MAUT Decomposition, Real Options Aggregation

Multi-Attribute Utility Theory has been standard in high-stakes infrastructure decisions for decades. Keeney and Raiffa used it to evaluate airport development strategies for Mexico City. Keeney, Renn, and von Winterfeldt used it to structure Germany's energy policy. In each case, the evaluation was decomposed into independent attributes, scored separately, and aggregated. The decomposition prevents a strong score on one attribute from biasing the assessment of another. Each attribute gets its own evaluation pass. The structure enforces discipline.

Real options theory has been standard in pharmaceutical R&D and energy exploration for just as long. When a pharma company evaluates a drug pipeline, each clinical phase is an option: the right, not the obligation, to invest further. A Phase I success purchases the option to run Phase II. A Phase II failure terminates the chain. The pipeline's value is the sum of option values across the entire staged sequence, each weighted by its probability of advancing.

Neither framework has been applied to venture creation with any structural rigor. Venture evaluation still runs on gut feel, pattern matching, and opaque "proprietary playbooks." Our Venture Studio Operating System (VSOS) borrows the decomposition principles of MAUT and the aggregation logic of real options. The decomposition ensures reliable inputs. The aggregation produces a valuation that respects the staged, conditional nature of venture creation.

Decomposition Principles from MAUT

The 33 indicators across six dimensions are designed on MAUT principles. Each indicator is scored independently through a separate evaluation pass. Willingness to Pay is assessed without reference to Technical Feasibility. Customer Acquisition Cost is assessed without reference to Market Size. The structural requirement is preferential independence: the score on one attribute can be assessed without knowledge of the scores on any other.

This is the precondition for the decomposition to mean anything. When a single evaluator scores an entity holistically, "this looks like a 7 out of 10," every attribute contaminates every other. A compelling founder story inflates the demand score. A large market estimate inflates the feasibility score. The halo effect produces systematically mispriced evaluations.

Separate scoring calls eliminate this. The evaluator assessing Willingness to Pay sees only the information relevant to Willingness to Pay. The evaluator assessing Regulatory Complexity sees only the information relevant to Regulatory Complexity. Each indicator produces a score and an epistemic state that reflects the quality of evidence behind the score. The decomposition is clean. The scores are uncontaminated.

In standard MAUT, these scores would then be aggregated (using a weighted additive or multiplicative utility function) into a composite that ranks competing alternatives. That aggregation step does not happen here. A venture studio at a gate is not ranking alternatives against each other. It is evaluating a single entity and asking whether the evidence supports advancement. The decomposed indicator scores do not get rolled into a composite. They flow directly into a different framework entirely.

rNPV for Aggregation

The value of a venture is not a static evaluation of attributes at a point in time. It is a sequential chain of conditional decisions, where the value at each stage depends on the probability of surviving to the next. MAUT (in any form) evaluates an alternative. What a venture studio needs is a framework that values a sequence of options.

Each entity in the VSOS pipeline passes through a sequence of gates: AI screening, human screening, opportunity review, investment committee. Each gate is an option in a chain. At every stage, the studio spends capital to investigate. If the entity advances through all gates, it reaches spin-out and produces terminal value. The option chain value from any stage k forward is the probability-adjusted terminal value minus the probability-adjusted cost of each stage required to get there:

$$OV_k = TV \prod_{j=k}^{n} P_j - \sum_{j=k}^{n} \left[ C_j \prod_{l=k}^{j-1} P_l \right]$$

Where:

• $OV_k$ = option chain value from stage $k$ forward (the rNPV of the entity from where you stand now)
• $TV$ = terminal value (the equity value realized at spin-out)
• $P_j$ = probability of advancing through gate $j$
• $\prod_{j=k}^{n} P_j$ = compound probability of surviving every gate from $k$ to exit
• $C_j$ = cost of investigation at stage $j$
• $\prod_{l=k}^{j-1} P_l$ = compound probability of reaching stage $j$ from stage $k$

The first term is the terminal value weighted by the probability of reaching it. The second term is the sum of each stage's investigation cost weighted by the probability of incurring it. The option chain value is what remains.

When $P_k$ goes to zero at any stage, the compound probability of reaching every subsequent stage goes to zero with it. The terminal value contribution disappears. The costs of downstream stages disappear (you will never incur them). The chain is truncated.

This is the risk-adjusted net present value (rNPV) framework, adapted from pharmaceutical R&D where the same logic has governed drug pipeline valuation for decades. A Phase I trial purchases the option to run Phase II. A Phase II failure terminates the chain. The structural parallel to a venture studio's gate sequence is exact, and the math transfers directly.

A standard rNPV in pharma would discount each term for time, dividing future cash flows by $(1+r)^t$ where $r$ is either the weighted average cost of capital or a risk-free rate (an ongoing debate among rNPV practitioners, since the probability adjustments already capture development risk and a WACC-based discount may double-count it). The formula above omits this. In a lean venture studio where the time between gates is measured in weeks rather than the years-long clinical phases of drug development, the discount factor is negligible. Studios with abnormally long development cycles or high cost of capital may want to reintroduce it.

A deliberate simplification: the model treats $TV$ as a single terminal value at spin-out, which means it assigns zero recovery value to killed entities. In practice, a kill produces portfolio-level information value: market intelligence that redirects other entities, validated dead ends that save future investigation costs, domain knowledge that transfers. That value is real, and the kill/advance framework accounts for it. But it operates outside this equation. The rNPV values the entity. The portfolio benefits of a kill accrue to the studio. The formula is also conservative: it treats all value as realized at terminal exit, ignoring intermediate liquidity events (secondary sales at funding rounds) that have a higher probability of occurring than the full exit. This makes the formula a lower bound on entity value.

Where the Indicator Scores Go

The 33 indicators that MAUT principles decompose are the same objects that drive the rNPV calculation. Each indicator feeds at least one of three components in the valuation: probability of advancing at a given stage, terminal value at exit, or cost to operate through a given stage.

An indicator like Willingness to Pay feeds probability of advancing. If customers will not pay, the entity fails a downstream gate. An indicator like Market Size feeds terminal value. A validated market does not affect whether the entity advances through the next gate, but it determines how much the entity is worth if it survives the chain. An indicator like Regulatory Complexity feeds cost. Higher complexity means more expensive investigations to resolve the uncertainty, which reduces the net value at each stage.

The mathematical structure of the rNPV equation (multiplicative across probability and terminal value, additive for cost) determines which indicators have leveraged impact and which have bounded impact.

This is where the indicators do triple duty. Each indicator is simultaneously a decomposition axis (scored in isolation on MAUT principles), an rNPV input (feeding P, TV, or Cost at a specific stage in the option chain), and an information value driver (its current epistemic state determines where the next dollar of investigation should go). The decomposition ensures the scores are reliable. The rNPV determines what those scores are worth. The resolution value calculation determines which score to resolve next.

Compensatory and Non-Compensatory Factors

The rNPV structure reveals which factors are compensatory and which are not.

Some factors are genuinely compensatory: a smaller market can be offset by stronger unit economics, because both feed terminal value additively. A weaker competitive moat can be offset by lower customer acquisition costs, because both affect the magnitude of the payoff without threatening the entity's ability to reach it. Within the same component of the rNPV equation, tradeoffs are real.

The non-compensatory factors are the ones that feed probability of advancing. Because probability and terminal value interact multiplicatively, a near-zero probability at any stage drives the entire downstream chain toward zero. No amount of terminal value compensates for a stage where the entity almost certainly fails.

Consider two entities. Entity A has moderate Market Size (moderate TV) but strong unit economics and low customer acquisition costs. The terminal value indicators partially offset each other. The rNPV reflects the tradeoff accurately. These factors are compensatory.

Entity B has strong Market Size indicators (large TV), strong Economics indicators (favorable Cost), but a Willingness to Pay indicator that has been Tested and scored near zero. A weighted composite score would look passable. The strong dimensions pull the average up. The weak demand indicator gets diluted. The rNPV calculation produces a different answer. Willingness to Pay feeds probability of advancing at the opportunity review gate. Near-zero probability at that gate means $P_k \approx 0$, which means the entire downstream chain (investment committee, spin-out, terminal value) is effectively eliminated. The large market and favorable economics are real, but they are multiplied by a probability that collapses the chain. Willingness to Pay is non-compensatory.

The rNPV equation identifies the structural positions where non-compensatory factors live: inside the $\prod$ operator, feeding $P_j$ at each gate. VSOS maps specific indicators to those positions. These are the kill criteria: named indicators that, if they resolve negatively at a given stage, drive $P_j$ toward zero and truncate the downstream chain. They are encoded in the system before the investigation runs, not decided after the results come in. When an investigation resolves a kill criterion negatively, the entity is terminated because the math says no configuration of the remaining indicators can recover the lost value.

The pharma analogy is precise. A Phase II trial that demonstrates the drug is toxic does not reduce the drug's value. It eliminates Phase III. The company does not run a weighted average of toxicity against efficacy and decide the drug is "somewhat viable." Toxicity is non-compensatory. The trial result truncates the chain.

Kill criteria in VSOS work the same way. They are the indicators whose resolution, if negative, truncates the rNPV chain at that stage. The rNPV structure identifies which positions are non-compensatory. VSOS specifies which indicators occupy those positions.

But a score alone does not trigger a kill. A Willingness to Pay score of 2 at Hypothesized (State 1) is a weak signal resting on inference. The true value could be anywhere. Acting on it would mean killing entities based on guesses. The same score of 2 at Tested (State 3), backed by primary evidence from a landing page test or customer interviews with purchase commitment, is a finding. The epistemic state determines when a score is trustworthy enough to act on. The rNPV structure determines whether that trusted score is fatal. Kill criteria operate at the intersection: an indicator that feeds $P_j$, whose score has reached a sufficient epistemic state for the current gate, and whose value at that epistemic quality drives $P_j$ toward zero. All three conditions must hold. The position in the equation, the quality of the evidence, and the score itself.

Decomposition and Aggregation

MAUT's contribution to this framework is its decomposition principles: preferential independence, separate scoring, clean inputs. Without these, the indicator scores feeding the rNPV would be contaminated by halo effects and holistic bias, and the valuation would be built on noise.

The rNPV's contribution is the aggregation: a forward-looking, probability-weighted valuation that respects the sequential nature of venture creation, distinguishes compensatory from non-compensatory factors, and maps kill criteria to structural consequences rather than arbitrary thresholds.

Neither framework is used in its standard form. MAUT's aggregation step is replaced. The rNPV's inputs are produced by a decomposition architecture borrowed from decision theory rather than estimated holistically. The result is a system where resolving any single indicator simultaneously updates the evaluation, the option chain valuation, and the investigation priority across the portfolio.