AI Is Putting a Price Tag on Bureaucracy

How AI Changed What the Cost of Failure Really Measures

By Dan Toma

Artificial intelligence has fundamentally changed the economics of product development. According to Cursor’s latest Developer Habits Report, developers are coding significantly faster than they were a year ago, while Nvidia recently reported that AI has enabled its engineering teams to produce roughly three times more code than before. Whether the exact productivity gain is twofold or threefold is almost beside the point. Building software has become dramatically cheaper and faster than it was only a few years ago.

Naturally, this raises an interesting question: what happens to the Innovation Accounting indicator, Cost of Failure, when the cost of building collapses?

Recently, I’ve heard arguments that Cost of Failure is becoming a less relevant Innovation Accounting KPI in the age of AI. If the cost of building continues to fall, then surely the cost of failing falls with it.

The opposite may be true.

AI hasn’t made Cost of Failure less relevant. It has simply changed where the cost comes from – and, as a result, what the metric tells us.

Cost of Failure was never just about building

The idea that Cost of Failure only measures the cost of building something is the root of the misconception.

Cost of Failure has never been solely about development costs. It is a compound indicator made up of three distinct costs that every organization incurs while bringing new products to market. The first is the Cost to Build the product or service. The second is the Cost to Learn whether customers actually value it by generating sufficient evidence. The third is the Cost to Decide what to do with that evidence – whether to stop, pivot, persevere or scale.

Cost of Failure = Cost to Build + Cost to Learn + Cost to Decide

AI hasn’t changed the equation. It has changed which part of the equation dominates.

Historically, the Cost to Build was by far the largest component of the equation. Software development was expensive, engineering teams were large and development cycles lasted months or even years. It was therefore natural to interpret a high Cost of Failure primarily as an engineering problem.

AI has changed what Cost of Failure measures

There is little debate that AI has dramatically reduced the Cost to Build. Code generation, automated testing, AI-assisted design and low-code platforms have compressed development timelines and reduced the effort required to create digital products. As organizations successfully adopt these capabilities, the Cost of Failure should logically decline.

The mistake is assuming that every component of the equation declines at the same rate.

Learning still requires customers. AI can summarize interviews, synthesize research and generate hypotheses, but it cannot tell you whether customers will actually buy your product. That still requires real-world evidence. While the Cost to Learn may be lower today because prototypes can be built and tested much faster, generating meaningful evidence and allowing experiments to run their course still takes time.

The same is true – perhaps even more so – for the Cost to Decide.

The Cost to Decide isn’t the cost of the meeting. It’s the cost of waiting for the meeting.

Once a team has generated sufficient evidence to justify a decision, every additional day before that decision is made continues to consume resources. Teams remain allocated, budgets remain committed and opportunities to redeploy people to higher-potential initiatives are delayed. In practice, the Cost to Decide is the fully loaded cost of keeping a team and its initiative alive while the organization decides what to do next.

A simple example illustrates the point. If a team costs €8,000 per day to run and the decision to stop or continue the initiative is delayed by 45 days while waiting for the next portfolio review, the organization incurs an additional €360,000 in Cost to Decide. That money isn’t spent building. It isn’t spent learning. It is simply the cost of waiting.

With or without AI, most organizations still review innovation portfolios quarterly, allocate funding annually and require multiple approval layers before teams receive permission to continue or stop an initiative. AI has compressed execution, but many organizations still govern innovation at the same pace they did before AI.

Ironically, AI may amplify this problem. As the Cost to Build falls, organizations can pursue far more ideas in parallel. Instead of reviewing ten initiatives, leadership may suddenly find itself reviewing fifty. The engineering bottleneck shrinks. The decision-making bottleneck grows. AI increases an organization’s capacity to build faster than its capacity to make decisions.

As the Cost to Build shrinks, the Cost to Learn and the Cost to Decide account for an increasingly larger share of the total Cost of Failure.

A high Cost of Failure no longer tells us primarily that engineering is expensive. Increasingly, it tells us that organizations are taking too long to learn – or taking too long to act on what they have already learned.

Twenty years ago, a high Cost of Failure often reflected the economics of software development. Today, it increasingly reflects the economics of organizational decision-making.

That is precisely why AI hasn’t made Cost of Failure less important. It has made it a far better diagnostic tool.

The new bottleneck is organizational

Most executives already know their organizations are slower than they would like. They know investment decisions take too long. They know projects remain alive long after the evidence suggests they should be stopped.

What they rarely know is what that bureaucracy actually costs.

Organizations that continue to govern innovation through quarterly portfolio reviews, annual funding cycles and rigid approval processes will simply build unsuccessful products faster than ever before. AI won’t eliminate waste – it will amplify the consequences of slow governance.

The difference has nothing to do with technology. It has everything to do with governance.

The greatest opportunity AI creates isn’t faster coding. It’s faster decision-making.

Otherwise, organizations simply replace one bottleneck with another. Instead of waiting for engineering teams to build products, they wait for leadership teams to make decisions about products that have already been built and validated.

The companies that benefit most from AI won’t necessarily be those with the best models or the fastest developers. They’ll be the ones that make better decisions, faster.

AI didn’t kill Cost of Failure. It made it a better diagnostic tool, putting a dollar value on slow and inefficient governance.

Dan Toma is CEO and Partner at OUTCOME and a leading voice on how established companies build sustainable top-line growth. His work focuses on the systems that enable growth – from governance and funding to portfolio management and execution – helping organizations move beyond “innovation theatre” to measurable business results. He is the award-winning co-author of The Corporate Startup, Innovation Accounting, and Open Innovation Works, and has advised executive teams across the financial services, energy, pharmaceutical, healthcare, and engineering sectors.

Dan was named to the Thinkers50 Radar in 2020.