{"site":{"name":"Koji","description":"AI-native customer research platform that helps teams conduct, analyze, and synthesize customer interviews at scale.","url":"https://www.koji.so","contentTypes":["blog","documentation"],"lastUpdated":"2026-11-18T09:00:21.455Z"},"content":[{"type":"blog","id":"d2f2333b-e2e5-4cc3-a5e3-f37cde7e68f9","slug":"should-you-build-the-feature-customers-asked-for","title":"Should You Build the Feature Customers Asked For? Test a Mockup First","url":"https://www.koji.so/blog/should-you-build-the-feature-customers-asked-for","summary":"A playbook for early-stage founders deciding whether to build a feature customers requested. Rewrite the request as a problem and set kill criteria, interview 12 to 15 people across three groups (requesters, active customers who did not ask, recent signups), ask about the problem first, show a rough mockup for five seconds then openly, and end with a commitment ask such as a pilot or upgrade. Build when the problem appears beyond the requesters, the mockup is understood and people outside the requesters commit; change, shrink or drop it otherwise. Cites Pendo feature adoption data (80% of features rarely or never used), Atlassian State of Product 2026, Hennink and Kaiser on saturation (9 to 17 interviews), Morwitz et al. on purchase intentions, Schmidt and Bijmolt on 21% hypothetical bias and Camuffo et al. on scientific decision-making in startups.","content":"Should you build the feature customers asked for? Not yet. First find out how many of your users have the problem behind the request, show 12 to 15 of them a mockup of your fix, and ask for something that costs them a little, like a date in their calendar or an upgrade. This playbook walks an early-stage founder through validating a feature request in about a week, with a question bank, a worked example and the rules for deciding to build, change or drop it.\n\n## When you need this\n\nYou have a product, some paying customers and a backlog that grows faster than you can ship. A customer on a call says, \"I'd pay for this if it just had X.\" Two more say something close. It feels like a signal, and with fewer than a hundred users every voice is loud.\n\nThat's the moment to slow down for a week. The signs you need this playbook:\n\n- A request came from one or two vocal customers and you're about to give it a sprint.\n- People phrase the request as a solution (\"add a client portal\") and you aren't sure what problem it solves.\n- The request would take more than two weeks of engineering, or would change what your product is.\n- You hear \"I would pay if...\" more often than you see people paying.\n\nThe cost of guessing is high and it's mostly invisible. Pendo's analysis of 615 software products found that 80% of features were rarely or never used, and that about 12% of features produced 80% of daily usage [1]. Its 2024 benchmarks put it more starkly: for every 100 features shipped, 6.4 drive 80% of clicks, and even the best products only reach 15.6 [2]. At a seed-stage company, one unused feature can be a month of your only engineer.\n\n![Stat card: 80% of features in the average software product are rarely or never used, from Pendo's analysis of 615 products.](/visuals/features-rarely-or-never-used-pendo-stat/features-rarely-or-never-used-pendo-stat.png \"Most of what gets built goes unused, so test the request before you spend a sprint on it · Source: Pendo, Feature Adoption Report, 2019\")\n\nYou're not alone in finding this hard. In Atlassian's State of Product 2026 survey of more than 1,000 product professionals, close to 40% named prioritizing features as their biggest challenge, and 84% said they worry the product they're building won't succeed in the market [3].\n\n## The method in brief\n\nFeature requests are solutions customers have designed for problems they really have. Your job is to separate the two, then test your own fix for the problem. Three methods do most of the work at this stage:\n\n1. **Problem interviews** about the last time the problem came up. These tell you whether the pain is real, frequent and shared. The [Mom Test](/docs/mom-test-methodology) is the rulebook for asking without leading.\n2. **A mockup shown in the same interview**, after the problem questions. A screenshot, a Figma frame or even a sketch is enough. A timed first look tells you whether the idea is clear on sight; the open questions after it tell you whether it fits how they work. Koji's [prototype testing guide](/docs/prototype-testing-concept-validation) covers the craft.\n3. **A commitment ask** at the end: an upgrade, a pilot date, a deposit. Talk is cheap; the [stated vs revealed preferences](/docs/stated-vs-revealed-preferences) doc explains why.\n\nIf you have the traffic, a [fake door test](/docs/fake-door-testing-guide) (a button for a feature that doesn't exist yet) measures clicks from your whole user base, and a [concierge MVP](/docs/concierge-mvp-guide) (doing the feature's job by hand) tells you whether people use the outcome. For a founder with 30 to 300 users, interviews with a mockup are usually the fastest honest answer, and you can run a fake door alongside them. For the full menu of methods, see our [feature validation guide](/docs/feature-validation-guide) and the [methods compared for product managers](/blog/feature-validation-methods-compared-product-managers).\n\nWhy not just count requests? Because requests travel. By the time a request reaches you through sales, support and a Slack thread, it tends to be bigger and more certain than the problem behind it, a pattern we cover in [why the feature request that reaches you is bigger than the problem](/docs/feedback-relay-amplification-bullwhip). 37signals made the opposite point years ago: they read requests and let them go, because \"the ones that are really important will keep bubbling up\" [4]. Both lessons point the same way. Count the problem, not the request.\n\n## Step by step: validate a feature request in a week\n\n**Day 1: write the request as a problem and a bet.** Turn \"add a client portal\" into \"freelancers lose time chasing clients for approvals, and a shared page would cut that.\" Then write down your kill criteria before you talk to anyone: what would you need to hear to build it, and what would make you drop it?\n\n**Day 1: pick who to talk to.** You want three groups, about five people each:\n\n- The customers who asked for it.\n- Active customers who fit your best customer profile and did not ask.\n- Recent signups or trial users in the segment the feature is meant to win.\n\nThe second group matters most. If only the requesters have the problem, you've found a niche, not a feature. Leave out friends, investors and other founders; they're kind, and kindness is noise here.\n\n**Day 2: invite them.** Email them yourself from your own address, with a short, honest ask: \"We're deciding what to build next and your answer will shape it. It takes 10 minutes, by voice or text, whenever suits you.\" Invite about two to three times as many people as you need. Active customers reply far more often than strangers, but plan for silence anyway.\n\n**Day 2: make the mockup.** One screen is enough. It should show the outcome (the approval page a client sees), not your settings menu. Don't polish it. A rough mockup invites honest criticism, a glossy one invites compliments.\n\n**Days 3 to 6: run 12 to 15 conversations.** Why that many? A systematic review of empirical saturation studies found that interview studies with a fairly uniform group and a narrow question reached saturation within 9 to 17 interviews [5]. A feature decision is a narrow question. Fifteen spread across your three groups gives you enough to see whether the problem is shared, and few enough to finish in a week. If the groups disagree sharply, add five more to the group that surprised you. Our [how many user interviews](/docs/how-many-user-interviews) doc goes deeper.\n\nEach conversation runs in this order: problem first, then the mockup, then the ask. Never show the mockup first. Once they've seen your solution, every answer about the problem is coloured by it.\n\n**Day 7: read the results and decide** (rules below).\n\n## What to ask: a question bank for feature requests\n\nThese are open questions, in the order you'd ask them. Ask about the last real time, not about what they'd do in general.\n\n![Question bank for validating a feature request: problem questions, a first look at the mockup, fit questions and a commitment ask.](/visuals/feature-request-validation-interview-questions/feature-request-validation-interview-questions.png \"Problem first, mockup second, the ask last. Reuse these in your next round\")\n\n**The problem**\n\n- Tell me about the last time [the problem] came up. What were you doing?\n- What did you do about it? Walk me through it.\n- How often does that happen in a normal month?\n- What does it cost you when it happens: time, money, a client?\n- Have you tried anything else to fix it? What happened?\n\n**First look at the mockup** (shown for 5 seconds, then hidden)\n\n- What do you remember seeing?\n- What do you think this does?\n\n**Fit**\n\n- Where would this fit in the way you work today? What would it replace?\n- What's missing, or what would stop you from using it?\n\n**The ask**\n\n- If this were ready on [date], would you move up to the plan that includes it?\n- Would you try an early version with us for two weeks and give us 20 minutes of feedback after?\n\nThe first look is the interesting part. If people can't say what the screen does after five seconds, your fix may solve the right problem but won't be understood on sight, and that's cheaper to learn from a mockup than from a launch.\n\n## A worked example\n\nSay you're a two-person team selling invoicing software to freelance designers. You have 140 paying customers. Over a month, three of them ask for a client portal where their clients can approve invoices and see project files. Your developer estimates five weeks.\n\n![Two founders at a laptop and a whiteboard, sketching a feature mockup before deciding whether to build it](https://images.unsplash.com/photo-1621570168077-befe8b7eab56?crop=entropy&cs=tinysrgb&fit=max&fm=jpg&ixid=M3w5NDUzMDN8MHwxfHNlYXJjaHw2fHx0d28lMjBwZW9wbGUlMjBsYXB0b3AlMjB3aGl0ZWJvYXJkJTIwc2tldGNofGVufDF8MHx8fDE3OTQ5OTE4MTB8MA&ixlib=rb-4.1.0&q=80&w=1080)\n\n_Photo by [Surface](https://unsplash.com/@surface?utm_source=koji&utm_medium=referral) on [Unsplash](https://unsplash.com/?utm_source=koji&utm_medium=referral)_\n\nYou write the bet: \"Freelancers lose days waiting for client approval, and a portal would cut that.\" Your kill rule: build only if at least half the non-requesters describe a recent approval delay that cost them money, and at least four people agree to a pilot.\n\nYou invite the three requesters, ten active customers who never asked and ten recent signups. Fifteen people answer within five days.\n\nHere's what you might hear. All three requesters describe chasing approvals. Of the seven active customers, two do; the other five say their clients approve by replying to the invoice email and it's fine. Of the five signups, three mention approvals, but all three are designers working with agencies, where an account manager has to sign off.\n\nOn the mockup, most people remember \"a page for the client,\" but several think it's a file-sharing tool and ask whether it replaces their Google Drive folder. That confusion matters: the screen led with files, and the problem was approval.\n\nAt the ask, two requesters and two agency freelancers agree to a pilot. Nobody outside that group does.\n\nThe decision: you don't build the portal. You build a one-week version of the part that matters, an \"Approve\" button on the invoice email, and offer the pilot to the four who said yes. You also note that freelancers who work with agencies look like a segment worth a closer look, and you queue an [ICP round](/blog/find-your-first-customer-segment-before-you-build) for them.\n\n## Reading the results and deciding\n\nCount before you interpret. For each person, write down three things: did they describe the problem from a real, recent event; did they understand the mockup in five seconds; and did they accept the ask. Then work through the decision.\n\n![Flowchart for deciding whether to build a requested feature, from the problem interviews to the mockup test and a commitment ask.](/visuals/should-you-build-the-feature-flowchart/should-you-build-the-feature-flowchart.png \"Work through it with your co-founder before anyone opens the code editor\")\n\nRules of thumb that hold up well for a founder with 12 to 15 conversations:\n\n- **Build** when the problem shows up beyond the people who asked (a good bar is at least half of the non-requesters, from real recent events), most people understood the mockup on sight, and at least a few people outside the requesters accepted the ask.\n- **Change the fix** when the problem is widespread but the mockup confused people or solved the wrong part. Redesign, then show the new version to five more people.\n- **Build small** when the problem is real but narrow. Do the smallest version that solves it for that group, or do it by hand for them, and watch usage before you expand it.\n- **Drop it** when only the requesters have the problem, or when people describe it in the hypothetical (\"I guess that could be useful\"). Tell the requesters what you learned and why. They'll respect it more than a feature that never arrives.\n\nWhatever you decide, the commitments are your strongest evidence. A founder who builds on interest alone is betting against the research below.\n\n## Run it this week with Koji\n\nYou can run all of this by hand, with a calendar link, a Figma file on screen share and notes in a doc, and many founders should. Koji makes it faster when your customers won't book calls, because they answer an AI-moderated interview by voice or text whenever suits them, and you read 15 conversations in an evening instead of spending a week on calls. The setup:\n\n1. **Start a study from one sentence.** On the home screen, type your bet (\"Find out whether freelancers lose time waiting for client approvals, and test a mockup of an approval page\") and Koji drafts a research brief with questions you can edit.\n\n![Koji home screen asking what you want to research, with one research question typed in and a Start study button.](/email/shots/shot-brief-from-one-message.jpg \"Type the bet behind the request and Koji drafts the interview brief\")\n\n2. **Add the problem questions** from the bank above as open questions. Koji asks follow-ups on each answer, so \"it happens sometimes\" turns into the story of the last time.\n3. **Attach your mockup to a question.** Any question can carry one image (PNG, JPG or WebP). Describe what the image shows in the field under it, so Koji's follow-ups are about the screen. Set it to \"Visible for 5 seconds\" for the first-look question: the participant sees it once, full screen, and then Koji asks what they remember. Add the same image with no time limit to the fit questions that follow. This works in text and voice interviews.\n4. **Add the ask as a choice question,** such as \"Would you try an early version for two weeks?\" with Yes, Maybe and No. If the feature would be a paid upgrade, you can add a price test question to see how many would still say yes at each price, with their reasons. It reads best from about 30 interviews, so treat a 15-person result as directional.\n5. **Invite your three groups.** Share the interview link in your own email, or add people to Koji's Participants list and send the invitation from there, with reminders. A screening question can make sure only active customers get through.\n6. **Read the report.** Koji groups what people said into themes with the quotes behind each one, shows how many chose each answer to the ask, and lets you download the report as editable slides or a PDF for your co-founder or your next investor update.\n\nYou can build the study, attach an image and preview the interview yourself from a free account before you send anything.\n\n## Common mistakes\n\n- **Showing the mockup first.** It anchors every answer that follows. Problem first, always.\n- **Only talking to the requesters.** They already believe in their solution. The customers who didn't ask tell you whether it's a feature or a favour.\n- **Counting votes as demand.** Ten upvotes from one account, or from free users who never activated, are one data point.\n- **Asking \"would you use this?\"** Almost everyone says yes. Ask what they did last time, and ask for a commitment.\n- **Deciding the rules after the results.** Write the kill criteria on Day 1.\n\n## When research will mislead you\n\nInterviews are good at telling you whether a problem is real and whether a fix makes sense. They are poor at predicting what people will do. Morwitz, Steckel and Gupta's analysis of purchase intention studies found that stated intentions track real purchases more closely for existing products than for new ones [6]. A feature nobody has used yet is the hard case.\n\nMoney makes it worse. A meta-analysis of 77 studies found that people's hypothetical willingness to pay runs on average 21% above what they actually pay [7]. \"I would pay for it\" is a soft yes.\n\nSo pair the interviews with behaviour. A pilot people actually show up to, an upgrade they actually click, a fake door your users actually press, or usage of a small version after two weeks. There's also good evidence that this discipline pays: across four randomised trials with 759 startups, founders trained to treat their ideas as hypotheses and test them dropped weak ideas more often, and made a few deliberate pivots rather than none or constant ones [8].\n\n## Where this fits\n\nOn the [founder's research playbook](/learn/founders#early-stage), this is the first decision when you're finding product-market fit: does the concept land? Before it, [interview your existing customers without booking thirty calls](/blog/interview-existing-customers-without-thirty-calls) to know who your best users are. After it, if the feature changes what you sell, [price it with 30 customer conversations](/blog/price-your-first-product-with-30-conversations). If your [PMF score is below 40%](/blog/pmf-score-below-40-percent), fix that first: a new feature rarely rescues fit. And if you're pre-product, [test your landing page](/blog/test-your-landing-page-before-you-launch) or [pre-sell the product](/blog/how-to-pre-sell-before-you-build) instead. See how Koji works for founders at [/for/founders](/for/founders).\n\n## FAQ\n\n**How many customers need to ask before I build a feature?**\nThere's no safe number, because requests are about solutions. Count how many users have the problem behind the request. If at least half of a sample of active customers who never asked describe it from a real, recent event, it's worth testing a fix.\n\n**Should I build a feature a big customer is asking for?**\nOnly if the problem shows up beyond that customer, or if they'll commit to it (a paid pilot, a contract term). Otherwise you're building custom software for one account, which may be fine, but price it that way.\n\n**How do I say no to a feature request without losing the customer?**\nTell them what you learned, why you're not building it now and what you're doing about the problem instead. Customers who see their feedback taken seriously tend to keep giving it.\n\n**Can I validate a feature without interviews?**\nWith enough traffic, a fake door test measures interest across your whole user base. It tells you how many clicked, not why, so pair it with five or six conversations.\n\n**What should the mockup look like?**\nOne rough screen that shows the outcome the customer gets. A sketch or a quick Figma frame is enough. Polish invites compliments.\n\n**One next step:** write the request you're most tempted to build as a problem and a bet, then [start a study from a Koji template](/for/founders) and send it to five customers who didn't ask for it.\n\n## References\n\n1. Pendo. \"The 2019 Feature Adoption Report.\" Pendo.io, 2019. https://www.pendo.io/resources/the-2019-feature-adoption-report/\n2. Pendo. \"Product Benchmarks.\" Pendo.io, 2024 (data from May 2023 to March 2024). https://www.pendo.io/product-benchmarks/\n3. Atlassian. \"State of Product 2026.\" Atlassian, September 2025; figures as reported by diginomica, \"Product teams under pressure,\" 2025. https://www.atlassian.com/software/jira/product-discovery/resources/state-of-product-2026 and https://diginomica.com/product-teams-under-pressure-atlassian-state-product-report-role-in-flux\n4. 37signals. \"Forget Feature Requests.\" Getting Real, Basecamp, 2006. https://basecamp.com/gettingreal/05.7-forget-feature-requests\n5. Monique Hennink and Bonnie N. Kaiser. \"Sample sizes for saturation in qualitative research: A systematic review of empirical tests.\" Social Science & Medicine 292, 2022. https://pubmed.ncbi.nlm.nih.gov/34785096/\n6. Vicki G. Morwitz, Joel H. Steckel and Alok Gupta. \"When do purchase intentions predict sales?\" International Journal of Forecasting 23(3), 2007. https://ideas.repec.org/a/eee/intfor/v23y2007i3p347-364.html\n7. Jonas Schmidt and Tammo H. A. Bijmolt. \"Accurately measuring willingness to pay for consumer goods: a meta-analysis of the hypothetical bias.\" Journal of the Academy of Marketing Science 48(3), 2020. https://core.ac.uk/download/519031117.pdf\n8. Arnaldo Camuffo, Alfonso Gambardella, Danilo Messinese, Elena Novelli, Emilio Paolucci and Chiara Spina. \"A scientific approach to entrepreneurial decision-making: Large-scale replication and extension.\" Strategic Management Journal 45(6), 2024. https://econpapers.repec.org/RePEc:bla:stratm:v:45:y:2024:i:6:p:1209-1237\n\n## Further reading\n\n- Rob Fitzpatrick, *The Mom Test* (2013). The best short book on asking customers about their lives instead of your idea, with practical advice on digging behind feature requests. Our summary: [Mom Test methodology](/docs/mom-test-methodology).\n- 37signals, [Getting Real](https://basecamp.com/gettingreal) (2006). Old, short and still the clearest argument for building less.\n- Koji, [Feature validation: how to validate a feature before you build it](/docs/feature-validation-guide). The method reference behind this playbook, with fake doors, prototypes and price tests side by side.\n","category":"Founders","lastModified":"2026-11-18T08:50:10.797339+00:00","metaTitle":"Should You Build the Feature Customers Asked For?","metaDescription":"How to validate a feature request before you build it: who to interview, what to ask, a 5-second mockup test and rules to build, change or drop it.","keywords":["validate feature request","should I build this feature","feature validation","mockup testing","customer interviews","startup product decisions","early-stage founders"],"aiSummary":"A playbook for early-stage founders deciding whether to build a feature customers requested. Rewrite the request as a problem and set kill criteria, interview 12 to 15 people across three groups (requesters, active customers who did not ask, recent signups), ask about the problem first, show a rough mockup for five seconds then openly, and end with a commitment ask such as a pilot or upgrade. Build when the problem appears beyond the requesters, the mockup is understood and people outside the requesters commit; change, shrink or drop it otherwise. Cites Pendo feature adoption data (80% of features rarely or never used), Atlassian State of Product 2026, Hennink and Kaiser on saturation (9 to 17 interviews), Morwitz et al. on purchase intentions, Schmidt and Bijmolt on 21% hypothetical bias and Camuffo et al. on scientific decision-making in startups.","aiKeywords":["feature request validation","mockup test","five-second test","commitment ask","customer interviews","founder playbook","product-market fit"],"aiContentType":"guide","faqItems":[{"answer":"There is no safe number, because requests are about solutions. Count how many users have the problem behind the request. If at least half of a sample of active customers who never asked describe it from a real, recent event, it is worth testing a fix.","question":"How many customers need to ask before I build a feature?"},{"answer":"Only if the problem shows up beyond that customer, or if they will commit to it with a paid pilot or a contract term. Otherwise you are building custom software for one account, which may be fine, but price it that way.","question":"Should I build a feature a big customer is asking for?"},{"answer":"Tell them what you learned, why you are not building it now and what you are doing about the problem instead. Customers who see their feedback taken seriously tend to keep giving it.","question":"How do I say no to a feature request without losing the customer?"},{"answer":"With enough traffic, a fake door test measures interest across your whole user base. It tells you how many clicked, not why, so pair it with five or six conversations.","question":"Can I validate a feature without interviews?"},{"answer":"One rough screen that shows the outcome the customer gets. A sketch or a quick Figma frame is enough. Polish invites compliments.","question":"What should the mockup look like?"}],"relatedTopics":["feature validation","concept testing","customer interviews","founder playbooks"]}],"pagination":{"total":1,"returned":1,"offset":0}}