of one scenario; I did not run the course with property managers. A working branch shows that the branch works, not that anyone's judgement improves. The consequence the scenario wrote for my first choice also differed from the one in my brief, so the consequences an AI writes need the same read as the facts.

To fix the case data, I typed the six documents' facts into the cards in the page editor: the procedure extract word for word, the maintenance log, the $500 limit and the owner's three-quote rule, the $380 valve and the $1,650 tank, the rent ledger and the people. Then I reloaded the learner view and checked each figure was there. With the cards filled, the limit, the quote rule and the prices are on the page where learners classify the repair.

Putting scenario-based learning to work

Scenario-based learning puts learners in a work role, hands them a realistic situation with the facts they need, and lets their decisions play out, with guidance and feedback along the way. It works best for judgement tasks that are rare or risky on the job, with learners who already know the setting. Ruth Colvin Clark's six components are the most useful check on whether a "scenario" really is one: if it lacks a task deliverable, a trigger event or feedback after each choice, it is not a scenario in her sense.

Use it where the job turns on judgement, keep directive teaching for procedures, and give new starters a worked example before they decide alone. Do not judge a scenario course by how confident learners feel afterwards. Judge it by what they do in a fresh case, rated by someone who knows the job.

Your next step: take one judgement task from a course you are building, write its trigger event and list the documents a learner would need to decide, then show both to the person who does that job best before you write a single option.