{"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":"2026-06-26T10:28:31.462Z"},"content":[{"type":"documentation","id":"2d0c7576-8708-4858-85ac-86c157560306","slug":"riskiest-assumption-test-guide","title":"The Riskiest Assumption Test (RAT): Validate the One Thing That Can Kill Your Product","url":"https://www.koji.so/docs/riskiest-assumption-test-guide","summary":"A practical guide to the Riskiest Assumption Test (RAT): the lean method of isolating and validating the single highest-risk, lowest-evidence assumption behind a product before building — including assumption mapping, writing testable assumptions, choosing experiments, and running the desirability test with AI interviews in days.","content":"# The Riskiest Assumption Test (RAT): Validate the One Thing That Can Kill Your Product\n\n**Bottom line:** The Riskiest Assumption Test (RAT) is a lean validation method that isolates the single assumption most likely to sink your product — the one that is both highly important and least supported by evidence — and tests it first, before you write a line of code. Where a Minimum Viable Product (MVP) asks \"what is the smallest thing we can build?\", a RAT asks the sharper question: \"what is the cheapest experiment that tells us whether we should build anything at all?\" Run well, a RAT compresses months of speculative engineering into days of focused learning.\n\nThis guide covers what a RAT is, how it differs from an MVP, how to surface and prioritize your riskiest assumption, how to design the right experiment, and how AI-moderated interviews let you run the customer-facing version of a RAT in days.\n\n## What Is the Riskiest Assumption Test?\n\nEvery product idea is a stack of beliefs. \"Customers have this problem.\" \"They will switch from their current solution.\" \"They will pay this price.\" \"They can figure out the workflow without training.\" Most of these beliefs are untested — they feel obvious from inside the building, but you have no real evidence for them.\n\nA Riskiest Assumption Test focuses your validation energy on the *one* belief that matters most. The logic is simple: if an assumption is critical to success **and** you have little evidence it is true, that is the assumption that should keep you up at night. Testing anything else first is wasted motion, because if the riskiest assumption is wrong, nothing else you build matters.\n\nThe cost of skipping this step is well documented. In CB Insights analysis of startup post-mortems, roughly **35% of failed startups died because there was no market need** for what they built — they shipped a polished answer to a question nobody was asking. The RAT exists to catch that failure mode while it is still cheap to fix.\n\n> \"There are no facts inside the building, so get the hell outside.\" — Steve Blank, originator of Customer Development and a foundational voice of the Lean Startup movement.\n\n## RAT vs. MVP: Why Order of Operations Matters\n\nThe Minimum Viable Product is one of the most misused ideas in product. Teams hear \"MVP\" and build a small, real, shippable product — which still takes weeks or months of engineering. If the core assumption was wrong, the team has simply found an expensive way to fail.\n\nThe RAT reorders the work:\n\n- **MVP mindset:** *Build* the smallest version of the solution, ship it, and see if people use it.\n- **RAT mindset:** *Test* the riskiest belief behind the solution with the cheapest possible experiment — often no product at all — and only build once that belief survives.\n\nA RAT frequently needs zero code: a set of customer interviews, a fake-door landing page, a concierge walkthrough, or a smoke test. The MVP comes *after* the riskiest assumptions have been retired, not before. As product leader Marty Cagan of the Silicon Valley Product Group puts it, the vast majority of product ideas simply do not work out — so the goal of discovery is to find that out quickly and cheaply, not slowly and expensively.\n\n## The Three Risk Categories\n\nDavid J. Bland, co-author of *Testing Business Ideas* with Strategyzer, groups assumptions into three classic categories. Naming the category helps you choose the right test:\n\n- **Desirability** — Do customers actually want this? Is the problem real, painful, and frequent? Will they switch? (Tested primarily through customer conversations and demand experiments.)\n- **Viability** — Does this work as a business? Will customers pay enough, often enough, for the unit economics to make sense?\n- **Feasibility** — Can we actually build and deliver this with our technology, skills, and constraints?\n\nFor most early-stage products, the riskiest assumption lives in **desirability** — and desirability is exactly what customer interviews are built to test. You cannot A/B test your way to knowing whether a problem matters to people; you have to talk to them.\n\n## How to Find Your Riskiest Assumption\n\nThe standard tool is **assumption mapping** — a 2x2 grid that plots every assumption on two axes:\n\n- **Importance** (vertical): If this is wrong, does the whole idea collapse, or is it a minor detail?\n- **Evidence / certainty** (horizontal): How much real evidence do we have that this is true?\n\nThe riskiest assumptions sit in the **high-importance, low-evidence** quadrant — critical to success, and currently unproven. That quadrant is your RAT queue. Everything else either gets deferred (low importance) or noted as a known strength (high evidence).\n\nRun assumption mapping as a 60–90 minute team workshop. Get product, design, and engineering in the room, brain-dump every belief onto sticky notes, then place them on the grid together. The disagreements are the gold: when one person is certain and another is nervous, you have found an assumption worth testing.\n\n## How to Write a Testable Assumption\n\nA vague assumption (\"users will like this\") cannot be tested. Rewrite each one as a falsifiable statement with explicit success criteria:\n\n> \"We believe that **[specific customer segment]** experiences **[specific problem]** frequently enough that they have **[tried to solve it / paid for a workaround]**. We will know this is true if **at least 7 of 10 interviewees** describe the problem unprompted and can point to a recent instance.\"\n\nGood testable assumptions name the audience, the belief, the evidence you will accept, and the threshold that counts as a pass or fail *before* you run the test. Setting the threshold in advance is what stops confirmation bias from creeping in afterward.\n\n## Choosing the Right RAT Experiment\n\nMatch the experiment to the assumption and the budget:\n\n| Assumption type | Cheapest valid test |\n| --- | --- |\n| Problem exists / is painful | Customer discovery interviews |\n| Customers will switch | Switch / Jobs-to-be-Done interviews |\n| There is demand | Fake-door or smoke-test landing page |\n| The solution actually helps | Concierge MVP or Wizard of Oz |\n| Willingness to pay | Pricing interviews + a real \"buy\" intent signal |\n\nNotice how many of the highest-leverage tests are conversations. Demand experiments tell you *whether* people click; interviews tell you *why* — and the \"why\" is what lets you adjust the idea rather than just kill it.\n\n## The Modern Approach: Running a RAT with AI Interviews\n\nThe historical bottleneck of the RAT is the desirability test. Recruiting, scheduling, moderating, transcribing, and analyzing 10–15 discovery interviews has traditionally taken **2–3 weeks** — long enough that teams skip it and \"just build the MVP.\" That is precisely the trap the RAT is meant to prevent.\n\nAI-native research platforms collapse that timeline. With **Koji**, you launch an AI-moderated interview study, share a link, and the AI interviewer conducts every conversation — asking your core questions, then probing follow-ups in real time when an answer is vague or interesting. Voice or text, available 24/7, in dozens of languages. The same validation that took weeks now takes **3–5 days**.\n\nThree Koji capabilities make it especially well suited to a RAT:\n\n1. **Structured questions for crisp pass/fail signals.** Koji supports six [structured question types](/docs/structured-questions-guide) — `open_ended`, `scale`, `single_choice`, `multiple_choice`, `ranking`, and `yes_no`. For a RAT you mix them deliberately: an `open_ended` question to hear the problem in the customer's own words, a `scale` question to quantify pain severity, and a `yes_no` or `ranking` question to measure switching intent. That gives you both the narrative *and* the threshold metric your success criteria depend on.\n2. **Automatic thematic analysis.** Instead of hand-coding transcripts, Koji aggregates themes across every interview and surfaces how many participants raised each one — so you can see at a glance whether \"7 of 10 described the problem unprompted\" actually happened.\n3. **Real-time reporting and quality scoring.** Each interview is quality-scored, so low-effort responses do not pollute your signal, and the report updates live as responses arrive — you can often call the result before the study is even full.\n\nThe strategic shift is this: when validation is fast and cheap enough, there is no longer an excuse to skip the riskiest assumption test and build on faith.\n\n## Common Mistakes to Avoid\n\n- **Testing the comfortable assumption instead of the risky one.** Teams gravitate to assumptions they are confident about because the test will pass. Discipline means pointing the RAT at the belief you are most afraid to examine.\n- **No pre-defined success threshold.** If you decide what counts as a pass after seeing the data, you will rationalize almost any result.\n- **Leading the witness.** Asking \"Wouldn't a faster version be useful?\" guarantees a yes. Neutral, open questions — and an AI moderator that does not have a stake in the answer — reduce this bias.\n- **Treating one RAT as the finish line.** Retiring the riskiest assumption just promotes the next-riskiest one. Validation is a sequence, not a single gate.\n\n## Related Resources\n\n- [Assumption Testing: Validate Product Assumptions Before You Build](/docs/assumption-testing-guide)\n- [Structured Questions Guide: The 6 Question Types](/docs/structured-questions-guide)\n- [Pretotyping: Test Demand Before You Build](/docs/pretotyping)\n- [MVP Validation Guide](/docs/mvp-validation-guide)\n- [Problem Validation Guide](/docs/problem-validation-guide)\n- [Customer Discovery Interviews](/docs/customer-discovery-interviews)\n- [Fake Door Testing Guide](/docs/fake-door-testing-guide)","category":"frameworks","lastModified":"2026-06-24T07:48:31.995981+00:00","metaTitle":"The Riskiest Assumption Test (RAT): A Practical Guide | Koji","metaDescription":"Learn the Riskiest Assumption Test (RAT): how to find your most dangerous unproven belief, design a cheap experiment, and validate it with AI interviews in days instead of months.","keywords":["riskiest assumption test","RAT product","riskiest assumption","assumption mapping","leap of faith assumption","product validation","RAT vs MVP"],"aiSummary":"A practical guide to the Riskiest Assumption Test (RAT): the lean method of isolating and validating the single highest-risk, lowest-evidence assumption behind a product before building — including assumption mapping, writing testable assumptions, choosing experiments, and running the desirability test with AI interviews in days.","aiPrerequisites":["Basic familiarity with product development","Some exposure to lean startup or product discovery concepts"],"aiLearningOutcomes":["Define what the Riskiest Assumption Test is and how it differs from an MVP","Use assumption mapping to find your highest-risk, lowest-evidence belief","Write falsifiable assumptions with pre-set success criteria","Match the right cheap experiment to each assumption type","Run the desirability RAT with AI-moderated interviews in days"],"aiDifficulty":"intermediate","aiEstimatedTime":"15 minutes"}],"pagination":{"total":1,"returned":1,"offset":0}}