A venture can narrow at every stage and still arrive with its riskiest belief untouched.
Every founder who follows a structured venture process must make choices that narrow their focus. The process itself demands this discipline. Founders do not leave the customer stage with four target segments, test six hypotheses at the product stage, or track eleven metrics equally at the financial stage. At five key points, the structure guides us to focus on what matters most.
The real question is not whether founders narrow their focus, because they do. Instead, we should ask which direction those choices take them.
As we observe founders at the financial stage, this narrowing becomes easiest to track because each decision leaves a clear numerical trail. Across two separate coaching cohorts, we observed several recurring failure patterns. Some teams presented revenue models that no longer matched the evidence from earlier stages. Others submitted cash flow statements that omitted sales- cycle metrics they had already measured. Some dashboards included figures that contradicted the teams’ own inputs. A few funding requests were actually smaller than the capital their models required.
On their own, these look like ordinary modeling errors.
What stood out was their consistent direction. None of the teams produced unreconciled figures that worsened their outlook, such as deepening losses, larger funding needs, or weaker unit economics. Every single error made the numbers look better: revenue arrived sooner, costs appeared lower, and ratios improved. When mistakes consistently slant in a favorable direction, it points to something beyond simple oversight.
These teams worked carefully and reconciled their numbers across most areas. Yet the figures that would have made their outlook less favorable were precisely the ones left out.
This is the pattern we want to highlight. The financial stage simply makes it easiest to spot. The same tendency appears at four earlier points in the process, where it easily disguises itself as disciplined decision-making.
Protective Subtraction
Why the direction of a cut reveals more than the cut itself
Leidy Klotz opens Subtract with a problem that looks too simple to be interesting. People trying to improve a structure add pieces to it. People fixing a schedule add steps. People streamlining a process add another review. Across a series of experiments, his research team found that subtraction rarely enters the running. It isn’t that people weigh removal against addition and pick addition. Often, removal never comes up.
That holds in most settings. It changes when teams work inside a structured venture process, and the reason is worth spelling out.
A structured process removes the option of not subtracting. A team can’t leave the customer stage carrying every segment they find interesting, because the next stage requires a single beachhead. They can’t leave the product stage testing every claim their evidence supports, because the first version tests one behavioral assumption. They can’t leave the financial stage watching every metric they built, because the stage asks for a single core metric. The narrowing isn’t a choice the founder makes or fails to make. It’s a condition of moving forward.
So every founder in our cohorts subtracts. The process guarantees it. What it doesn’t tell them is which direction to cut.
Look closely at what teams keep, and a pattern shows up. They rarely drop the segment they care about most. They rarely defer the feature that took the most effort to design. The assumption labeled low risk is often the one the venture can’t afford to get wrong. Each choice is defensible, and founders can explain any single cut with confidence. What comes out the other end is a venture that looks much narrower, with the founder’s core belief still untested, now surrounded by decisions that give the impression a great deal has been examined.
Watching this happen across cohort after cohort, we started calling it protective subtraction. It’s confirmation bias working under a constraint the bias didn’t anticipate.
We usually think of confirmation bias as a problem of gathering. People look for evidence that supports what they believe and skip past what doesn’t. That assumes there’s always room to keep gathering. In a structured process, the room is gone. Founders can’t keep everything, test everything, or watch everything. Something has to go. When the instinct that shapes what people gather also shapes what they discard, the result is a cut that narrows scope without reducing risk.
When the instinct that shapes what people gather also shapes what they discard, the result is a cut that narrows scope without reducing risk.
The distinction matters because protective subtraction goes unnoticed by the people doing it. A founder who narrows from four segments to one has done what the process asked. The output looks right. The difference only surfaces when you ask what the cut cost the founder in belief. One founder chose the segment with the strongest evidence. Another chose the segment that felt safest for their assumptions. Both submitted one segment.
It’s also hard to catch through self-reflection. Founders experience the decision as discipline, and in most ways it is. Something real was given up. The question that goes unasked is whether what they gave up was ever holding any weight.
Ranking What Could Break the Venture
The assumption a founder rates low risk is often the one everything else rests on
The first narrowing comes early, and it’s the one founders least recognize as a narrowing at all.
Before any customer conversation, the team lists what must be true for the venture to succeed. The list runs long. Some assumptions concern the customer, some the market, some the money, some the founders themselves. Then the team rates each one, because nobody tests twenty assumptions in a discovery cycle. Four or five get tested. The rest wait, and many of them wait forever.
That rating is a subtraction. It decides what gets examined and what goes unchallenged for the life of the venture.
Sung-jin is building a smart-home health-monitoring system for aging parents. His customer is the adult child, the one who lives three hours away and lies awake wondering whether a fall would go unnoticed until morning. He’s spoken to enough of these adult children to know the worry is real and the willingness to pay is real. His assumption list reflects that confidence. Purchase intent, rated medium. Price tolerance, medium. Integration with existing home systems, high, because that’s the part he finds technically uncertain. Regulatory questions, high.
Near the bottom, rated low, sits a single line about the parent accepting the sensors in their home.
Ask Sung-jin why that one is low, and he has an answer ready. The sensors are unobtrusive. There’s no camera. Adult children he’s spoken with are confident their parents will be fine with it. Every one of those points is true, and none of them is evidence about what a parent will actually do.
Look at what the low rating protects. Sung-jin’s venture rests on a proxy relationship in which the payer isn’t the person living with the product. If aging parents quietly turn off the sensors, or refuse installation, or agree in front of their child and unplug the unit the following week, nothing else on the list matters. Purchase intent means nothing if the product comes out of the wall. That assumption isn’t low risk. It’s the floor the entire venture is standing on.
The rating didn’t happen despite its importance. It happened because of it.
The pattern is consistent enough that we now watch for it directly. Teams over-assign the lower ratings, and the assumption most likely to be underrated is the one closest to the founder’s original conviction. Rating something high risk means saying out loud that the venture might rest on unverified assumptions, and many founders find a reason not to say it. The reasons are usually good ones. Sung-jin’s reasons were good ones.
There’s a structural fix that helps: it keeps two questions from collapsing into each other. Rate confidence separately from consequence. How sure are we that this is true, answered on its own. How much breaks if it turns out false, answered on its own. Sung-jin would rate his confidence about the parent high, and rightly so, since everything he’s heard supports it. Then he’d have to rate the consequence of being wrong, and the only available answer is that the venture ends.
Those two answers together produce a different list than either produces alone. High confidence paired with severe consequence is the cell that deserves the hardest look, and it’s invisible to any single-axis rating.
It deserves a closer look when the confidence rests on indirect evidence. Sung-jin’s confidence comes from adult children’s predictions about what their parents will accept. That’s testimony about someone else’s future behavior, which is the weakest kind of evidence a founder can hold and the easiest kind to mistake for the real thing.
What makes this first narrowing so consequential is that it governs every narrowing after it. The assumptions rated high are the ones that get customer conversations. Those conversations produce the evidence that shapes segment choice, then product decisions, then the numbers. An assumption rated low in the first week doesn’t just go untested. It gets built on, again and again, by work that has no reason to question it. By the time the venture reaches the financial stage, the untested belief is load-bearing across four decision stages, and nothing in the model can distinguish it from the verified parts.
Sung-jin will find this out eventually. The question is whether he finds it out from a parent in week three or from a churn rate in month fourteen.
Choosing Where to Enter
The segment a team gives up is rarely the one they were holding onto
The second narrowing is the one founders find hardest, and the one they’re proudest of once it’s over.
After initial discovery, a team usually has more potential customers than it can handle. Their interviews surfaced multiple groups with real pain, and they could reasonably build a solution for any of them. But the process asks for just one: a single beachhead segment where the first version gets built and early proof gets made.
Every founder gets the logic. Focus beats breadth. Serve one niche completely rather than three partially. The argument is well understood, and nobody pushes back on it.
Then comes the moment of choice, and the metric founders instinctively grab is market size.
The Meridian team is building an AI-native compliance platform for mid-market supply chains, and their discovery turned up something they weren’t looking for. The buying decision splits. Compliance owns the problem and feels the pain daily. IT owns the infrastructure and holds veto power over anything touching the data layer. Two decision-makers, different incentives, different vocabularies, different reasons to say no.
They’d gone in assuming compliance frustration drives the purchase. What they learned is that compliance frustration starts the conversation and IT ends it.
So they narrowed, and they narrowed hard. Out of everything cross-border compliance could mean, they chose Canadian manufacturers navigating rules-of-origin determinations under the current trade agreement. A specific regulatory obligation, a specific national context, a specific type of firm. Gone were pharmaceutical supply chains, food and agriculture, European operations, and the entire enterprise tier. That’s a substantial cut, and they made it deliberately.
Now ask what the cut removed and what it left standing.
It removed reach. Everything Meridian gave up was market they couldn’t have served well anyway at their size and stage. That’s a real sacrifice in the way a wider aperture is a real sacrifice, and it feels like discipline because writing it down is uncomfortable.
What it left standing was the belief that frustration with compliance moves a purchase forward. Their own discovery had already complicated that belief. The narrowing never touched it. Meridian picked their entry segment by asking where their solution fits best, and the answer to that question was shaped by the problem they set out to solve rather than by the decision structure they’d just uncovered.
Consider what a different criterion would have produced. Had Meridian asked where the purchase decision is easiest to observe and resolve, they might have landed somewhere else entirely. Firms where compliance and IT report to the same executive. Firms where a recent audit failure had already forced the two functions into the same room. That segment might be smaller. It would also produce evidence about the thing most likely to kill the venture.
They cut toward fit. The alternative was to cut toward risk.
Two forces meet at this moment, and both push the same way. Larger numbers hold attention better than smaller ones, so the segment with the bigger population keeps surviving the shortlist. And giving up market feels like a loss in a way that giving up certainty doesn’t. Once a team has done something that hard, the narrowing feels finished.
It’s worth being precise about what Meridian did well, because a lot was done well. The narrowing is specific enough to act on. The rationale is written down. The segment is reachable. Set them beside a team that leaves this stage still describing its customer as mid-market companies with compliance needs, and Meridian is far ahead.
The challenge is not about how well the team makes decisions at this stage. It is about whether they are focusing on the right target. Even when a team follows the process carefully, they may leave with the same core assumption they started with. If that assumption is now tied to a smaller segment, it may seem that testing is no longer needed.
At this point, the most helpful question is not about where we might succeed. Instead, we should ask where we can learn the fastest if our assumptions are incorrect.
Naming What the First Version Must Test
Everything the evidence demands, and one thing the first build can prove
The third narrowing is the sharpest in the whole process, and it’s the one teams argue about longest.
By this stage, a team has earned a long list of what the solution must do. Every requirement on it traces directly back to something a customer said or did during discovery. Then the process asks a question that feels like it contradicts all that hard work: out of everything on this list, what does the first version have to prove?
Not build. Prove. Those two concepts split apart here, and the space between them is where teams get stuck.
Take Amara, who is building a real-time clinical decision support tool for rural community health workers in Ghana. Her discovery was thorough and hard-won. Health workers walk between villages carrying paper protocols that go out of date, making judgment calls on patient referrals with no one to consult. Her findings reshaped the product entirely: phones are shared across teams, grid power is spotty, cellular connectivity comes and goes, supervisors want visibility into decisions, and field workers are wary of being watched.
All of that is real. All of it belongs in the final product. Her list of requirements runs fourteen items deep, and she can defend every single one.
Then she has to isolate the one core hypothesis the first version must test, and her list gives her nothing to go on. Fourteen valid requirements will not rank themselves.
Watch what happens when a founder in Amara’s position makes this choice. The natural instinct is to test whatever took the most effort to figure out. Offline functionality was her hardest engineering hurdle, and shared-device authentication was her cleverest design solution. Both feel essential because both cost her significant time. Neither, however, is the core assumption on which her venture rests.
The entire venture hinges on a single action: whether a health worker, standing in front of a patient, actually changes a referral decision based on the tool’s recommendation. That is the critical behavior. Every other feature is merely a condition to make that behavior possible, and a condition isn’t the behavior itself. A tool can run offline, log in three workers on one phone, and sync flawlessly when back in range, yet still sit untouched in a pocket while the worker defaults to old habits.
Effort is a terrible guide here, but it’s a persistent one. Sunk work makes a feature feel far more critical than it is. The requirements that soak up the most design hours are rarely the ones carrying the real behavioral risk. They’re just the ones the team found interesting to build.
There is an even worse version of this failure pattern. A team can complete every subsequent stage of the venture process without ever closing their initial behavioral test. We reviewed a project that reached the financial stage with the result of its first user test completely blank. Twelve iterations of the deck. The outcome placeholder was still empty. Revenue projections, cost structures, cash flows, dashboards, and funding plans were all fully built and internally sound, yet every line rested on a user action nobody had actually verified.
That team didn’t omit the test out of laziness. They were among the most diligent in the cohort. They subtracted the single test result that could have invalidated their entire premise, then built out all the downstream work anyway.
They subtracted the single test result that could have invalidated their entire premise, then built out all the downstream work anyway.
This is protective subtraction at full extension: scope narrows, effort stays high, and core exposure never drops by a fraction.
What makes this stage so difficult is that doing it right feels irresponsible. Amara has fourteen documented needs, but the process asks for only one behavioral question. Selecting it means putting out a prototype that visibly fails to meet most of what she knows her users require. Any reasonable founder resists that. It means putting her name on something she already knows is inadequate.
The key is realizing that a first version isn’t a final product. It’s just an instrument designed to capture one specific piece of evidence. That might mean a manual prototype, a concierge delivery model, or a supervisor answering calls in real time while Amara observes whether those answers alter clinical decisions in the field. Technical fidelity should be reduced only to the point where the core behavior can still occur and be measured. Below that line, the test loses all meaning. Above it, every extra feature is comfort armor for the founder.
Every extra feature is comfort armor for the founder.
The question isn’t “What does our complete solution need?” It’s “What specific action must happen right in front of us for us to know this works?”
Deciding Who to Reach First
Willingness to talk is not the same as willingness to act
The fourth narrowing looks like the second one repeated, and teams treat it that way. It isn’t.
The entry segment describes a population. These firms, this size, this situation. It’s a filter you could apply to a list. The early adopter is something else: the subset within that segment whose circumstances have already forced them into motion. Same segment, different evidence, different question. One describes who fits. The other identifies who is out of time.
Teams collapse them constantly. Roughly half the submissions we review at this stage hand back a demographic profile where a behavioral one was required, and the distinction is spelled out for them plainly beforehand. It still happens.
Fatima is building an executive access service for former Gulf-region consultants rebuilding their networks after returning home. Her segment is well drawn. Senior operators, ten to twenty years of regional experience, now in markets where their credibility doesn’t transfer and their contacts don’t answer. She’s spoken with dozens of them, and the conversations were good.
That’s the problem. The conversations were good.
Ask Fatima who she should approach first, and she’ll name the people who took her calls, engaged with the idea, offered introductions, and asked to be kept posted. They’re warm, and they’re enthusiastic, and there is nothing dishonest about any of it. But warmth is not urgency, and she’s about to select her first customers based on how the conversation felt.
Consider what’s actually available to her. Some of those contacts are eighteen months into the transition and have made peace with it. They find her idea interesting the way you find a good article interesting. Others are four months out, watching savings drain, with a specific deal that requires a specific introduction they cannot get. Both groups take her calls. Only one of them will change what they do next week.
The second group is frequently harder to reach, and it isn’t a coincidence. People in the acute phase are busy solving the problem. They cancel. They’re less generous with their time precisely because their time has a price on it right now. The people most available to a founder are often the ones least pressed by the thing the founder is solving.
That’s the mechanism. Availability doubles as a selection filter, selecting against urgency.
Notice that Fatima has done nothing wrong by any process standard. She has a defined segment, real discovery, and an early-adopter list she can justify. What she has subtracted is everyone outside her existing conversations, and the cut removed exactly the evidence she needed. Her list will produce positive signals. It will not produce a purchase pattern she can build on, and the difference won’t be visible until she tries to repeat what worked.
The key question focuses on behavior and looks at what has already happened, not what might happen. Instead of asking if someone would use a solution or how urgent it feels, we ask what actions they have taken. For example, we might ask what they have already tried, how much they spent, or when they last worked to solve the problem themselves. These questions help us understand real patterns and needs.
Someone who has cobbled together a workaround, paid for something that half-worked, or asked a favor they’d rather not have asked has already demonstrated the urgency. That’s a record of behavior, not a forecast of it. Fatima’s warmest contacts may have no such record at all.
The early adopter isn’t the person most excited about the venture. It’s the person already spending something to solve the problem without it.
Choosing the One Number
A metric that can only rise cannot tell you anything
The last narrowing is the smallest in scope and the hardest to execute because, by this point, there is nothing left to give up but attention.
A team arrives at the financial stage with a full instrument panel. Revenue by month, acquisition cost, lifetime value, contribution margin, runway, burn, conversion at every step. All of it built, all of it defensible. Then the process asks for one. Not the only number tracked, but the one number the team checks first, the one that would change what they do this week if it moved.
Renée is building a leadership-readiness program for officers transitioning out of military service. She’s a retired Army colonel, which means she knows this customer the way few founders know theirs. She has watched capable people leave the service and struggle to translate what they can do into language a civilian employer recognizes. The program exists because she has seen the gap up close for twenty years.
Ask Renée for her one number, and she’ll say enrollments.
It’s the obvious answer, and it is the wrong one, for a reason that has nothing to do with her judgment. She means total enrollments to date, the way most founders do, and a running total only goes up. Every month it rises; it tells her the same thing: that some people signed up. It cannot fall. It cannot warn her. She could watch it faithfully for a year while the thing her program exists to produce fails to happen, and the chart would climb the entire time.
Enrollments per cohort would at least be able to drop. Even that only reports demand.
Now consider what Renée’s venture actually rests on. Not enrollment, and not completion either, though completion is closer and at least it can fall. What the program exists to cause is a participant doing something differently in an interview or a negotiation because of what she taught them.
That behavior is observable if she builds a way to see it. The share of participants who complete a mock interview and demonstrate a defined improvement in translating their service record into terms a civilian employer recognizes. Specific, time-bounded, and capable of getting worse. That last property is the one that matters.
This is where the last cut gets made, and it’s not a cut in scope. Every metric she built stays on the dashboard. What she’s choosing is where her attention goes by default, and attention is the scarcest thing a founder has. The number checked first shapes what gets noticed, which shapes what gets fixed. Choose a number that only rises, and the venture can deteriorate in a direction the founder is not looking.
We see a clear pattern in the numbers. At this stage, roughly half of the submissions include dashboard figures that do not match the team’s financial model. When the numbers differ, the dashboard almost always presents the more favorable result. For example, a team may report 57 months of runway when its own inputs produce 5.7 months. Customer acquisition costs may be shown using the lower marginal rate rather than the actual blended rate. The same ratio may also appear three different ways in three different places, with the most favorable version given the greatest visibility.
None of these are lies. They are recollections, carried over from an earlier draft, a benchmark, or a moment when the number was better. What’s notable is that the drift has a direction, and it is the same direction as every other cut in this article.
Renée is not careless. She’s the most disciplined founder in this piece, and the metric she’d pick is the one nearly every founder in her position picks. That’s the thing worth considering. Protective subtraction doesn’t require a lapse in rigor. It operates through the choices that feel most natural to a person who cares about the venture, which is why rigor alone never catches it.
The test for a metric is not whether it captures something important. It’s whether it can move against you. A number that cannot deliver bad news isn’t a metric. It’s reassurance with a chart attached.
A number that cannot deliver bad news isn’t a metric. It’s reassurance with a chart attached.
What the Cut Costs
There are five key points at which a venture must narrow its focus, each requiring a different decision. Founders decide which assumptions to test, which customer segment to enter, what the first version must demonstrate, whom to approach first, and which number to watch.
These decisions do not follow a single method. For example, rating risk is a different process than choosing a metric. What connects them is the moment when a founder must decide what to set aside. Often, there is a tendency to let go of whatever feels least costly to believe.
Klotz found that people often overlook the option to subtract. Our observation is more specific and less reassuring. When people are placed in a process that requires subtraction, they will usually do it, and often do it well. The instinct to avoid subtraction does not vanish. Instead, it shifts. It no longer determines whether a cut gets made. It determines where.
The teams whose work informs this article were not careless. They were often thorough, conscientious, and able to explain every choice they made. What they deferred, reframed, or crowded out with less consequential work was the part most likely to challenge the belief on which the venture was built. As a result, their ventures became more focused without becoming more exposed. From the outside, those two things look identical.
The pattern is difficult for a founder to catch on their own. Each decision feels disciplined, and in fact, each one is a form of discipline. Something real is given up at every step. The question that often goes unasked is whether what was given up was actually important.
It is important to ask directly at each of the five points, and to pay attention to how you phrase the question. Instead of asking what you decided to focus on, ask what you chose not to examine. Then consider what it would cost if your decision were incorrect.
A founder who can answer this question at all five points has achieved something more valuable than simply narrowing their focus. They have identified the factor that could cause their venture to fail. This is the only kind of narrowing that produces real evidence.
Everything else just makes the venture smaller.
A note on the examples. The founders in this article are composites, built from patterns that recurred across behavioral venture process work and coaching discussions.
Klotz, Leidy. Subtract: The Untapped Science of Less. New York: Flatiron Books, 2021.
Venture for All® · © 2026 · Learn · Innovate · Thrive