Measurable Risk vs. Fundamental Uncertainty in Venture Decisions

Welcome. This course examines contrarian VC not as a temperament of “disagreeing with the crowd,” but as a way to reason about how collective beliefs form, become self-reinforcing, and sometimes reverse. We begin at the deepest level: what exactly is unknown when one makes an early-stage investment decision?

The distinction between measurable risk and fundamental uncertainty matters because venture capital often gives both the appearance of numerical precision. A model can assign a probability of success, a valuation can imply a return multiple, and a market map can show comparable companies. Yet none of these automatically means that the relevant probabilities are genuinely known or even coherently knowable. In this lesson, you will learn to identify which parts of a venture decision are probabilistic risks, which are fundamentally uncertain, and what that distinction should change in an investment process.

---

1. Risk has a distribution; uncertainty challenges the distribution itself

Frank Knight’s original distinction is still the cleanest starting point.

A risk is a situation in which relevant outcomes can be specified and probabilities have a defensible foundation. That foundation might be a stable physical mechanism, a large and relevant data set, or a repeatedly observed process. The probabilities need not be perfectly known. They may be estimated with error. But it makes sense to speak of a probability distribution and improve its estimate.

A simple risk decision can be represented as:

$$\mathbb{E}[X]=\sum _{i=1}^n{p}_i{x}_i$$

where $x_i$ are defined outcomes and $p_i$ are meaningful probabilities attached to them.

Fundamental uncertainty, often called Knightian or radical uncertainty, arises when that representation is not secure. The issue is not merely that $p_i$ is difficult to estimate. It may be unclear whether:

• the set of relevant outcomes is complete;
• historical observations belong to a useful reference class;
• the mechanism generating outcomes will remain stable;
• knowledgeable people even share the same model of causation.

In these circumstances, assigning a single probability such as “there is a 20 percent chance this company becomes a category leader” can conceal rather than clarify disagreement. One investor may mean “20 percent conditional on the market existing in five years”; another may be implicitly assuming that distribution will solve itself; a third may believe a regulatory shift makes the category structure unknowable. The numerical agreement is superficial.

This is especially important in pre-seed and seed investing. A company may have an observable burn rate, known contractual terms, and a testable product milestone while still depending on a future market architecture that does not yet exist.

A useful corrective follows from this distinction:

Risk is not the same as volatility, and uncertainty is not the same as lack of data.

A volatile but mature public-market security can be risky without being fundamentally uncertain: its returns may vary widely, but decades of observations, market microstructure, and a reasonably stable business model make probabilistic analysis meaningful.

Conversely, an emerging category may have a small number of apparently excellent data points without being reducible to risk. Two early AI-research labs achieving extraordinary outcomes do not establish a stable distribution for all “AI labs.” The apparent sample may reflect an unusual conjunction of technical timing, founder reputation, access to compute, strategic partnerships, and temporarily abundant capital. The category itself may change in response to the initial winners.

---

2. The venture decision contains both risk and uncertainty

The practical mistake is to classify an entire investment as either risky or uncertain. In reality, an early-stage decision is a bundle of claims, and each claim deserves a different epistemic label.

Consider a pre-seed company developing software for regulated clinical workflows.

Decision component	Usually closer to	Why
Monthly cash burn under a specified hiring plan	Measurable risk	Costs can be modelled from contracts, salaries, and reasonably known operating assumptions.
Time required for a defined integration	Risk mixed with uncertainty	Past implementations provide some evidence, but the customer’s internal politics and systems may differ materially.
Probability that a pilot converts under an established procurement process	Potentially measurable risk	A relevant cohort and a stable sales process can support an estimate, albeit with wide error bars.
Whether regulators, buyers, and providers will converge on a new workflow standard	Fundamental uncertainty	The future institutional arrangement is itself unsettled.
Whether specialist investors will fund the next round in three years	Fundamental uncertainty	It depends partly on future narratives, comparables, capital markets, and the category’s social legitimacy.

The distinction is therefore decision-relative. Even a familiar metric can become uncertain when the surrounding regime changes. Customer acquisition cost may be a manageable risk in an established channel. It becomes much less stable if a platform changes its rules, a new model changes user behavior, or competitors with abundant capital distort the economics.

The inverse is also true. A seemingly speculative startup may contain narrow questions that can be made tractable. A laboratory company might face radical uncertainty about eventual category economics, but a particular technical experiment can still test whether a specific process reaches a target yield or cost threshold.

The disciplined investor separates these layers instead of averaging them into a single “probability of success.”

A spectrum, not a binary

It is helpful to use three working categories:

1. Estimable risk
   There is a reasonably relevant reference class, an agreed outcome space, and a mechanism expected to remain sufficiently stable. The estimate may be noisy, but probability modelling is useful.

2. Model uncertainty
   Several credible models describe the same situation. Outcomes may be specified, but the right probabilities or causal structure are contested. For example, a software category may have plausible but sharply different adoption curves depending on whether buyers see it as a budget-saving tool or as a strategic system of record.

3. Radical uncertainty
   Key outcomes, mechanisms, or categories are not yet sufficiently defined. The question “what is the probability?” may be premature, because one cannot state the relevant event without embedding a speculative narrative.

The boundary is not fixed. A seed-stage climate-software company may move from radical uncertainty toward model uncertainty as regulation crystallizes, early buyers behave consistently, and its implementation process becomes repeatable. Conversely, a company can move the other way when an external technological discontinuity invalidates the reference class on which its forecast depended.

---

3. Why a range of probabilities is not merely a wider forecast

Suppose an investment produces value $V$ if successful and zero if it fails, while the initial capital required is $C$. Under a conventional risk model, with estimated success probability $p$, the crude decision criterion is:

$$pV>C$$

This is not a complete VC model, but it illustrates the logic: a single probability gives a ranking.

Under fundamental uncertainty, the investor may regard the success probability as plausibly lying in a range:

$$p\in [\underline{p},\overline{p}]$$

Now the decision can fall into three regions:

• If $\underline{p}V>C$, the investment looks attractive even under the pessimistic probability consistent with the investor’s understanding.
• If $\overline{p}V<C$, it looks unattractive even under the optimistic case.
• If

$$\underline{p}V\le C\le \overline{p}V$$

then the model does not rank the options decisively.

That third region is central to early venture investing. It is not a calculation error. It is an honest result: plausible interpretations of the world support opposing choices.

A decision still has to be made, of course. But the ultimate choice then reflects something beyond expected value: a default toward action or inaction, a view on the value of learning, a belief about the founder’s capacity to adapt, or a judgment about whether a specific milestone can turn uncertainty into more measurable risk. These are substantive judgments. Treating them as if they had been produced mechanically by a spreadsheet only obscures where the investment thesis actually lives.

This also explains why a portfolio of apparently similar companies can be misleading. The companies may share a sector label, but their uncertainty structures may differ radically. One may be taking ordinary execution risk in an established market; another may be attempting to create the very market whose size the model assumes.

---

4. Exploration has value precisely because the distribution is unclear

Fundamental uncertainty does not imply passivity. It creates a reason to value learning.

An investment in a relatively known category mainly offers its direct expected financial return. An investment in a genuinely novel technology or category may also generate information that changes future decisions: whether to make follow-on investments, whether to form a more ambitious thesis, and which technical or commercial direction is viable.

This insight has a contrarian implication. During a period of category neglect, the market may underprovide experimentation because the informational benefits spill over. If one investor funds a difficult technical experiment and it succeeds, competing investors may learn from the result without having paid for the experiment. The private expected return can therefore look inferior to the broader informational value created.

But “learning value” is also easily abused. It should not be a euphemism for a weak thesis. It has genuine force only when all three conditions hold:

• the investment can produce a discriminating signal, rather than a vague demonstration;
• that signal would change a meaningful future capital-allocation decision;
• the investor can plausibly act on the information before it becomes fully commoditized.

A well-designed technical milestone, paid pilot, or narrowly scoped regulatory result can be valuable not because it guarantees success, but because it removes a decision-relevant ambiguity. By contrast, a generic press launch or vanity partnership may create attention without making the underlying distribution more knowable.

---

5. The danger of false precision

A spreadsheet is often useful in VC. It forces assumptions into view, shows which conditions must hold for an outcome to be investable, and exposes hidden dependence on a future financing round or exit multiple. Its danger begins when the output is mistaken for a measurement of reality.

Watch John Kay and Mervyn King’s discussion of radical uncertainty before turning this into an operational discipline.

The appropriate response is not “never quantify.” It is quantify conditionally and label the conditions.

For a seed memo, that can mean distinguishing among:

• Measured inputs: current burn, contracted revenue, conversion rates from a relevant cohort, technical test results.
• Model assumptions: a future pricing structure, sales-cycle duration after a product change, likelihood of a follow-on round under specified market conditions.
• Uncertain propositions: whether a new buyer category will emerge, whether regulation will define a market boundary, whether a celebrated early winner will legitimize or crowd out adjacent companies.

A point estimate may still be useful for comparing scenarios, but it should never erase the provenance of the number. “A 15 percent probability of category leadership” is much less informative than: “This estimate assumes incumbents remain fragmented, implementation costs fall below a stated threshold, and the next financing market remains open to companies with this profile.”

The second formulation tells you what evidence would matter. It also makes it possible to notice when the thesis has changed.

---

6. A practical uncertainty map for an early-stage decision

Before relying on a probability-weighted return model, make an uncertainty map. This is not generic investment-process bureaucracy; it is a way to prevent a category narrative from smuggling speculative claims into apparently hard numbers.

For each central claim in the thesis, record four items:

1. The proposition
   State it in a falsifiable form. For example: “Mid-market manufacturers will adopt autonomous quality-control software without requiring a systems-integrator-led deployment.”

2. Its status
   Is it estimable risk, model uncertainty, or radical uncertainty?

3. Its evidence base
   Identify whether the evidence is direct observation, an adjacent reference class, founder testimony, a comparable financing, or a social signal such as a prestigious investor’s involvement.

4. The most informative observation
   Specify what would materially change the view: repeated paid deployment, a technical benchmark, a procurement-policy shift, a competitor’s failure for a diagnostic reason, or the absence of follow-on financing after a defined interval.

This discipline makes a key distinction visible: an event can be informative without being confirmatory. A brand-name VC investment, for example, may signal that a sophisticated actor has performed diligence. But it may also change the company’s own future by improving recruiting, customer access, and financing availability. Later lessons will examine that reflexive effect. At this stage, the essential point is simpler: do not treat the signal as an independent observation of the category’s fundamental value.

For contrarian work, the map should include a further question: What is the market currently treating as measurable that is actually uncertain? In a boom, this is often the durability of demand, the availability of future capital, or the transferability of one winner’s economics to an entire category. In a bust, the inverse error occurs: a salient failure is treated as proof of a stable negative distribution even though the failure may reflect one business model, one financing environment, or one premature implementation path.

---

Key takeaways

• Measurable risk exists when outcomes and their probabilities have a defensible basis. Estimates can be imperfect without losing their meaning.
• Fundamental uncertainty concerns the outcome space, causal mechanism, or probability distribution itself. It cannot be solved merely by using a wider confidence interval or a more elaborate spreadsheet.
• Early-stage VC decisions combine both. Decompose the investment into specific claims rather than labelling the whole company “high risk.”
• Quantitative models remain valuable when treated as conditional reasoning tools. Their assumptions, not their precision, deserve the most scrutiny.
• Under uncertainty, an investment can possess learning value if it produces information that would alter future decisions. That is distinct from simply owning a speculative asset.
• A strong contrarian thesis identifies not only what consensus believes, but also which of consensus’s apparently numerical assumptions lack a sound probabilistic foundation.

Next, we turn to Keynes’s beauty-contest model: once investors recognize that fundamental value is uncertain, they may increasingly try to anticipate what other investors will believe and fund. That shift from first-order judgment to higher-order beliefs is a major source of venture-category booms and reversals.