By Alex Adamopoulos, Chairman and CEO, Emergn

A board asking what its AI investment has delivered needs more than a list of pilots, licenses and training sessions. It needs to understand what has improved and whether that improvement justifies the cost. Those are reasonable questions. A busy program can still struggle to answer them.

Emergn’s 2026 study, The Value Gap, surveyed 700 senior leaders in the US and UK. On average, they estimated that their organizations lose 2.4% of annual revenue to transformation and AI initiatives that fail to deliver the expected value. For a business with $1 billion in annual revenue, that would be $24 million a year.

The research points to a management problem that buying more technology cannot resolve. Organizations need a way to decide which problems deserve investment, test whether the work is helping and change course when the evidence calls for it. AI makes those decisions more urgent.

The mistake is treating the purchase as the change. Buying a tool gives people access to a capability. Getting value from it may require a different workflow, better data or a decision that currently takes weeks to move through the business. People need to learn how to use it in their own work. Without that attention, the company can spend more and see little improvement.

Where investment loses focus

Consider a company with AI pilots running separately in finance, HR, marketing and operations. Each may have a reasonable purpose. But if each sponsor defines success differently, leadership has no reliable basis for comparing the investments or deciding where to put the next dollar.

The cost is also more than the software bill. The same people may be asked to support several experiments while keeping the business running. Data teams become stretched. Managers spend time reconciling reports. A pilot that looks inexpensive on its own can be costly when those demands are included.

There is nothing wrong with running experiments that don’t succeed. An experiment can be useful because it establishes that an idea should go no further. The waste begins when work continues after the evidence no longer supports it, or when nobody has agreed what evidence would be enough to make that decision.

Making decisions in the work

Before funding an initiative, leaders should be clear about the problem it is meant to solve and who owns the result. Establish a baseline and agree how progress will be assessed. Early discovery may need to test an assumption about customer demand. A service already in use should be judged on what it does for customers and the business. The evidence should fit the stage of the work.

Funding should reflect that evidence. Release enough to answer the next important question, then review the result before committing more. Give someone explicit authority to continue, change or stop the work. A review without a decision leaves the same costs running.

Teams also need shared standards for using data, selecting tools and managing risk. These should make the boundaries clear enough for people to work within them. If every small experiment has to repeat the same approval process, governance becomes another delay. Equally, faster delivery is no improvement if it creates unreliable services or exposes customer information.

Test the approach on live work. Put a few initiatives through a full cycle, from an initial problem to an investment decision and a review of what happened. Adjust the process where it adds effort without improving the decision. The people doing the work should be able to explain how it helps them.

That capability needs to stay with the client. External partners can bring expertise and help teams make progress, but the engagement should leave the organization better able to operate and improve the work itself. Learning belongs in the delivery process, with people applying it to real decisions while experienced support is available.

A more useful account of value

Boards need reporting that connects investment to outcomes. Revenue and cost matter, with the full cost of implementation and operation included. Customer retention, service quality and reduced risk can also justify investment. Where the benefit is not yet visible in the accounts, explain how it is being measured and when a financial effect is reasonably expected.

Hours saved, for example, are not automatically cash saved. Leaders need to know whether the capacity has reduced overtime, increased useful output or allowed people to improve service. Otherwise, a productivity claim can remain a theoretical benefit that nobody has converted into value.

The choice of model and supplier matters. So do the skills of the people using them and the way their work is organized. Leaders are responsible for making those choices work together and for acting when the results fall short. Buying AI does not discharge that responsibility.