ICP Discovery: Finding Your Buyer When No Playbook Exists

Every GTM tool assumes you already know your ideal customer profile. Apollo needs it to filter leads. Clay needs it to enrich contacts. Outreach needs it to personalize sequences. But what if you genuinely do not know who your buyer is yet? This is not a niche problem. CB Insights analyzed 156 failed startups and found that 42% failed because there was no market need, making it the single most common cause of startup death. What that statistic obscures is how many of those companies actually had a viable product but sold it to the wrong people.

The distinction matters. No market need does not always mean the product was useless. It often means the founders never found the right buyer. They built something valuable, pitched it to the wrong audience, heard no over and over, and concluded the market did not exist. The market existed. They were just looking in the wrong place.

Category-creating startups face this problem at a structural level. When you are building something genuinely novel, there is no existing category to study, no competitor customer lists to reverse-engineer, and no analyst reports profiling the typical buyer. You are not choosing between known segments. You are discovering whether segments even exist. The traditional ICP frameworks that ask you to define firmographics, demographics, and technographics assume the answer to the question you are trying to ask.

This is why ICP discovery, not ICP definition, is the critical skill for category creators. The difference is not semantic. Definition is a top-down exercise where you decide who your buyer should be. Discovery is a bottom-up process where you let the data and early signals reveal who your buyer actually is.

Traditional ICP frameworks follow a predictable pattern. Start with firmographic data: industry, company size, revenue band, geography. Layer on demographics: buyer title, seniority, department. Add technographics: what tools they already use, what systems they need to integrate with. Score and rank. Build your outbound list. This works beautifully when your category exists and you are choosing which segment to focus on within an established market.

For category creators, this approach fails for three fundamental reasons. First, you cannot filter on adoption signals that do not exist yet. If no one has ever bought a product like yours, there is no purchase history to analyze, no intent data to mine, and no technology install base that predicts readiness. Second, the buyer title is often wrong. When Slack launched, the assumed buyer was the engineering team lead. The actual buyer turned out to be entire teams, crossing every department and function. The buying unit was not a person but a social graph. Third, the problem you solve may not map to an existing budget category. When Gong started selling conversation intelligence, there was no line item in any sales department budget for call recording and AI analysis. The budget had to be created, not redirected.

The Jobs to Be Done framework, pioneered by Clayton Christensen, offers a better foundation for category creators than traditional ICP methods. JTBD asks not who is your customer but what job is the customer hiring your product to do. This reframe is powerful because the same job can be done by buyers across wildly different firmographic profiles. A 50-person startup and a 5,000-person enterprise might both be hiring your product to do the same job, just at different scales and with different willingness to pay.

The practical implication: start with the job, not the demographic. Your ICP will emerge from understanding who has the job to be done most urgently, most frequently, and with the least satisfactory alternatives available today.

Instead of defining your ICP from a conference room whiteboard, discover it from the ground up using the signals that are already in front of you. This is what we call bottom-up ICP discovery, and it inverts the traditional process entirely. Rather than starting with a hypothesis about who should buy and then seeking confirmation, you start with observed behavior and work backward to the profile.

The signals you need are hiding in plain sight. Who responded to your cold outreach with genuine curiosity, not just politeness? Who signed up for a demo without being chased? Who, during a demo, leaned forward and started describing their workflow in detail? Who converted from free trial to paid without a sales touch? Who referred you to a colleague unprompted? Each of these micro-signals carries more ICP information than a thousand rows in a firmographic database.

The method works in three phases. Phase one is signal collection, where you capture every interaction from first touch through conversion and record not just the outcome but the behavioral context. Phase two is pattern clustering, where you look for non-obvious commonalities among your best signals. Maybe your most engaged prospects are not from the same industry but they all have the same organizational structure, or they all recently went through a specific change event like a funding round or leadership transition. Phase three is hypothesis formation, where you build a testable ICP hypothesis that includes both the traditional dimensions and the behavioral and situational triggers you discovered.

Arnen's Buyer Persona Synthesizer automates this process by clustering your early signals and pattern-matching them against analogous markets. It surfaces the non-obvious commonalities that human analysis misses. The output is not a static persona document. It is a living hypothesis with confidence scores that updates as you close or lose deals.

Rahul Vohra, the founder of Superhuman, developed one of the most rigorous frameworks for connecting ICP discovery to product-market fit. His approach, which he detailed publicly, centers on a single survey question borrowed from Sean Ellis: How would you feel if you could no longer use this product? The key metric is the percentage of users who answer very disappointed.

When Vohra first ran this survey, only 22% of Superhuman users said they would be very disappointed without the product. The conventional wisdom would be to try to make the product appeal to everyone. Vohra did the opposite. He segmented the responses and found that a specific subset of users, founders and managers who received 100+ emails per day and valued speed above all else, had a very disappointed rate far above the average.

Instead of trying to please the users who were indifferent, Vohra doubled down on the users who already loved the product. He narrowed the ICP, focused development on the features that mattered most to that segment, and systematically ignored feedback from users outside the core profile. The result: the very disappointed score climbed from 22% to 58%, well past the 40% threshold that Ellis identified as the benchmark for product-market fit.

The lesson for category creators is counterintuitive: your ICP is not everyone who could use your product. It is the narrow segment that would be devastated without it. Finding that segment requires active discovery, not passive observation. Vohra's framework gives you a repeatable method: survey, segment, identify the lovers, understand why they love it, and then find more people who look like them.

Gong's ICP discovery story is instructive because it reveals how narrow your initial target should be. When Gong started selling in 2016, they did not try to sell to every company with a sales team. They targeted roughly 5,000 specific companies that met a tight set of criteria: B2B, technology sector, with inside sales teams of a certain size, in markets where conversation data would be most immediately valuable. This surgical approach let them saturate a small segment rather than spray across a broad market.

Gusto, the payroll and HR platform, took an even more extreme approach to initial ICP narrowing. Despite building a product that could theoretically serve any small business that pays employees, Gusto started by targeting only companies with roughly 5 employees located in California. This hyper-specific ICP allowed them to perfect their onboarding, understand the regulatory landscape deeply, and build word-of-mouth within a tight network before expanding. The expansion came later, after they had thoroughly understood the needs and behaviors of their initial segment.

Figma's ICP evolution tells a different but equally important story. Figma launched as a design tool, and the obvious ICP was designers. But the product's collaborative, browser-based nature meant that the actual usage pattern was much broader. Product managers, engineers, marketers, and executives all ended up in Figma files, not as designers but as commenters, reviewers, and stakeholders. The ICP expanded beyond designers into what Figma identified as anyone involved in the product development process.

The common thread across these stories is that the initial ICP was far narrower than the eventual market. Gong started with 5,000 companies and now serves tens of thousands. Gusto started with 5-employee California companies and now serves hundreds of thousands of businesses nationwide. Figma started with designers and now serves entire product organizations. The narrow start was not a limitation. It was the strategy.

Bangaly Kaba, who led growth at Instagram and later at Instacart, developed the Adjacent User Theory to explain how products grow beyond their initial user base. The theory states that at any given time, there is a set of users who are almost but not quite able to get value from your product. They are adjacent to your current users. They have similar needs but face slightly different barriers to adoption. Understanding and removing those barriers is how you systematically expand your ICP.

For category creators, the Adjacent User Theory is essential because it provides a structured framework for ICP expansion without losing focus. Your initial ICP is your beachhead. Your adjacent users are your next market. But the key insight is that adjacent users are not simply a broader version of your current users. They often have different onboarding needs, different value perceptions, and different purchasing processes.

Slack's ICP expansion illustrates this perfectly. The initial users were developer teams who adopted Slack because it replaced IRC and integrated with their development tools. The adjacent users were non-technical teams within the same companies who saw developers using Slack and wanted in. But these adjacent users had different needs: they cared less about GitHub integrations and more about file sharing, search, and channel organization. Slack had to evolve the product to serve adjacent users without alienating the core. The key buyers shifted from individual teams to IT departments making company-wide purchasing decisions.

The practical application: once you have validated your initial ICP with 20-30 customers, start mapping your adjacent users. Who is trying to use your product but failing? Who is asking for features that your core users do not need? Who is finding your product through word of mouth rather than your marketing? These signals point to your next ICP segment.

Just as important as knowing your ideal customer is knowing who is explicitly not your customer. The anti-persona concept, popularized by HubSpot and later formalized by product-led growth practitioners, defines the characteristics of buyers who will waste your time, drain your resources, and churn within months.

For category creators, anti-personas are especially important because when you are building something novel, the temptation is to sell to anyone who shows interest. In the early days, every hand raised feels precious. But selling to the wrong customer is worse than not selling at all. Wrong-fit customers generate misleading product feedback, distort your roadmap, consume disproportionate support resources, and lower your NPS. Their churn pulls down your retention metrics, making it harder to raise your next round.

Building an anti-persona requires examining your worst customer experiences. Which customers churned fastest? Which ones generated the most support tickets per dollar of revenue? Which ones asked for features that would have taken the product in a direction you did not want to go? The patterns in these negative signals are just as valuable as the patterns in your best customers.

Common anti-persona characteristics for category-creating B2B startups include: companies that have no budget authority for your category and cannot create a new line item, buyers who are looking for a feature-by-feature replacement of an existing tool rather than a new approach, organizations where the decision-making process requires more than 6 months because your runway cannot absorb those sales cycles, and customers whose primary motivation is price rather than value because they will churn the moment a cheaper alternative appears.

Front CEO Mathilde Collin has said publicly that she wishes they had defined their anti-persona earlier. In the early days, Front tried to serve both individual users and enterprise teams, which pulled the product in conflicting directions. The clarity that came from defining who was not their customer was as valuable as defining who was.

Clayton Christensen's Jobs to Be Done framework provides the deepest theoretical foundation for ICP discovery in novel categories. The core insight is that customers do not buy products. They hire them to do a job. The job exists independent of any product, and understanding the job illuminates who the real buyer is.

For category creators, JTBD is especially powerful because it decouples your ICP from existing product categories. When you define your ICP by traditional firmographics, you are implicitly anchoring to existing market structures. When you define it by the job to be done, you can discover buyers that no firmographic filter would surface.

The practical application is to conduct JTBD interviews with your earliest users and prospects. The interview format asks people to walk you through the last time they encountered the problem your product solves. Not hypothetically, but the actual last time. Where were they? What triggered the need? What did they try first? Why did that fail? What did they try next? How did they ultimately resolve it, or did they give up? This narrative approach surfaces the contextual and emotional dimensions of the buying decision that surveys miss entirely.

The output of JTBD interviews is not a demographic profile. It is a situation profile: a specific set of circumstances that creates urgent demand for what you are building. Your ICP then becomes defined not by who the buyer is in the abstract, but by the situation they find themselves in. This is why you often see category-creating products adopted by buyers who look nothing alike on paper but who share the same urgent situational need.

Rahul Vohra's product-market fit framework, which he open-sourced through his writing, deserves a deeper examination because it provides the most actionable bridge between ICP discovery and product development. The framework has four steps that every category creator should follow.

Step one: survey your users with the Sean Ellis question. How would you feel if you could no longer use this product? Very disappointed, somewhat disappointed, or not disappointed. You need at least 40 responses to get meaningful signal. Step two: segment the responses. Break them down by user type, use case, company size, or any other dimension you can identify. Look for the segment where the very disappointed percentage is highest. This segment is your high-expectation customer, your true ICP.

Step three: analyze the qualitative responses from your high-expectation customers. What do they say is the main benefit of the product? What would they use as an alternative if your product disappeared? How would they describe the product to a friend? The answers to these questions tell you what your positioning should be, not in your words but in the words of the people who love you most.

Step four: build a roadmap that doubles down on what high-expectation customers love while addressing the concerns of users who are somewhat disappointed, but only when those concerns do not conflict with what the core segment values. This is the hardest part because it requires saying no to feedback from users who are outside your ICP. Vohra found that ignoring the feedback from not-disappointed users and focusing development entirely on the very-disappointed and somewhat-disappointed segments was the key to moving from 22% to 58%.

The framework is not a one-time exercise. Run the survey quarterly, track the very-disappointed percentage over time, and watch how it changes as your ICP evolves. A declining score means you are drifting away from your core. A rising score means you are deepening product-market fit within your ICP.

An ICP hypothesis needs validation through at least 15 qualified conversations. The goal is not to confirm what you believe. It is to discover what you missed. The conversations should be structured around three axes: problem validation, which confirms the buyer actually has the problem you solve and it is painful enough to warrant a purchase; solution validation, which confirms your specific approach resonates more than the alternatives they have tried; and buying process validation, which confirms they have the budget, authority, and timeline to actually purchase.

The most common surprise in these conversations is that the buyer is rarely who you assumed. Founders building developer tools discover their real buyer is the VP of Engineering, not the individual developer. Founders building compliance tools discover their champion is the CFO, not the compliance officer. Founders building design tools discover their economic buyer is the Head of Product, not the Head of Design. The user and the buyer are often different people, and your ICP needs to account for both.

After 15 conversations, you should be able to answer five questions with confidence. Who has this problem most acutely? What triggers them to start looking for a solution? Who in the organization champions the purchase? Who controls the budget? And what is the typical timeline from first conversation to signed contract? If you cannot answer all five, you need more conversations, not more hypothesizing.

Once validated, your ICP becomes operational. Export it to your go-to-market tools: Clay for lead enrichment, Apollo for prospecting, and your outbound platform for personalized outreach. Arnen is the discovery layer that feeds your execution stack, turning ambiguous early signals into a sharp, testable buyer profile that your entire team can act on.

The question every category creator eventually faces is when to expand beyond the initial ICP. Expand too early and you dilute focus, confuse your product roadmap, and slow growth. Expand too late and you miss the market window while competitors catch up.

The signals that indicate readiness for ICP expansion include: net revenue retention above 120%, which means your existing segment is healthy enough to sustain itself while you invest in a new one; a saturated addressable market within your current ICP, which means you are running out of net-new accounts to pursue; inbound interest from outside your ICP, which means adjacent users are pulling you into new segments; and a product platform that can support segment-specific customization without forking the codebase.

The Adjacent User Theory provides the expansion methodology. Identify the adjacent segment, understand the barriers they face that your current users do not, remove those barriers through targeted product and go-to-market changes, and measure whether the new segment achieves the same engagement and retention benchmarks as your core. If it does, you have found your next ICP. If it does not, the segment may be adjacent in theory but not in practice.

Figma's expansion from designers to the broader product development organization is the gold standard example. They did not simply start marketing to product managers. They built features like FigJam, a whiteboarding tool, that addressed the specific jobs to be done by non-designer users. The product expanded, and the ICP expanded with it. The ICP did not expand first. This sequencing matters: product changes enable ICP expansion, not the other way around.

ICP discovery is not a one-time project. It is a continuous system that should be embedded in your go-to-market operations. The system has four components that work together.

The first component is signal capture: every customer-facing interaction should feed structured data back into your ICP model. Win/loss analysis after every deal. Churn interviews within two weeks of cancellation. NPS and satisfaction surveys quarterly. Product usage data correlated with customer attributes. The second component is pattern analysis: monthly reviews of who is buying, who is churning, and why. Look for shifts in the profile of your best customers over time. Are they getting larger? Shifting industries? Coming from different channels?

The third component is hypothesis testing: quarterly ICP reviews where you update your target profile based on accumulated evidence. Treat your ICP like a product: version it, document the changes, and communicate updates to every team that touches go-to-market. The fourth component is expansion planning: semi-annual reviews of adjacent segments using the Adjacent User Theory framework. Which new segments are showing organic interest? What product changes would unlock them? What is the expected impact on your unit economics?

Arnen's Buyer Persona Synthesizer is built to serve as the engine of this system. It continuously ingests your signal data, clusters patterns across your customer base, benchmarks your ICP against analogous companies, and surfaces expansion opportunities you might miss through manual analysis alone. The goal is to make ICP discovery as rigorous and data-driven as the product development process itself.

The founders who build this system early, treating ICP discovery as a first-class operational function rather than a one-time strategic exercise, consistently outperform those who define their ICP once and never revisit it. Your market is not static. Your ICP should not be either.