The Best Way to De-Risk Innovation Is to Make Failure Cheaper
Corporate innovation works better when companies make uncertainty cheaper to resolve, stage capital, and let evidence determine what scales.
Large companies have spent decades becoming extraordinarily good at reducing risk, which is one of the reasons they became large companies in the first place. They standardize processes, create governance, establish procurement and security requirements, build sophisticated financial controls, and introduce layers of accountability around decisions that could affect millions of customers, thousands of employees, or billions of dollars of enterprise value. These mechanisms are not bureaucratic accidents; they are rational responses to scale, and in the parts of the business where reliability matters more than discovery, they work remarkably well.
The problem is that most of these systems were designed to protect businesses whose economics, customers, and operating models are already reasonably well understood. They are much less useful when the question in front of management is whether an entirely new product, capability, or business should exist at all, yet companies regularly apply the same institutional logic to both. An uncertain idea is researched, modeled, socialized, budgeted, and moved through successive layers of approval before anyone has had the opportunity to test the assumptions on which the business case depends. This is usually described as de-risking innovation; in many cases, it does precisely the opposite.
The central problem with early-stage innovation is not that some ideas will fail. Any company pursuing genuinely new opportunities should expect that. The problem is discovering that an idea does not work only after the organization has committed significant capital, time, and political energy to it. Once an initiative has a budget, a team, an executive sponsor, a roadmap, and a place in the strategic plan, the economics of shutting it down have changed; the financial sunk cost matters, but the organizational sunk cost often matters more because people become invested in proving that the original decision was correct. A better approach to de-risking innovation begins with a different objective: not reducing the probability that an experiment fails, but reducing the cost of discovering that it should.
The difference between execution risk and discovery risk
Mature corporations are generally very good at managing execution risk. If a company knows that it wants to enter a new geography, migrate a critical system, construct a facility, or roll out an established product to another customer segment, planning and coordination can meaningfully reduce the probability of a bad outcome because the underlying objective is relatively well understood. Innovation contains a different category of risk: management may not yet know whether customers care enough about a problem to change their behavior, whether a technically viable product can support attractive economics, whether an AI system performs reliably in the environment where it will actually be used, or whether a new capability belongs inside the existing business at all.
Research can narrow these questions, but eventually many of them cannot be resolved in a conference room; the information required to make the decision is produced by the experiment itself. This distinction matters because governance systems designed for execution risk can become counterproductive when applied to discovery risk. Requiring greater certainty before an experiment begins sounds prudent, but when the uncertainty is irreducible without experimentation, the organization can end up spending substantial amounts of money trying to predict something it could have learned more cheaply by building. The effort to make innovation safer can therefore make being wrong considerably more expensive.
Venture capital offers a useful model for allocating capital under uncertainty
There is an obvious place to look for an alternative. Venture capital has spent decades allocating money to businesses whose outcomes are profoundly uncertain, and its record is hardly evidence of perfect foresight; quite the opposite, the model is interesting precisely because it assumes foresight will be imperfect. The fundamental mechanism is staged commitment. A seed investor does not generally finance a company on the assumption that every important question has already been answered; the investment is intended to allow the company to answer enough of those questions to determine whether a larger investment is warranted. As evidence accumulates, capital follows; when the evidence deteriorates, capital should stop. Venture investors do not eliminate uncertainty before investing; they attempt to price it and increase their exposure as uncertainty falls.
Corporate innovation often follows the reverse sequence. Significant organizational work goes into creating enough confidence to authorize an initiative, after which the initiative receives the people and capital required to execute against a relatively fixed plan. The distinction may appear procedural, but economically it is substantial: one model treats investment as a series of options that can be exercised as information improves, while the other risks making a substantial commitment before some of the most important information exists. This is why venture thinking has relevance well beyond venture capital. Its most transferable insight is not that startups move quickly or tolerate failure; it is that uncertain opportunities should be financed differently from established ones.
What a failed experiment is worth
Once innovation is understood as a process of reducing uncertainty, failure itself needs a more precise definition. Consider two companies exploring the same new business: the first spends ten weeks establishing that customers value the product but will not pay enough to support the required economics, and management shuts the initiative down; the second spends eighteen months building the product, integrates it into core systems, hires a dedicated team, and reaches exactly the same conclusion. Both ideas failed commercially, but only one process failed economically.
The first company purchased valuable information at relatively low cost; it now knows something about the market that it did not know ten weeks earlier and can redeploy its capital and management attention elsewhere. The second company purchased essentially the same information at a dramatically higher price. This suggests that one of the more useful measures of an innovation system is not simply the percentage of initiatives that succeed, but the amount of capital and time required to determine which initiatives deserve further investment. An organization in which nearly every experiment becomes a scaled initiative may not have developed extraordinary foresight; it may simply be experimenting too close to what it already knows, or allowing organizational commitment to substitute for evidence.
The objective should therefore be neither to celebrate failure nor to eliminate it. Both ideas are intellectually lazy; the objective is to make uncertainty less expensive to resolve.
AI changes the economics of experimentation
The timing of this matters because AI is changing more than the products companies can build; it is also changing the cost of discovering whether those products should be built. Historically, software was expensive evidence. Even a serious prototype could require months of work across product, design, and engineering before a company had something credible enough to expose to customers, and when the cost of experimentation is high, it is rational to spend more time deciding which experiments deserve to be run.
AI changes that calculation by increasing the productive capacity available to relatively small teams. A 2026 quasi-experimental study examining developers using Ant Group’s CodeFuse coding model found an increase in code output of more than 50 percent among developers with access to the tool, with the strongest effects concentrated among less-experienced developers. Code output is obviously not equivalent to customer value, product quality, or innovation success, so it would be irresponsible to translate that finding directly into a claim that AI makes venture building 50 percent faster; what the research does provide is evidence that generative AI can materially increase software-development capacity under real operating conditions.
The strategic implication is more interesting than the productivity statistic. As the cost of producing functional software falls, the relative economics of prediction and experimentation change: a company that previously had to choose one idea because building three was prohibitively expensive may increasingly be able to test several. Management can move some of the resources historically spent trying to determine, in advance, which idea is correct toward generating evidence about several ideas and allowing reality to make more of the decision.
There are also indications that the economics of corporate venture creation are shifting. McKinsey’s research on corporate venture building found that respondents in its 2025 survey reported lower investment requirements before new ventures reached break-even than respondents to the previous year’s survey, with the weighted average falling from approximately $125 million to $77 million. The same research found advanced technological capabilities, including generative AI, appearing frequently among companies reporting successful venture-building programs. These are survey findings, not a controlled experiment, and they do not establish that AI caused the reduction in investment requirements; what they do suggest is that the technical capabilities available to venture builders and the economics of venture creation are changing at the same time.
For CEOs and boards, the implication is that AI should not only appear on the list of opportunities being considered by the innovation portfolio; it should change the architecture of the portfolio itself.
From project governance to portfolio governance
Most large organizations still manage innovation primarily as a collection of projects. Projects have sponsors, budgets, milestones, teams, and expected deliverables; once approved, the management system is largely designed to help them reach completion. A venture portfolio has a different objective: its purpose is not to maximize the percentage of projects completed, but to continuously reallocate scarce capital and talent toward the opportunities generating the strongest evidence.
That distinction has implications for governance. Before an experiment begins, management should be explicit about which uncertainty the investment is intended to resolve and what evidence would justify the next commitment of capital; the criteria for increasing investment should be established while the organization is still relatively dispassionate about the outcome. Projects that fail to generate the necessary evidence should lose resources, while those that do should gain them. The process is closer to capital allocation than traditional program management.
Experienced corporate venture builders appear to benefit from repetition. McKinsey has found persistent differences between companies that build ventures serially and those that approach venture creation episodically, including stronger reported growth and scale among experienced builders. This should not be surprising: a company that launches a new venture every few years has relatively few opportunities to improve the institutional capability required to do it well. Venture building, like M&A integration or operational excellence, becomes more sophisticated through repeated practice.
This is why the conversation about corporate innovation needs to move beyond whether a company has an innovation lab, a venture arm, or an AI strategy. Those are organizational artifacts; the more consequential question is whether the company has developed a repeatable mechanism for allocating capital to opportunities whose outcomes cannot yet be known.
De-risk the commitment, not the ambition
None of this argues for corporations taking smaller strategic risks; in many industries, the opposite will be required. AI is altering cost structures, customer expectations, competitive boundaries, and the technical feasibility of products quickly enough that incrementalism may itself become one of the more dangerous strategies available to an incumbent. The distinction is between the size of the ambition and the size of the initial commitment: companies should be willing to pursue opportunities capable of becoming material businesses without behaving as though every promising hypothesis already deserves the cost structure of one.
That requires a different conception of prudence. In an established business, prudence often means reducing variance and preventing mistakes; in an innovation portfolio, prudence means maintaining the ability to change your mind cheaply as new information arrives. It means preserving optionality until evidence warrants commitment and then being willing to commit aggressively when it does.
For senior leadership, this ultimately makes corporate innovation less a question of methodology than one of capital allocation. Every company has a finite amount of money, talent, and management attention available for pursuing growth outside the core; the strategic question is how much information the organization can purchase with those resources before it is required to make larger commitments.
The companies that become exceptional at innovation over the next decade may not be the ones that predict the future most accurately. In periods of technological discontinuity, that is an unrealistic standard even for excellent management teams; they will be the companies that build systems capable of learning faster than their competitors, abandoning weak assumptions before they become expensive, and concentrating capital quickly when the evidence becomes compelling. The objective is not to make innovation safe, because genuinely new businesses cannot be made safe in advance; the objective is to make uncertainty economically manageable.
The best way to de-risk innovation is not to eliminate the possibility of being wrong. It is to make being wrong inexpensive enough that you can afford to find out.
--
Sources
- McKinsey & Company, “How CEOs Are Turning Corporate Venture Building Into Outsize Growth”
https://www.mckinsey.com/capabilities/business-building/our-insights/how-ceos-are-turning-corporate-venture-building-into-outsize-growth - McKinsey & Company, “The Way to Win in Corporate Venturing: Serial Building and AI”
https://www.mckinsey.com/capabilities/business-building/our-insights/the-way-to-win-in-corporate-venturing-serial-building-and-ai - Leonardo Gambacorta, Han Qiu, Shuo Shan, and Daniel M. Rees, “Generative AI and Labour Productivity: A Quasi Experiment on Coding,” Journal of Financial Stability, 2026
https://www.sciencedirect.com/science/article/pii/S1572308926000458