07 · Risk and its management

"Risky" is a useless adjective until you specify which risk. The operational breakdown distinguishes four components, each with its own question and its own instruments of reduction.

Technological risk: does the mechanism work? In real conditions, at the foreseen costs, reproducibly? It is reduced with the progression of evidence of chapter 06: proof of concept, validation in the laboratory, demonstration in a relevant environment.

Market risk: does anyone want it, enough to pay for it, in sufficient numbers, reachable at sustainable costs? It is reduced with the validation of the problem (chapter 02) and with the demand experiments that this chapter deals with; chapter 05 showed that it includes the risk of adoption and of timing.

Execution risk: does the organisation know how to do what the plan requires, in the time and at the cost foreseen? It includes team, production (the MRL axis), operations, and the availability of the necessary competencies; chapter 13 will deal with the organisational side, and the law of the minimum of chapter 16 is its synthesis: the project yields according to the competency that is missing.

Ecosystem risk: does success depend on actors outside the project's control? Ron Adner (The Wide Lens, 2012) has shown that many failures attributed to the product are in reality failures of co-innovation (the necessary complements were not ready: the case of 3D televisions without 3D content) or of downstream adoption (an intermediary in the chain had no incentive to adopt). The diagnostic question: for the value promised to the end customer, who else has to innovate or adopt, and why would they do so within our timescale?

The breakdown produces the rule of sequence that governs the whole chapter: work first on the largest residual risk, because that is where a euro of validation buys the most information. A project that perfects the technology while the dominant risk is market risk is buying the wrong information, however well it buys it.

The strand of customer development (Steve Blank, The Four Steps to the Epiphany, 2005) and of lean startup (Eric Ries, 2011) starts from a diagnosis: in the conditions of a new company, the plan is a set of unverified hypotheses, and treating it as a plan to be executed is the founding error. The method that follows treats every element of the model (the problem exists, this segment feels it, it would pay this price, this channel reaches it) as a hypothesis to be verified with the least expenditure, in the build-measure-learn cycle.

The key instruments, in their correct use. The minimum viable product is not a shoddy product: it is the smallest experiment capable of verifying the riskiest hypothesis, and it does not need to be sellable (the ladder of product versions, from MVP to MMP, is in chapter 06); it may be a page that measures conversion, a series of structured interviews, a service delivered by hand behind a product facade (the "concierge"), even before it is a prototype. Validated learning distinguishes evidence from vanity metrics: what counts is behaviour that costs those who perform it something (time committed, data provided, money paid, contracts signed), not applause; the hierarchy of evidence of chapter 02, from the stated to the contractual, is the same. The pivot is the structured change of one component of the model that preserves the accumulated learning: the technology that changes market, the product that becomes a component, the channel that changes; it is the rational outcome of a refuted hypothesis, distinct both from obstinacy and from abandonment.

The practical breakdown that connects this method to section 1 distinguishes three families of hypotheses: desirability (do they want it?), feasibility (do we know how to build it?), economic sustainability (does the model hold: margins, acquisition costs, recurrence?). The sequence of experiments follows the largest residual risk, and the operational literature (Bland and Osterwalder, Testing Business Ideas, 2019) catalogues for each family the available experiments with their cost and the strength of the evidence they produce.

The two methods combine without friction once their respective roles are clarified: hypothesis validation defines what counts as evidence and how to produce it at low cost; stage-gate defines when the evidence is examined, by whom, and with what consequences for capital. A mature process is a funnel of experiments with decision gates: the phases produce validated learning, the gates convert it into investment decisions.

The chapter has insisted on buying information before committing capital; this last section fixes the limit of the principle, because information too has a price and protracted analysis is itself a decision (to delay) with its own costs.

The conceptual frame comes from decision theory: additional information is worth as much as its expected capacity to change the decision. If every plausible scenario of further investigation leads to the same choice, the investigation has zero value and its cost is a dead loss. To the direct costs of analysis are added the costs of delay: the time windows of chapter 05 close, competitors move, the best team does not wait. And the accumulation of analysis beyond the useful threshold is not neutral cognitively either: it feeds the feeling of control more than the quality of the choice, and becomes a socially acceptable form of postponement.

A window follows: before it, deciding is rash (the missing information would plausibly change the choice, and it costs little to obtain); after it, deciding is a duty (the residual information is costly, slow, or incapable of changing the choice). The discipline consists in declaring in advance which evidence would decide (which is then the structure of the gates of section 2: criteria first, examination afterwards) and in recognising that the residual risk at the moment of decision is not a defect of the method but its object: risk management does not promise certainty, it promises that the capital committed is proportionate to the available evidence, and that the evidence has been bought at the best price.

• R.G. Cooper, "Stage-Gate Systems: A New Tool for Managing New Products" (Business Horizons, 1990): the original formulation of the method; for the hybridisation with agile, Cooper and Sommer (2016).
• E. Ries, The Lean Startup (2011): the build-measure-learn cycle, MVP, validated learning, pivot; to be read together with S. Blank, The Four Steps to the Epiphany (2005), which is its foundation.
• D.J. Bland and A. Osterwalder, Testing Business Ideas (2019): the operational catalogue of validation experiments, classified by cost and strength of the evidence.
• R. Adner, The Wide Lens (2012): ecosystem risk, with the cases of missing co-innovation and downstream adoption.