GreenBox is a produce-box startup that delivers weekly boxes of local farm produce to subscribers in Perth. They’ve used discovery workshops to build shared understanding and reach 200 subscribers – but now they need to grow to 1,000, and the techniques that got them here won’t answer the questions they’re facing next.
GreenBox has hit 200 subscribers. It took longer than anyone expected and involved more rework than anyone wants to admit, but the number is real. Two hundred people paying real money every week for a box of local produce.
Maya secured the next funding round. The board’s new target: 1,000 active subscribers within six months. That’s five times the current base in half a year. The discovery techniques from the earlier series – Event Storming, Example Mapping, BDD, Impact Mapping, User Story Mapping – got them here. The team knows the domain. The software works. The delivery logistics are solid.
Maya tells herself this on the coastal track at 5:45am on Monday, feet landing on packed sand, breath steady. The number is real. Two hundred. She’s proved something. She should feel good about it. But the board call last Thursday sits in her chest like a stone she swallowed. The new target isn’t a vote of confidence. It’s a test. Five times the current base. Six months. Diane had said it kindly enough – “We’re excited about the trajectory, Maya” – but the slide behind Diane’s head had a red line showing where the funding ran out if they didn’t hit it.
She showers, makes coffee, sits at the kitchen table with her laptop. Nadia is still asleep. The photo of her parents’ farm catches the morning light from the window – the one where her father is standing in front of the converted dairy shed, the year they switched from dairy to mixed produce. He’s smiling but his eyes look tired. Maya remembers that year. Two seasons of no income while they learned new skills. Her mother picking up extra shifts at the local school canteen. Her father sleeping in the shed some nights because he was irrigating by hand.
She opens the subscriber dashboard. Two hundred and six. Net gain of three last week.
Three.
But there’s a problem hiding in the numbers.
The churn problem
Churn is 8% monthly. For every ten new subscribers the team signs up, they lose three or four existing ones.
Sam pulls up the spreadsheet on Monday morning and walks the team through it. “We added forty-two new subscribers last month. We lost sixteen. Net gain: twenty-six. If that ratio holds, we need to sign up roughly sixty new subscribers every month just to net the growth we need. That means our acquisition has to triple – while we’re also haemorrhaging paying customers.”
Maya does the maths on the whiteboard. At 8% monthly churn, even if they doubled their acquisition rate, they’d plateau somewhere around 600 subscribers. The churn eats the growth. They’d never hit 1,000.
“We need to understand why people leave,” Maya says.
Tom nods. “Let’s Event Storm it.”
The wrong tool for the job
It’s a natural instinct. Event Storming worked brilliantly when the team needed to understand the subscription domain. It mapped every business event from farm availability to box delivery. The wall of sticky notes gave everyone a shared picture of what happens in the business.
So the team books the meeting room, grabs the sticky notes, and starts mapping the cancellation flow.
Orange notes go up. “Subscriber Cancelled.” “Cancellation Feedback Requested.” “Feedback Received.” “Refund Processed.” “Subscriber Paused Instead.” Blue command notes: “Cancel Subscription.” “Submit Feedback.” “Process Refund.”
After an hour, the wall has a clean timeline of what happens when someone cancels. The process is well mapped. The events are in order. The commands are clear.
Lee stands back and looks at it. He’s been quiet for most of the session, which is unusual for him.
“This is a good map of the cancellation process,” he says. “But I don’t think it answers your question.”
Maya frowns. “We’re mapping why subscribers leave.”
“No,” Lee says. “You’re mapping what happens when they leave. Those are different questions. Event Storming tells you what happens in the business. It doesn’t tell you what’s going on in the customer’s head.”
He’s right. The map shows that subscribers cancel, that some provide feedback, that the system processes the cancellation. But it says nothing about why they decided to cancel in the first place. The motivation lives outside the system. It lives in the customer’s life – in their kitchen on a Tuesday evening, in the moment they decide this subscription isn’t worth it any more.
Event Storming is designed to map domain events and business processes. It’s extraordinarily good at that. But customer motivation isn’t a domain event. It’s psychology. And you need a different lens.
Priya tries to push the Event Storm further. She adds a pink hotspot note next to “Subscriber Cancelled” that reads “Why?” But the pink note just sits there. In a normal Event Storm, pink notes trigger conversations about process ambiguity or domain disagreements. This one triggers shrugs. Nobody in the room knows why subscribers cancel. The information isn’t in the system. It’s in the heads of people who’ve already left.
Tom tries another angle. He adds a blue command note: “Decide to Cancel.” But that’s not a system command – it’s a human decision that happens before the subscriber ever opens the app. There’s no event that precedes it in the domain model. The decision happens over dinner, or in the shower, or during a frustrated scroll through the fridge. By the time the subscriber clicks “Cancel,” the decision was made days ago.
The Event Storm has hit a wall. Not because the technique is flawed, but because the question lives outside its scope.
Tom is frustrated. “So we wasted an hour?”
“Not wasted,” Lee says. “You now have a clean map of the cancellation flow, which you’ll need when you build retention features. And this is a useful lesson: knowing when a technique has reached its limit is as valuable as knowing how to use it. You tried the tool you knew. It showed you what it could. Now you know you need something else.”
Jobs to Be Done
Lee draws a simple diagram on the whiteboard. A stick figure, an arrow, and a box labelled “GreenBox.”
“Clayton Christensen’s framework,” he says. “Jobs to Be Done. The core idea is this: customers don’t buy products. They hire them to do a job in their life. Your product isn’t competing with other produce boxes. It’s competing with whatever else the customer could hire to do the same job.”
He lets that sink in.
“So the question isn’t ‘why do people subscribe to GreenBox?’ The question is ‘what job are they hiring GreenBox to do?’”
Priya tilts her head. “Isn’t the job obvious? They want fresh local vegetables.”
“Maybe,” Lee says. “But if that were the job, they could go to a farmers’ market. Or join a food co-op. Or grow their own. What does GreenBox do that those alternatives don’t?”
Nobody answers immediately. It’s a harder question than it sounds.
Lee tells the team about Christensen’s famous milkshake study. A fast-food chain wanted to sell more milkshakes. They did focus groups, improved the recipe, added flavours. Sales didn’t move. Then they watched what actually happened at the counter. Half the milkshakes were sold before 8am, to commuters. The commuters didn’t want a delicious milkshake. They wanted something thick enough to last a boring forty-minute drive, something they could consume with one hand, something more interesting than a banana but less messy than a bagel. The job was “make my commute less tedious.” Once the chain understood that, they made the milkshake thicker and added bits of fruit for texture. Sales went up 40%.
“The milkshake wasn’t competing with other milkshakes,” Lee says. “It was competing with bananas, bagels, and boredom. GreenBox isn’t competing with other produce boxes. It’s competing with whatever else your customers could do to solve the same problem in their lives.”
Lee explains the framework. In JTBD, every job has three dimensions:
Functional: What does the customer need to accomplish? (Get food on the table.)
Emotional: How do they want to feel? (Like a good parent. Like they’re eating well. Like they’re not contributing to industrial agriculture.)
Social: How do they want to be seen? (As someone who cares about local food. As someone who supports farmers.)
“The functional job is usually the least interesting,” Lee says. “Everyone can get food on the table. Supermarkets are functional. The emotional and social jobs are where your competitive advantage lives – or doesn’t.”
Talking to actual humans
Lee suggests interviews. Not surveys, not analytics, not A/B tests. Actual conversations with actual people.
“You need three groups,” he says. “Five people who are active, loyal subscribers. Five who cancelled. And five who considered subscribing but didn’t. Each group tells you something different.”
Maya is sceptical. “Fifteen interviews? That’s going to take weeks.”
“It’ll take three days,” Lee says. “Thirty minutes each. You and Sam can do five a day. The hard part isn’t the time – it’s asking the right questions.”
Lee coaches Maya and Sam on interview technique. The key rules:
Don’t ask “why do you subscribe?” People will give you a rationalised answer that sounds reasonable but isn’t the real reason. Instead, ask about the timeline: “Walk me through the moment you decided to sign up. What was happening in your life? What were you doing before? What changed?”
Don’t ask “what features would you like?” People will invent features they’d never actually use. Instead, ask about struggles: “Tell me about the last time you were frustrated with dinner. What happened?”
Listen for the switch. The moment someone moved from their old solution (supermarket, takeaway, cooking from whatever’s in the fridge) to GreenBox. What triggered it? What were they giving up? What were they hoping for?
Don’t try to validate your product. You’re not looking for compliments. You’re looking for the truth about what role GreenBox plays – or fails to play – in someone’s week. If the truth is “I signed up because my sister-in-law guilted me into it and I cancelled after two weeks,” that’s useful. Uncomfortable, but useful.
Lee also explains why three groups matter. Active subscribers tell you what job GreenBox is doing well. Churned subscribers tell you where GreenBox failed the job. Non-subscribers tell you what barriers prevent people from even trying. Together, the three groups give you the full picture: what works, what breaks, and what never starts.
Sam is sceptical. “Can’t we just send a survey? We could reach all two hundred subscribers.”
“You could,” Lee says. “And you’d get two hundred shallow answers. Surveys tell you what people think. Interviews tell you why they think it. You need the why.”
The team also uses LLMs during this process. Maya records each interview on her phone (with permission), and they use an LLM to transcribe the recordings. That saves hours of note-taking. After the first batch of five interviews, Maya pastes all five transcripts into a conversation with the LLM and asks it to identify recurring themes, unexpected phrases, and emotional language. The LLM spots patterns across transcripts faster than a human reader can – not because it understands the customers better, but because it can hold five long conversations in context simultaneously and cross-reference them.
It’s a tool, not a replacement for judgement. Maya reads every transcript herself. But the LLM highlights things she might have skimmed past. In one case, it flagged a phrase Maya had glossed over: a churned subscriber who said “I felt guilty every time I opened the fridge and saw the kale wilting.” Maya had coded that as “didn’t like kale.” The LLM tagged it as emotional language related to guilt and waste anxiety – a different and more useful categorisation.
The interviews themselves are harder than anyone expected. The first two feel stilted. Maya keeps asking leading questions (“So you subscribed because you like local produce, right?”) and Sam has to gently redirect her.
The third interview is a disaster.
His name is Greg. He cancelled six weeks ago. He arrives at the cafe ten minutes late, already irritated. Maya buys him a flat white and starts with the script: “Walk me through the moment you decided to sign up for GreenBox. What was happening in your life?”
“I’ll tell you what was happening. My wife found you on Instagram and signed us up without asking me. Then I was the one dealing with the box every week.”
Maya nods. Good start – unscripted, honest. She should listen. Instead, she says: “And what was the experience like for you?”
“Terrible. You sent me beetroot three weeks running. Three weeks. I told your support team after the second time. The third week I opened the box and there it was again. Purple. Staring at me.”
Maya feels heat rise in her neck. “We track all dietary preferences and – “
“No you don’t.” Greg puts down his coffee. “Or if you do, your system is broken. I sent two emails. Nobody responded to the second one.”
“I’m sorry about that. We’ve improved our – “
“I’m not here for an apology. You asked to talk. I’m talking. You want to know why I left? I left because I spent more money on your box than I would have at Hartland Group and I got ingredients I didn’t want that nobody helped me cook. I switched to Freshly. Seven dollars cheaper and the delivery tracking is better.”
Maya blinks. “Freshly?”
“Yeah. The Sydney mob. They launched in Perth last month. The produce isn’t as good but at least I know what I’m getting.”
Maya writes down “Freshly” on her notepad and underlines it twice. She wants to ask more about this company she’s never heard of, but Greg is still talking.
“Look, I could tell you cared. The little notes in the box about which farm the carrots came from – that was nice. But nice doesn’t matter when I’m standing in my kitchen at six o’clock with a kohlrabi and no bloody idea what to do with it.”
The interview ends after twenty minutes. Greg shakes her hand and leaves. Maya sits at the cafe table, staring at her notepad. Lee, who’d been observing from the next table with his laptop open, walks over and sits down.
“That was rough,” he says.
“He was rude.”
“He was honest. And you broke the first rule.”
Maya knows what he’s going to say.
“You got defensive,” Lee says. “The moment he said the system was broken, you stopped listening and started defending. ‘We track all preferences’ – you were protecting the product instead of hearing the customer.”
“Because what he said wasn’t true. We do track preferences.”
“Do you track his preference? Did anyone action his emails?”
Maya opens her laptop and searches the support inbox. She finds Greg’s first email – Sam had responded with a template apology. The second email, sent four days later, has no reply. It sits in the inbox, unread, between forty other messages from the same week.
“We missed it,” Maya says quietly.
“You missed it. And Greg’s complaint about receiving beetroot three weeks in a row – that’s the same kind of problem Tom found with the turnip substitution. The system doesn’t track implicit preferences unless you personally remember them. Mrs Patterson’s beetroot aversion, Greg’s beetroot aversion – same gap, different customers.”
Maya is stung. But Lee isn’t finished.
“That was the most useful twenty minutes of the whole batch,” he says. “Greg gave you a system failure, a competitor name, and the clearest articulation of the core problem anyone’s said yet. ‘I was standing in my kitchen at six o’clock with a kohlrabi and no idea what to do with it.’ That’s your answer. And you almost missed it because you were defending instead of listening.”
Maya nods slowly. She writes down Greg’s kohlrabi line and circles it.
That evening, she goes home and searches for Freshly. She finds a polished website with a slick onboarding flow, an app with real-time delivery tracking, and a pricing page that makes her stomach tighten: $18 per week for a standard box. She clicks through their Instagram – sixty thousand followers, professional photography, a VC funding announcement from four months ago. Twelve million dollars. Series A.
She sits at the kitchen table for a long time. Nadia comes in from a late physio session and finds her there, laptop open to Freshly’s website, a glass of wine untouched beside her.
“What’s that?” Nadia asks.
“Competition. Well-funded competition.”
Nadia looks at the screen. “Their boxes look nice.”
“They’re not local. They buy wholesale from the markets.”
“Does that matter?”
Maya doesn’t answer. It’s the question she’s been asking herself for the last hour. She closes the laptop and goes to bed, but she doesn’t sleep. At midnight, Nadia finds her in the kitchen, reorganising the cupboards. Tins arranged by expiry date. Spices alphabetised. The jars of preserved lemons that Maya’s mother sent from Margaret River lined up on the counter like soldiers.
Nadia leans against the doorframe. “You’re doing the cupboard thing.”
“The cupboard thing” is what Nadia calls it when Maya processes something painful by imposing order on something physical. The last time was when Maya’s father had his heart scare. The time before that was when GreenBox’s first pilot subscriber cancelled.
“I’m fine,” Maya says.
“You’re alphabetising cumin at midnight. You’re not fine.”
Maya puts down the jar. “The customers don’t care about local sourcing, Nadia. We interviewed fifteen people. Three of them – three – mentioned local as the main reason they subscribe. I built the whole brand around it. The fifty-kilometre promise, the farm stories, all of it. They don’t care.”
Nadia is quiet for a moment. “They care about something, though?”
“Convenience. They care about not having to think about dinner. That’s it. That’s the product.”
“Is that a bad thing?”
“It’s a different thing. It’s a completely different business than the one I thought I was building.”
Nadia sits down at the kitchen table. “You built the brand around what matters to you. Now you’re finding out what matters to them. Those can both be true.”
Maya looks at the preserved lemons. Her mother made them last summer, in the kitchen of the small house in Margaret River that her parents moved to after the farm was sold. The recipe is her grandmother’s, from Taiwan. Three generations of women preserving food with their hands.
“I know,” Maya says. “I just need a minute.”
She calls her mum the next morning, before the coastal run. Her mother answers on the second ring – she’s always been an early riser.
“Mum, did it bother Dad that people didn’t care about where their food came from? When you were farming?”
Her mother laughs. “Your father didn’t farm because people cared about farming. He farmed because people needed to eat. The caring was his. The eating was theirs.”
Maya stands at the kitchen window watching the sky lighten over Fremantle. Her mother’s words land somewhere deep.
By the fourth interview, Maya finds her rhythm. She learns to sit with silence – the pauses after a question where the interviewee is actually thinking, not waiting to be prompted. Those pauses produce the most honest answers.
One churned subscriber, a man named Patrick, gives them a fifteen-minute story about his Tuesday evenings that becomes the team’s touchstone for the rest of the quarter. He describes getting home at six, tired, opening the GreenBox, seeing ingredients he doesn’t recognise, googling recipes on his phone while his kids argue about homework, giving up, ordering pizza, and then feeling guilty about the $25 box of vegetables sitting unused on the counter. “I was paying twenty-five dollars a week to feel bad about myself,” he says. That sentence ends up on a sticky note in the office.
What the interviews reveal
Three days later, the team has fifteen transcripts and a wall of extracted quotes. Maya and Sam present the findings on Thursday morning. The room goes quiet.
Active subscribers:
The loyal subscribers barely mention vegetables. They mention relief.
“I don’t have to think about what to cook on Tuesday. The box arrives and dinner is decided.”
“It’s one less thing to worry about. I get home, I open the box, and I know what we’re eating.”
“I used to spend twenty minutes in the supermarket just wandering, trying to figure out what to buy. Now I don’t go to the supermarket on Tuesdays at all.”
One of the active subscribers is Mrs Patterson – the same Mrs Patterson whose beetroot aversion Maya has been carrying around in her head since the Example Mapping sessions. Mrs Patterson is 63, lives alone on Stirling Highway, and has been subscribed since the second week of the pilot. In person, she’s warm and precise – the kind of woman who offers you tea before you’ve sat down and then tells you exactly what she thinks.
“I just open the box and trust what’s inside,” she says. “Except when there’s beetroot.” She smiles. “I’ve emailed about that. But honestly, it only happened twice. The rest of the time, the box is lovely. I don’t even look at what’s coming. I just know that when I get home on Tuesday, dinner is sorted.”
Maya writes down dinner is sorted and puts a star next to it. It’s the third time she’s heard that phrase, or something close to it, in four interviews.
“I don’t even know what’s in the box most weeks,” Mrs Patterson adds. “I just know I don’t have to think about it.”
Jas is sitting in on this interview – Lee suggested that each team member observe at least one. She’s in the corner with her Moleskine open, sketching while she listens. When Mrs Patterson says “dinner is decided,” Jas draws a quick napkin-style sketch: a box opening, a recipe card visible on top, and underneath it the words dinner decided. She underlines it. Then she underlines it again.
She doesn’t know yet that this phrase will become the tagline that reshapes the entire brand. Right now it’s just two words on a napkin. But the words came from a real person sitting across a real table, and that’s what gives them weight.
Another active subscriber, a woman named Danielle, mentions something in passing that Maya almost misses: “I check your website on my phone every Wednesday to see what’s coming in this week’s box.” Maya writes in her notebook: they want a preview on their phone. Not on a laptop – on their phone, standing in the kitchen, planning the week. Jas flips to a fresh page in her Moleskine and sketches a quick mockup next to the “dinner decided” napkin – a phone screen showing a push notification: “Your box this week: sweet potato, kale, leeks, apples.” The app idea isn’t born from a tech decision. It’s born from watching a customer check her phone on a Wednesday.
The job isn’t “get fresh local produce.” The job is “eliminate the mental load of deciding what to cook.” GreenBox removes a decision from their week. The vegetables are good, sure. But the relief is the product.
Churned subscribers:
The people who cancelled tell a starkly different story.
“The vegetables were great but I’d open the box and have no idea what to do with half of it. Kohlrabi? What do I do with kohlrabi?”
“I spent more time googling recipes than I would have spent at the shops.”
“It actually added stress instead of removing it. I had all these beautiful vegetables and the guilt of not knowing how to use them before they went off.”
The produce was good. But the box didn’t do the job. The mental load wasn’t reduced – it was just relocated. Instead of “what should I buy?” the question became “what on earth do I do with this?” The job GreenBox was hired for – eliminate dinner stress – wasn’t being done for these people. So they fired it.
And two of the five churned subscribers mentioned Freshly by name. Greg’s wasn’t the only defection. Another, a woman named Louise, said almost casually: “I tried that Freshly thing. It’s not as nice, but it’s easier.” Easier. Not better. Easier. That word sits in Maya’s notes like a splinter.
People who considered but didn’t subscribe:
This group is revealing in a different way.
“I looked at the website and I couldn’t tell what I’d actually get. It just said ‘seasonal produce.’ I need to plan meals for a family of four. I can’t plan around a surprise.”
“The price seemed fine but I didn’t know if we’d actually eat everything. I hate food waste.”
“I was interested but my partner was sceptical. I couldn’t explain what we’d be getting or why it was worth it.”
These people didn’t reject the product. They rejected the uncertainty. They couldn’t see how GreenBox would do the job they needed done.
One non-subscriber, a mother of three named Clare, put it perfectly: “I’m already drowning in decisions. I didn’t want to add another one – ‘will this box work for us this week?’ If I’d known exactly what was coming and what I could cook with it, I probably would have signed up.” She was describing the same job – eliminate dinner stress – but from the outside looking in. The product’s marketing didn’t communicate the job. It communicated the mechanism (“local produce”) without the outcome (“dinner, sorted”).
The insight that changes everything
Maya stares at the quotes on the wall. Priya is the first to say it out loud.
“We’ve been marketing this as ‘fresh local vegetables.’ But that’s not why people stay. They stay because we solve Tuesday night. And they leave because we don’t solve Tuesday night – we just make it a different kind of hard.”
Lee nods. “The job is ‘eliminate weeknight dinner stress.’ Not ‘provide fresh produce.’ The produce is the mechanism. The stress relief is the job.”
Tom leans forward. “So the next feature isn’t better produce selection or more variety or a fancier substitution algorithm. It’s…”
“Recipe cards,” Jas says. She pulls out the Moleskine and opens it to the napkin sketch from Mrs Patterson’s interview – the box opening, the card on top, the words dinner decided. “Simple, fast recipes that use exactly what’s in this week’s box. Open the box, pick a card, cook dinner. No thinking required.”
Sam pulls out his notebook. “And for the people who didn’t subscribe – they need to see what they’re getting before they commit. A preview. ‘This week’s box contains X, Y, Z, and here are three recipes you can make.’ That answers their uncertainty.”
The room is energised in a way it hasn’t been for weeks. Not because recipe cards are exciting technology – they’re not. They’re printed cards in a cardboard box. But they directly serve the job the customer is hiring GreenBox to do.
Priya pushes further. “If the job is ‘eliminate dinner stress,’ then the recipe cards aren’t just a nice-to-have feature. They’re the difference between the product doing the job and the product failing at the job. Without them, we’re delivering ingredients. With them, we’re delivering dinner.”
“Patrick’s kohlrabi problem,” Sam says.
“Exactly. Patrick didn’t need better kohlrabi. He needed someone to tell him what to do with it in twenty minutes while his kids did homework.”
Jas is already sketching card layouts. “One side: a photo of the finished dish. Other side: five steps, pantry staples listed, and a QR code to a video if they want it. Nothing fancy. Nothing that requires a mandoline or a sous vide machine.”
Maya adds a constraint: “Every recipe has to be doable by someone who considers themselves a bad cook. If Patrick can make it, anyone can.”
This is a small moment, but it matters. The team isn’t designing a feature. They’re designing around a specific human being they’ve actually talked to. Patrick isn’t a persona on a slide deck. He’s a real person who told them a real story about feeling guilty on a Tuesday evening. That specificity – that grounding in an actual life – is what JTBD gives you that feature brainstorming doesn’t.
Tom is quiet for a moment. “I was about to spend three weeks improving the substitution algorithm’s scoring model. That would have been the right kind of fast from Series 1 – well-understood problem, concrete examples, testable code. But it doesn’t serve the job. A slightly better substitution algorithm doesn’t reduce anyone’s dinner stress.”
“Right,” Lee says. “The techniques from the earlier series are still useful. You’ll Example Map the recipe card feature before you build it. You’ll connect it to the Impact Map. But JTBD tells you which problem to solve. The other techniques help you solve it well.”
Connecting JTBD to Impact Mapping
The team revisits their Impact Map from the earlier series. The goal is now 1,000 subscribers in six months. But the impacts look different through the JTBD lens.
Before, the impacts were product-focused:
- Subscribers stay subscribed
- Potential subscribers discover GreenBox
- Potential subscribers trust enough to try
Now, they’re reframed around the job:
- Subscribers feel dinner stress reduced (not just “stay subscribed”)
- Potential subscribers see how GreenBox eliminates dinner decisions (not just “discover GreenBox”)
- Churned subscribers get the missing piece that completes the job (not just “win back lapsed users”)
The deliverables shift too. “Improve substitution algorithm” drops down the priority list. “Recipe cards in every box” goes to the top. “Weekly box preview email” jumps up – it serves both retention (subscribers know what’s coming and can plan) and acquisition (potential subscribers can see the value before committing).
Lee points out something subtle. “Notice that the old impacts were about our metrics – subscriber count, discovery rate, trust. The new impacts are about the customer’s experience – stress reduced, decisions eliminated, job completed. That shift matters. When your impacts are about your metrics, you optimise for your dashboard. When they’re about the customer’s life, you optimise for value. The metrics follow, but the thinking is different.”
Tom connects this to something from the earlier series. “It’s like when we Impact Mapped to 300 subscribers and the pause feature came out as the obvious first build. We could trace a line from the feature to the behaviour change to the goal. This is the same structure, but the middle layer – the impacts – now comes from JTBD instead of guesses.”
“Exactly,” Lee says. “The Impact Map is a framework. JTBD fills it with truth instead of hope.”
Maya asks the LLM to help draft the first set of recipe cards. She pastes in this week’s box contents – beetroot, sweet potato, kale, leeks, apples – and asks for three simple recipes, each under thirty minutes, each using only box contents plus basic pantry staples. The LLM produces them in seconds. Jas designs a card layout. Sam sends the cards to the printer. They’ll be in next Tuesday’s boxes.
It costs almost nothing. The LLM writes the recipes. The printing costs cents per card. But if it completes the job – if subscribers open the box and feel relief instead of confusion – it could cut churn in half.
Maya reviews the LLM-generated recipes before they go to print. She catches one that calls for harissa paste – not a basic pantry staple in most Australian households. She asks the LLM to regenerate with a stricter constraint: “Only use ingredients that you’d find in a typical Australian kitchen. Salt, pepper, olive oil, garlic, butter, soy sauce, tinned tomatoes, basic spices. Nothing that requires a special trip to the shops.” The second attempt is better. This is the pattern with LLMs throughout the GreenBox story: they’re fast and capable, but they need human judgement to constrain and refine their output.
Tom raises a technical point. “If we’re generating recipes per box, and box contents vary weekly, we need a system that takes this week’s ingredient list and produces matching recipes. That’s actually a perfect LLM job – it’s pattern-matching and creative text generation with a clear input and clear constraints.”
He builds a prototype that afternoon. A simple script that takes the week’s box contents, sends them to an LLM with the recipe constraints (under thirty minutes, basic pantry staples only, feeds four people), and produces three formatted recipes. Jas drops them into her card template. The whole pipeline – from box contents to print-ready recipe cards – takes less than ten minutes per week.
This is LLMs at their most useful: not replacing human work, but making a specific, well-defined task fast enough that it’s viable as a weekly process. Without the LLM, writing three new recipes every week would require a food writer. With it, Maya can review and approve the output in fifteen minutes.
What Event Storming couldn’t see
This is worth pausing on. The team tried Event Storming first, and it wasn’t the wrong instinct – just the wrong tool for this particular question.
Event Storming maps what happens inside the business: domain events, commands, actors, process flows. It’s superb at building shared understanding of how the system works. But it can’t see outside the system. It can’t see the customer standing in their kitchen at 5:47pm, staring at a kohlrabi, wondering what the hell to do with it. That moment – that frustration, that decision to cancel – exists in the customer’s life, not in the domain model.
JTBD looks outward, at the customer’s world. What are they trying to accomplish? What are they struggling with? What would make them hire your product and keep hiring it?
The two techniques complement each other. JTBD tells you what problem to solve (eliminate dinner stress). Event Storming tells you how the business processes need to work to solve it (recipe content management, box-contents-to-recipe matching, card printing workflow). You need both.
There’s a useful analogy here. JTBD is a compass – it tells you which direction to walk. Event Storming is a map – it shows you the terrain you’ll cross. Impact Mapping is the route plan – it tells you which paths to take. Example Mapping is the step-by-step guide – it tells you exactly where to put your feet. You need all of them, but you need them in the right order. A compass without a map leaves you walking into swamps. A map without a compass means you’re charting terrain that doesn’t lead anywhere useful.
The GreenBox team had been using the map without the compass. They knew the domain inside out. They could Event Storm any process in the business. But they were mapping terrain that didn’t matter – improving substitution algorithms and farm portals while subscribers were quietly cancelling because the box didn’t do the job they hired it for.
When to use Jobs to Be Done
When churn is high and you don’t know why. Exit surveys give you surface reasons (“too expensive,” “didn’t use it enough”). JTBD interviews give you the real reason – the job wasn’t being done.
When you’re about to invest in a new feature. Before building, ask: does this feature serve the job customers are hiring us for? If you can’t draw a straight line from the feature to the job, you might be building the wrong thing.
When acquisition is hard and you don’t know your message. If you understand the job, your marketing writes itself. “Fresh local vegetables” is a product description. “Stop stressing about Tuesday dinner” is a job statement. One of those converts better than the other.
When you’re entering a new market or customer segment. Different segments may hire the same product for different jobs. A single person living alone might hire GreenBox to reduce food waste (they can’t buy small quantities at the supermarket). A busy family hires it to eliminate dinner decisions. Same box, different job, different marketing, different retention drivers.
When not to use it
When you already know the job and need to execute. JTBD is a discovery technique. If the team already has a clear, validated understanding of why customers buy, running JTBD interviews is discovery theatre. Move on to building.
When the problem is operational, not motivational. If subscribers are leaving because deliveries arrive late or boxes are damaged, you don’t need JTBD – you need to fix your logistics. JTBD is for understanding why customers hire and fire your product. It’s not for fixing broken processes.
When you have fewer than ten customers. JTBD interviews work because patterns emerge across multiple conversations. With a very small customer base, you’ll see individual quirks rather than patterns. Talk to your customers, absolutely – but don’t expect the framework to produce statistical confidence with a sample of three.
As a one-off exercise. The job customers hire you for can shift over time. GreenBox’s early subscribers might have hired it for novelty and local-food values. The next cohort might hire it purely for convenience. Revisit the job periodically, especially after growth spurts or market shifts.
When you’re using it as a substitute for quantitative data. JTBD interviews are qualitative. Fifteen conversations give you themes and hypotheses, not statistical certainty. Maya’s finding that subscribers value convenience over local sourcing is a strong signal, but it’s based on fifteen people. Before restructuring the entire business around it, the team needs to validate with the broader subscriber base. JTBD points you in a direction. Quantitative research tells you how far to walk.
The team tries the wrong tool, then finds the right one
This is a pattern that’s going to repeat throughout this series. The GreenBox team now has a toolkit from the earlier series – Event Storming, Example Mapping, BDD, sprints, Impact Mapping, User Story Mapping. Those tools are powerful. But they were designed for a specific set of problems: understanding the domain, making stories concrete, connecting work to goals.
The new challenge – scaling from 200 to 1,000 subscribers – involves different problems. Customer motivation. Business model viability. Pricing. Market positioning. The old toolkit doesn’t cover these. It’s not that the tools are broken – it’s that the team is reaching for a hammer when they need a screwdriver.
Lee recognises this because he’s seen it before. Teams that learn one set of techniques well tend to apply them everywhere, even where they don’t fit. He calls it “golden hammer syndrome” – when you’ve just learned to use a hammer, everything looks like a nail. Event Storming is a fantastic hammer. But churn analysis isn’t a nail.
The first sign of mastery isn’t knowing how to use a tool – it’s knowing when not to use it. The GreenBox team is learning this the hard way. They reached for Event Storming because it worked before, not because it fit the problem. That’s a natural mistake, and one they won’t make again.
The new series – “Finding the Fit” – is about the tools you need when the product works but the business needs to scale. Different problems, different techniques. JTBD is the first. The others will follow.
The JTBD interviews took three days and cost almost nothing. The insight they produced – that the job is dinner-stress elimination, not fresh-produce delivery – is worth more than any feature the team could have built in those three days. It reframes the entire product strategy.
Two weeks after the recipe cards ship, Maya checks the numbers. Churn has dropped from 8% to 5.5%. Three of the five churned subscribers they interviewed have re-subscribed after Sam emailed them about the recipe cards. Patrick – the man with the kohlrabi guilt – signed back up the same day he got Sam’s email. Greg did not. Louise did not. Maya checked.
She also checked Freshly’s website again. They’ve added a Perth delivery zone. The launch date is next month. Twelve million dollars in funding, a slick app, and $18 per week. Maya’s boxes cost $25 and come with a recipe card printed on a $0.12 piece of cardboard.
She tells herself the recipe cards are working. The churn is dropping. The direction is right.
She doesn’t tell anyone that she spent twenty minutes on Freshly’s sign-up flow that evening, getting as far as the payment page, just to see what the experience felt like. It was smooth. It was fast. It was everything GreenBox’s sign-up flow isn’t. She closed the tab and went for a run on the coastal track, even though it was dark and Nadia told her the path wasn’t lit.
The path was fine. The run helped. The knot in her chest loosened by half a turn.
It’s early. The sample is small. But the direction is clear. When you understand the job, even simple changes can have outsized effects. Recipe cards aren’t sophisticated technology. They’re pieces of cardboard with words on them. But they complete the job the customer hired GreenBox to do, and that matters more than any feature the team could have engineered.
Next week, the team takes that insight and asks an uncomfortable question: what else do we believe about this business that we haven’t actually validated?
That’s Assumption Mapping (coming 23 June).