The Funnel of Startup Failure

The previous article argued that a startup succeeds only when many independent things go right, and that this conjunction is why failure is the default. This article makes the conjunction precise. A startup is best pictured as a funnel, a sequence of stages that a venture must clear in order, each one conditional on the last, and survival is the chance of passing through all of them. The picture explains two things at once, why so few ventures emerge from the bottom, and where along the funnel most of them are actually lost. The treatment is general, and it is information rather than business advice.

A Brief History

The funnel is borrowed from sales and marketing, where it is older than any startup. Its ancestor is the attention, interest, desire, action model, proposed at the close of the nineteenth century to describe how a buyer moves from first notice to purchase. The same shape became the purchase funnel and then, in the digital era, the conversion funnel, which tracks how a population of visitors narrows at each step down to the few who pay. Startup practice adapted the metaphor twice. Once as a customer funnel, following a single user from acquisition through activation, retention, and revenue, and once, less formally, as a survival funnel, following the company itself through the existential stages it must clear to keep existing. This article concerns the second funnel. Its logic, that a chain narrows to the product of its stages, is the same in both, and it borrows one further idea from the theory of constraints, namely that a chain is governed by its narrowest link.

The Stages

A reasonable survival funnel has perhaps six stages, though the exact boundaries are a modeling choice rather than a law. First, a real problem, one that enough people genuinely have and would pay to remove. Second, a solution that works, a product that actually solves that problem. Third, product-market fit, the demonstrated pull of enough of the right people wanting the product strongly enough to return to it. Fourth, distribution, a repeatable and affordable way to reach those people. Fifth, monetization, charging a price they will pay that exceeds the cost of serving them. Sixth, durability, an advantage that lets the venture keep what it has won as others arrive to take it. A venture must clear every stage, and a failure at any one ends the journey regardless of strength elsewhere.

The Funnel as a Product

Let $A_i$ be the event that the venture clears stage $i$. Survival is the joint event that it clears all of them, $A_1 \cap A_2 \cap \cdots \cap A_n$. The chain rule of probability expresses the chance of that joint event as a product of conditional probabilities,

\[P(\text{survive}) = P(A_1)\, P(A_2 \mid A_1)\, P(A_3 \mid A_1 \cap A_2) \cdots P(A_n \mid A_1 \cap \cdots \cap A_{n-1}). \tag{1}\]

Write $p_i$ for the conditional probability of clearing stage $i$ given that every earlier stage was cleared. Then survival is simply the product

\[P(\text{survive}) = \prod_{i=1}^{n} p_i, \qquad p_i = P(A_i \mid A_1 \cap \cdots \cap A_{i-1}). \tag{2}\]

This identity is exact, and it assumes nothing about independence. The stages of a startup are deeply dependent, since a strong product helps distribution and a reached market helps reveal fit, but the conditioning absorbs every such dependence. Each $p_i$ already presumes that the venture stands where the earlier stages left it.

Why the Product Is Small

The product form carries the whole lesson of the previous article. A number of factors, each less than one, multiply to something far smaller than any of them. Suppose for a moment that every stage had the same probability $p$. Then survival is

\[P(\text{survive}) = p^{\,n}. \tag{3}\]

If each of six stages is cleared seven times in ten, survival is $0.7^6 \approx 0.118$, barely one venture in eight, even though no single stage looked especially hard. At even odds per stage, $0.5^6 \approx 0.016$, fewer than two in a hundred.

Real funnels are not uniform. Take an illustrative set of conditionals, a real problem at $0.6$, a working solution at $0.7$, product-market fit at $0.4$, distribution at $0.5$, monetization at $0.7$, and durability at $0.6$. Their product is

\[P(\text{survive}) = (0.6)(0.7)(0.4)(0.5)(0.7)(0.6) \approx 0.035. \tag{4}\]

About one venture in twenty-eight, a survival rate in keeping with the heavy failure the previous article reported. Failure is not a verdict on the founder. It is the arithmetic of multiplying many fractions.

The Bottleneck

The product form also says where effort is best spent. The marginal effect of improving a single stage follows from the product directly, since holding the others fixed,

\[\frac{\partial P}{\partial p_i} = \frac{P}{p_i}, \qquad \frac{1}{P}\frac{\partial P}{\partial p_i} = \frac{1}{p_i}. \tag{5}\]

The relative gain from improving stage $i$ is $1 / p_i$, which is largest for the smallest $p_i$. The leakiest stage offers the most leverage, exactly the lesson the theory of constraints draws for a production line. In the illustration, the weakest stage is product-market fit at $0.4$. Raising it to $0.5$ multiplies survival by $0.5 / 0.4 = 1.25$, a twenty-five percent improvement, from about $0.035$ to about $0.044$. Raising the working solution from $0.7$ to $0.8$, the same tenth of a point, multiplies survival by only $0.8 / 0.7 \approx 1.14$. The identical effort returns nearly twice as much when spent on the narrower stage. A founder who polishes a strong stage while a weak one leaks is improving the wrong number.

Where Startups Actually Die

The same product describes the shape of the funnel itself. The fraction of entering ventures that survive to the start of stage $k$ is the partial product

\[S_k = \prod_{i=1}^{k} p_i, \qquad S_0 = 1, \tag{6}\]

and the fraction of the original cohort lost at stage $k$ is the drop between successive levels,

\[L_k = S_{k-1} - S_k = S_{k-1}\,(1 - p_k). \tag{7}\]

A loss late in the funnel is a small conditional leak applied to a cohort already thinned, while a loss early falls on nearly everyone. With the illustrative conditionals above, the first stage removes forty percent of all entrants, the second eighteen, and the product-market-fit gate a further twenty-five, so more than four in five ventures are gone before they clear fit. The final three stages together account for about thirteen percent of the cohort, fewer than one in seven of all the deaths, for all the attention those late stages receive.

The model says leverage lies at the leakiest stage, and the evidence of the previous article says which stage that usually is. The dominant recorded cause of death, no market need, sits at the early stages, the real problem and product-market fit. These are also the widest parts of the funnel, where the most ventures enter and the largest fraction is lost. Yet founders, and the stories they read, dwell on the late stages, the scaling, the hiring, the fundraising, that only survivors ever reach. The funnel corrects the emphasis. Most ventures never reach the glamorous problems, because they are lost in the unglamorous early work of confirming that anyone wants the thing at all. The stages after fit, distribution and the advantage that makes a venture durable, matter greatly, but only for the ventures that survive long enough to face them.

Epistemic State

The per-stage probabilities here are illustrative, chosen to show the shape of the arithmetic, not measured from data. The decomposition into six stages is itself a modeling choice. Real ventures do not march through clean sequential gates. They iterate, revisit earlier stages, and pivot, so the funnel is a simplification of a process that loops. The chain-rule identity is exact given any chosen decomposition, but the numbers placed into it are not, and a different analyst would draw the stages differently.

What survives this caution is the structural conclusion, which does not depend on the particular numbers. Survival is a product of factors below one, so it is small, and the smallest factor carries the most leverage. That much follows from the form alone. Throughout, this is general information, and it is not business advice.

Out of Scope

The within-product customer funnel, its acquisition, activation, retention, revenue, and referral metrics, is a distinct and detailed subject left to the marketing literature. The cohort and retention mathematics that demonstrate product-market fit belong to the next article. The unit economics of the monetization stage, the cost to acquire a customer against the value that customer returns, belong to the article on distribution and getting paid. The work of clearing each stage, the remedies rather than the accounting, is the subject of the rest of the series and is not treated here.

Conclusion

A startup is a funnel, and survival is the product of the conditional probabilities of clearing each stage. Because the factors are many and each is below one, the product is small, and failure is the default for a reason no founder can wish away. The same form shows the remedy. Effort returns the most at the leakiest stage, which the record places early, in the search for a real problem and the pull of product-market fit. The next article takes up that stage directly, since it is both the widest part of the funnel and the place where the most ventures are lost.

Equations

This section collects the 7 display equations of the article in the order they appear, each under its number and reproduced exactly as it stands in the text. A title in bold names an established result, and a title in italics beginning with the word Bespoke names a construction made for this analysis. Of the 7, 5 are established and 2 are bespoke. Every quantity is a dimensionless probability or a fraction of the entering cohort, and each stage probability is conditional on clearing every earlier stage, so the stage figures cannot be compared with unconditional failure rates.

Equation 1, Chain Rule of Probability. The probability of clearing every stage is the product of the probability of clearing each stage conditional on clearing all earlier ones. chain rule of probability conditional probabilities

\[P(\text{survive}) = P(A_1)\, P(A_2 \mid A_1)\, P(A_3 \mid A_1 \cap A_2) \cdots P(A_n \mid A_1 \cap \cdots \cap A_{n-1}). \tag{1}\]

Equation 2, Survival as a Product of Stage Conditionals. Writing each conditional as a stage probability makes survival the product of the stage probabilities, an identity that assumes nothing about independence. chain rule of probability

\[P(\text{survive}) = \prod_{i=1}^{n} p_i, \qquad p_i = P(A_i \mid A_1 \cap \cdots \cap A_{i-1}). \tag{2}\]

Equation 3, Bespoke Survival with Uniform Stage Probabilities. Under the simplifying assumption that every stage has the same probability, survival is that probability raised to the number of stages.

\[P(\text{survive}) = p^{\,n}. \tag{3}\]

Equation 4, Bespoke Survival Through an Illustrative Six-Stage Funnel. Survival is evaluated for an illustrative set of assumed conditionals for problem, solution, product-market fit, distribution, monetisation and durability, giving about one venture in twenty-eight.

\[P(\text{survive}) = (0.6)(0.7)(0.4)(0.5)(0.7)(0.6) \approx 0.035. \tag{4}\]

Equation 5, Marginal and Relative Effect of One Stage. Differentiating the product with the other stages held fixed shows that the relative gain from improving a stage is the reciprocal of its probability.

\[\frac{\partial P}{\partial p_i} = \frac{P}{p_i}, \qquad \frac{1}{P}\frac{\partial P}{\partial p_i} = \frac{1}{p_i}. \tag{5}\]

Equation 6, Fraction Surviving to a Stage. The fraction of the cohort that survives to the start of a stage is the partial product of the stage probabilities up to it.

\[S_k = \prod_{i=1}^{k} p_i, \qquad S_0 = 1, \tag{6}\]

Equation 7, Fraction Lost at a Stage. The fraction of the original cohort lost at a stage is the drop between successive survival levels, the surviving fraction times the stage failure probability.