Case study · Scenario-based eLearning

Finding the Real Reason

A branching simulation that teaches salespeople to stop answering the question they were asked and start finding the one the customer didn't ask.

Experience the project → Jump to the process
A dialogue screen from the course: a customer stands in a kitchen showroom, her line in a blue speech bubble above her, and four possible replies laid out for the salesperson.

Audience

Showroom salespeople at a kitchen-remodeling retailer. New hires and anyone whose consultations aren't converting.

My role

Everything. Needs analysis, action map, script, characters, visual design, branching logic, build, voice production, publishing.

Tools

Articulate Storyline 360, ElevenLabs, Python, Freepik, Audacity. Published as a self-hosted web build.

The problem

Mark is good at his job. He knows the catalogue, he answers every question accurately, and customers leave the showroom happy. They just don't come back.

Mark stands alone in the showroom. A note in the foreground reads: Three months. 23 consultations. 2 sales. October: 0.
The opening screen of the course. The business problem stated before any training content appears.

Twenty-three consultations over three months produced two sales. Nothing in that number points at product knowledge, and more product training would not have moved it. The gap is in what happens in the first four minutes of a conversation.

A customer walks in and says "I need these cabinets, how long and how much." That sentence is a request for information, and a well-trained salesperson answers it well: six to eight weeks, twelve thousand and up. Accurate, helpful, and the end of the conversation. What the sentence is not is the reason the person drove to a showroom on a Saturday.

Close-up of Mark, thinking: I answer every question they ask. They leave happy. They don't come back.
Mark's own diagnosis, from the intro. He is not failing at the job as he understands it — which is exactly why telling him to try harder would not help.

This is a behavior problem, not a knowledge problem, and the distinction decided the whole design. People don't act on what they've been told; they act on what matters to them. You can tell a salesperson that discovery questions improve close rates and he will agree with you and keep quoting lead times, the same way a smoker agrees that smoking is harmful. The only thing that changes the behavior is practising the behavior, in a situation close enough to the real one that it transfers.

The solution

One shift in the showroom. Three customers. Each one says what they want and none of them says why they came. The learner has to find out.

A full-screen text slide: Three people are coming in today. Each of them has a reason for being here. None of them will tell you what it is. Your job isn't to tell them about kitchens. It's to find out why they came.
The entire instructional premise, stated once, in forty words, before the first customer walks in.

There is no lesson, no model to memorize, no list of discovery questions. The learner plays Mark through fifteen decision points — five per customer. At each one the customer has said something, and four possible replies sit on the screen. One of them moves toward the real reason. The other three are things a competent salesperson would plausibly say, and each of them gets its own answer from that customer.

I wrote the objectives against Bloom's taxonomy before designing anything, because the level decides the format. Everything here sits at analysis and above, which is precisely why a quiz could not have tested it. You cannot multiple-choice your way to "notice that she is embarrassed about her kitchen"; you have to put the learner in the conversation and let them miss it.

Learning objectives

  1. Distinguish a customer's stated request from the motive behind it, using evidence from what they say and how they behave. Analyze
  2. Evaluate four plausible responses and select the one that addresses the motive rather than the stated request. Evaluate
  3. Apply the same approach to an unfamiliar customer whose situation shares none of the surface details of the previous one. Apply · transfer

The third one is why there are three customers and not one. Denise hides embarrassment, Ray and Carol hide a disagreement, Tony hides his mother — if the learner could pass by recognizing a pattern from the first conversation, the course would be testing memory instead of judgement.

Denise came in for cabinets. Her daughter is getting married in April and the reception is at her house, and her son-in-law's mother's kitchen looks like a magazine. She will never say that to a salesperson. She says "how long does it take and roughly what am I looking at."

Close-up of Denise, unhappy: Yes. His mother's house looks like a magazine. Ours is a 1993 kitchen where the doors don't shut.
The moment the course exists for. Four replies earlier, this woman was asking about quartz countertops.

My process

Ten days of a software trial, one person, no budget. That constraint set the order of work: logic and a working branch first, graphics and polish after, and a publishable link from day one so there would be something real at the end regardless.

  1. Action mapping

    I started from the business number and worked backward to behavior, then to practice. Everything that didn't change a decision in the showroom got cut — which is why the course contains no history of cabinet construction, no materials glossary, and no model named after anyone.

    BUSINESS PROBLEM BEHAVIOR THAT MOVES IT HOW THEY PRACTISE IT 23 → 2 consultations to sales, three months Ask about the person before talking about the product Name the real reason out loud in the customer's own words Tie the product to that reason not to its features 15 decision points 3 customers 4 replies each 45 written consequences the customer's mood answers back Everything that didn't change one of these three got cut.
    The action map. The middle column is the course; the left column is why it exists and the right column is the only form of practice that teaches it.

  2. Writing the conversations

    The script came before anything was built, as a plain table: what the customer says, what they are actually thinking, four possible replies, the weight of each, and the customer's reaction to each. Writing it this way meant the hard work — making three wrong answers that a real salesperson would genuinely choose — was done before a single slide existed.

    The wrong answers are the design. If they are obviously wrong, the learner pattern-matches instead of thinking. So each distractor is something defensible: offering a cheaper line when cost was never the issue, promising the timeline will be met, reassuring someone that Shaker cabinets are a safe choice. All three are things that get said in showrooms every day, and all three end the conversation.

    The three customers also had to fail differently. Denise is embarrassed and will not admit it. Ray and Carol disagree with each other and neither one has said the real thing out loud. Tony is buying a forty-dollar faucet and the reason is his eighty-three-year-old mother. One script pattern, three different ways of hiding.

  3. Branching that never dead-ends

    My first branching map punished a second wrong answer by ending the conversation: the customer said "I'll think about it" and left, and the learner restarted that customer from the beginning. It was a reasonable-looking rule and I cut it.

    A learner who gets locked out stops trying. That is learned helplessness, and it is the opposite of what a practice environment is for. The mistake was also already punished three separate ways — the customer's mood drops, the reply counter climbs, and the final star rating falls — so the ejection added a penalty on top of a penalty and removed the learner from the only thing that teaches: the conversation.

    V1 — SCRAPPED Choice screen Mood −1 "I'll think about it." Customer leaves. Restart this customer from zero wrong wrong again Three penalties for one mistake, and the learner loses the conversation. V2 — SHIPPED Choice screen The customer answers that specific reply Mark's reply The customer opens up Next screen right · mood +2 wrong mood −1 or 0 · one reply spent back to the same screen, that option now gone
    The rule I removed, and what replaced it. In v2 a mistake still costs — mood, a reply, a star — but the conversation continues and the used option disappears, so the number of replies spent is the score. There is no way to get stuck.

    Then the harder half of the same problem

    Running my own build of Tony's scene, I hit something worse than the fail rule. Mark asks "How old is she?" — a decent question — and Tony replies "I'll think about it." Every wrong answer on a screen shared one generic brush-off line, so the dialogue read as nonsense.

    Before Close-up of Tony answering 'I'll think about it' to the question 'How old is she?'
    One brush-off line shared by every wrong reply on the screen.
    After The same close-up, now with Tony answering 'Eighty-three.'
    Tony answers the question that was actually asked.

    Feedback has to be aware of what the learner said. Otherwise all they learn is "that one wasn't it" — which is a quiz with better art, not a conversation. So every wrong reply got its own answer in the customer's voice: forty-five of them across fifteen screens.

    Forty-five lines did not mean forty-five new slides. Each wrong button now sets a text variable before it jumps, and the single return slide displays that variable. One extra trigger per button instead of thirty extra slides.

    Storyline trigger panel showing three wrong buttons, each setting the stallText variable to its own line before jumping to the same slide.
    Three wrong buttons, three different lines, one destination. The rule I settled on: variables where the content branches, static text on the slide where it doesn't.

  4. Characters, shots and what the picture is doing

    The learner's job is to read a person, so the people had to be readable. Each character got a set of poses and expressions, and the arc of a conversation is carried by which ones appear: Denise starts with a folder clamped to her chest, holds up a printout while the talk is about cabinets, lowers the folder when the real reason is named, and ends with open hands.

    Pose sheets for all six characters: Mark the salesperson, Denise, Ray, Carol, Tony and Linda the manager, each in six poses.
    Six characters, six poses each. Scroll the panel. Expressions were drawn as a second set of eight per character and swapped as object states.

    The scene is cut like film rather than laid out like a form. A customer enters in a wide shot, speaks in close-up, and the choice is made in the wide shot where both people are visible. The close-up is not decoration: it is where there is room for a long line, and it is where the gap between what someone says and what they mean is visible on a face.

    Wide shot: Denise speaks while a small thought bubble in the corner reads 'I don't want my daughter embarrassed by us...'
    The thought bubble sits on the choice screen, where the decision is made — not after it.
    Close-up of Mark delivering his reply about lighting rather than cabinet doors.
    Mark's correct reply gets its own close-up, so the learner hears their own choice spoken.

    Those thought bubbles appear only for the first customer, on three of her five screens. It's deliberately fading support: with Denise the course shows the learner what looking underneath the words means, and with Ray, Carol and Tony it stops helping. By the third customer nobody is narrating the subtext.

    Two people in one scene created its own problem. Ray and Carol are a couple who disagree, and the learner has to tell who is speaking without name tags. Carol always stands on the left and Ray on the right, her bubble is blue and his is dark, and when both speak her line sits above his so reading order matches speaking order.

    Ray and Carol both speaking: Carol's blue bubble above, Ray's dark bubble below.
    Two speakers, no labels. Position and color do the attribution.
    Ray turns toward Carol for the first time and says 'You never said your back was that bad.'
    Carol holds her lower back one screen before she admits anything. Ray only turns to look at her on the next one.

  5. Building it

    Five slide masters carry everything that repeats — the mood meter, the reply counter, the progress panel — so a change to the interface happens once rather than eighty-one times. Eighty-one slides, five scenes, named by customer and step (D1 choice, D1 stall, R3 mark) because at that count a mislabeled jump target is very hard to find.

    Storyline slide master panel showing five layouts: dialogue, close-up left, close-up right, achievement and office.
    Five masters. Shared interface lives here, not on the slides.
    Story view with five collapsed scenes: Intro, Denise, Ray and Carol, Tony, Linda.
    Five scenes. Each customer was built by duplicating the previous one.

    Eight variables run the whole thing: the customer's mood, replies spent on the current customer and overall, a star score per customer, a total, and the text variable that carries the right return line. By the end they were referenced about five hundred times from triggers.

    Day one Storyline variables window: seven variables, all with a use count of zero.
    Seven variables, used zero times.
    Day eight The same window: eight variables with use counts of 242, 108, 60, 28, 28, 28, 13 and 3.
    Eight variables, roughly five hundred references.

    That use-count column turned out to be an audit tool. repliesTotal showed 2 where it should have shown 3; one Add trigger had been deleted by accident while cleaning up a button in Tony's scene. The Usage window named the exact slide, which is considerably faster than replaying the course looking for a number that is quietly wrong.

    The Usage window for the repliesTotal variable, listing the three triggers that reference it.
    Finding a missing trigger by reading a count instead of replaying the course.

    A scale that stopped responding

    The mood meter originally ran 0–10, starting at 4, with a correct answer worth +2. A perfect run hit the ceiling on the third of five screens and the face stopped moving for the rest of the conversation — at exactly the point where Denise admits she's ashamed of her kitchen, which is not a moment for a smiley face already pinned at maximum.

    I kept the weights, because four levels of answer (+2, +1, 0, −1) map to the model the course is built on, and doubled the range instead: 0–20, starting at 8, face equals mood halved. The twenty-two expression triggers on the master changed from equality tests to ranges.

    Before Trigger list using equality tests: set face to m00 if mood = 0, m01 if mood = 1, and so on.
    One face per value. Ceiling reached halfway through.
    After The same triggers rewritten as ranges: set face to m00 if mood is between 0 and 1, m01 if between 2 and 3, and so on.
    One face per pair of values, so a perfect run still has headroom at the end.

    The lesson cost me an hour of hunting: the same logic had been duplicated onto three masters, and I fixed only one. The face behaved on the choice screens and jumped around on the close-ups. If identical logic lives in more than one place, the list of places is part of the build.

  6. Voice

    Without audio the course felt like a silent test. I tried Storyline's built-in text-to-speech first and marked up two dozen lines with pauses and emphasis; it got better and stayed flat. In a conversation that runs on embarrassment and a man's mother, a level voice kills the scene.

    So: six voices through ElevenLabs, one per character, plus a seventh narrator who is not one of them. Emotion is set by a tag before each line, and the tags follow the course's own logic rather than taste — a return line is tagged by the weight of the answer that triggered it. Pick something careless and the customer sounds annoyed; pick something merely unhelpful and they sound unconvinced. The feedback is in the voice, not only in the text.

    A spreadsheet of lines: file name, speaker, emotion tag, and text.
    One table: file name, speaker, emotion tag, line. Ninety-eight rows.
    Terminal output showing a Python script generating audio files one by one through the ElevenLabs API.
    A Python script turns the table into MP3s named after the slides. Re-running it only generates what's missing.

    Naming each file after its slide meant importing ninety-eight clips without checking a single one against a list. Re-recording a bad take is deleting one MP3 and running the script again.

How the learner knows how they did

Each customer ends with a result screen: the star rating, the exact number of replies it took, and one line that names what actually happened in that conversation. The scoring is simply efficiency — five replies is the minimum, so five replies is three stars.

Result screen: Denise signed, three stars, five replies, with the line 'She came in for cabinets. She left with a kitchen she won't be embarrassed to host forty people in.'
The line under the stars restates the whole point of the exercise.
Result screen after the third customer, with all three progress medallions filled and the button 'Linda wants to see you'.
The day's progress panel fills up as customers are closed.

The shift ends in the manager's office. Linda has three things she might say, chosen by the total across all three customers — and the weakest ending is the most useful one.

Three versions of the final conversation with Linda, at nine, seven and four stars.
Nine stars: a promotion. Seven: let's review the recordings together. Four: you closed all three, now count how many times you talked about cabinets while the person was talking about their life.

Then the model, once, at the very end — never before. The learner has already spent twenty minutes discovering that arguments move people less than motives do. Naming it first would have turned the experience into an illustration of a lecture.

The closing diagram on a phone: three circles labelled Knowing, Feeling and Doing, with a solid arrow from Feeling to Doing and a dashed one from Knowing, and the line 'Ask why they came, not what they want.'
The closing screen, on a phone. Knowing pulls weakly at Doing; Feeling pulls hard. That is also the course's own scoring: an argument is worth +1, reaching the real reason is worth +2.
81slides
8variables, ~500 references
45written consequences
98voiced lines

Results, and what I'd actually measure

This is a portfolio project. No one has run a sales team through it, so I am not going to claim a lift in close rates. What I can say is what the build produced and what I would put in place on day one of a real rollout.

The whole thing — fifteen branching decision points, three complete conversations, three endings and a working published link — took four days of build inside a ten-day software trial, with voice and polish added after. It runs in a browser on desktop, tablet and phone, with no LMS.

If this were deployed, here is what I'd instrument, against Kirkpatrick's four levels. Most training programs stop measuring at level 1, which is why so many of them survive without ever proving anything.

Levels 3 and 4 are the only ones that answer the question the business asked, and neither of them can be collected inside the course. That has to be agreed with whoever owns the sales numbers before the first slide is built, which is a conversation I'd have at the start of a real engagement rather than at the end.

What I'd do differently

Two things. The first screen I built was the one where the mechanics were still being worked out, and it quietly kept a bug the later screens didn't have — I should have rebuilt it from the finished template rather than patching it. And I would write the return lines at the same time as the options, not after: separating them is what produced a customer who answered a question nobody asked.

See it for yourself

Twenty minutes, three customers, no login. The fastest way to understand the design is to pick the wrong answer on purpose and watch what the customer does with it.

Experience the project → Back to all work