AI strategy and use case design in 2026: why investment fails without an owner and a measure

Worldwide AI spending reaches $2.67 trillion in 2026, yet only 37% of organisations attribute any EBIT impact to it. The difference is not model choice. Here is the test a use case has to survive before it gets funded.

Most AI strategies fail before anything is built, and the reason is almost always the same: nobody named the problem, the owner, or the measure. Four of the five causes Gartner gives for abandoned generative AI projects are decisions made before a line of code is written.

The gap between spend and result

Spending, and results

• Worldwide AI spending, 2026. $2.67 trillion. Up 49.5% on 2025. Gartner forecast, September 2026.
• Organisations attributing any EBIT impact. 37%. Essentially unchanged year on year. Only 6% are high performers with 5% or more EBIT impact.

Money is not the constraint, and has not been for some time.

Source: Gartner AI spending forecast, 16 September 2026; McKinsey The State of AI, 1,719 respondents across 97 nations, 2026.

Gartner forecasts worldwide AI spending at $2.67 trillion in 2026, up 49.5% year on year. Against that, McKinsey's 2026 survey of 1,719 organisations across 97 countries found 37% attribute any EBIT impact to AI, a figure that has barely moved, and 6% qualify as high performers with 5% or more EBIT impact.

Money is not the constraint. It has not been for a while.

What the organisations getting value do differently

What separates the organisations getting value

None of these findings is about model choice.

Source: McKinsey The State of AI 2026; Gartner survey of 353 data and analytics leaders, April 2026.

Two findings sit underneath all of this. Gartner's April 2026 survey of 353 data and analytics leaders found that organisations with successful AI initiatives invest up to four times more, as a proportion of revenue, in data quality, governance, AI-ready people and change management. McKinsey found 73% of high performers had fundamentally redesigned workflows because of AI, against 25% of everyone else, and high performers were twice as likely to say senior leaders demonstrated real commitment.

Neither finding is about models. Both are about the organisation around the model.

The question a use case has to survive

A use case is fundable when three things are true, and the third eliminates more candidates than the first two combined.

• Someone is losing something measurable today. Hours, dollars, cycle time, escaped defects. If the loss cannot be quantified, the benefit will not be either.
• The right answer is knowable from data you already hold. Not data you could collect, or data a vendor says exists. Data in your systems now.
• A named person will change what they do when the output arrives. Not "the business will have visibility." A person, an action, a Monday morning.

The third test is the one most strategy documents skip, and it is the one that predicts whether anything ships. A model whose output lands in a dashboard nobody has a reason to open has already failed, it just takes two quarters to find out.

Productivity is the smaller prize

Gartner's July 2026 survey of 204 finance leaders found 45% of CFOs say their AI investments lean towards productivity while 20% lean towards decision quality. In the same research, functions investing in initiatives that created new value propositions were more than twice as likely to report high realised value.

Productivity cases are easier to write and easier to approve, which is exactly why they crowd out the ones worth more. If every use case on your list saves time and none of them changes a decision, the list is incomplete rather than complete.

Governance is a multiplier, not a brake

Governance practices, ranked by measured effect on value

Likelihood of achieving high generative AI value, against organisations not doing each practice.

Source: Gartner survey of 360 organisations of 250+ employees, fielded May to June 2025, published November 2025.

This is the finding that surprises executives most. Gartner surveyed 360 organisations in late 2025 and found those conducting regular AI system assessments were three times more likely to achieve high generative AI value. Safely expanding rollouts scored 3.3 times. Role-based training doubled it.

The mechanism is not mysterious. Systems that are assessed get fixed, and systems that get fixed get used.

What we would do first

Write the decision record before the strategy document. A prioritised list of problems, a named owner per problem, the measure that says it worked, and a date to revisit. One page. If you cannot fill the owner column, that is the finding, and it is more valuable than another technology assessment.

Our AI readiness work produces exactly that, with the data-layer gaps mapped against each candidate so the sequencing is evidence-based rather than enthusiasm-based. The Data Strategy Roadmap Generator will give you a sequenced twelve-month view in a few minutes if you want a starting position first, and where to start with AI by function covers the operational version of the prioritisation question.

Frequently asked questions

Why do AI strategies fail before anything is built?

Because the investment is not tied to an outcome someone owns. Gartner's January 2026 assessment names five causes for generative AI projects abandoned after proof of concept: lack of business value, data not ready, escalating total cost of ownership, responsible AI as an afterthought, and poor change management. Four of those five are decisions made before a line of code is written.

What separates organisations getting value from AI?

Foundations and workflow redesign, not model choice. Gartner's April 2026 survey of 353 data and analytics leaders found organisations with successful AI initiatives invest up to four times more, as a share of revenue, in data quality, governance and change management. McKinsey's 2026 survey of 1,719 organisations found 73% of high performers had fundamentally redesigned workflows, against 25% of everyone else.

How much is being spent on AI, and is it working?

Gartner forecasts worldwide AI spending at $2.67 trillion in 2026, up 49.5% on 2025. Against that, McKinsey found only 37% of organisations attribute any EBIT impact to AI, essentially unchanged year on year, and just 6% qualify as high performers with 5% or more EBIT impact. Spending is not the constraint and has not been for some time.

Should AI investment target productivity or decision quality?

Most are targeting productivity, and the evidence suggests that is the smaller prize. Gartner's July 2026 survey of 204 finance leaders found 45% of CFOs say their AI investments lean towards productivity against 20% towards decision quality. Separately, functions investing in initiatives that create new value propositions were more than twice as likely to report high realised value.

How do we choose which use cases to fund?

Start from where the work already is rather than from the technology. Rank candidates by the size of the loss today, whether the right answer is knowable from data you hold, and whether a named person will change what they do when the output arrives. The third test eliminates more candidates than the first two, and it is the one most strategy documents skip.

What does good governance do for AI value?

More than most leaders assume. Gartner's November 2025 survey of 360 organisations found those conducting regular AI system assessments were three times more likely to achieve high generative AI value. Safely expanding rollouts scored 3.3 times, role-based training twice, and third-party governance products 1.9 times. Governance in this context is not a brake, it is a measured multiplier on outcomes.

Do we need an AI strategy document?

You need a decision record, which is a different artefact. A list of prioritised problems, the owner of each, the measure that says it worked, and the date the decision gets revisited. Thomson Reuters found in June 2026 that 18% of professionals say their organisation has no AI strategic direction and around half work where strategy is absent or misaligned, so the bar is lower than you might fear.