Key Takeaways
- Only 22 percent use AI productively: For the large majority of companies, artificial intelligence never leaves the testing stage.
- 95 percent have a strategy, 8 percent have a return: Writing the strategy turns out to be the easy part of the exercise.
- Governance triples the odds: Companies with clear accountability for AI results report measurable ROI three times as often.
- Weak data integration is the top failure cause: More than 60 percent name their own data foundation as the main reason projects stall.
- Scaled AI is worth about 20 percent: Where AI runs across core processes, productivity in those processes rises by roughly a fifth.
- Budgets are shifting from 3 to 5 percent: AI is moving out of the experiment column and into the regular annual budget.
Introduction
Almost every company now has an AI strategy. Only about one in five can show what that strategy produced. Seven studies published in recent months arrive at the same conclusion from different directions, and the pattern behind the numbers is remarkably consistent.
The obstacle is not the model. It is the plumbing around it: data that cannot be reached, responsibilities nobody claimed, and goals that were never connected to a business metric. Technology is rarely the problem. People and processes are.
This article sorts the findings into the decisions a management team can actually make this quarter — on governance, on data, on budget, and on the gap between what leadership believes and what the team experiences.
The 22 Percent Reality
The KPMG Global AI Pulse for the second quarter of 2026 puts the number bluntly: only 22 percent of companies use AI productively. Everywhere else it stays in pilots, proofs of concept and enthusiastic individual use that never becomes a process.
A companion finding from the first quarter makes the contrast sharper. While 95 percent of companies report having an AI strategy, only 8 percent report an actual return. More than half describe themselves as being at the very beginning, and just 13 percent are scaling.
The KPMG Global Tech Report 2026 adds the middle of that distribution: 74 percent of companies see value in AI, but only 24 percent achieve a return across more than one use case. One successful pilot is common. A portfolio that pays is rare.
Governance Decides, Not Technology
The single most useful number in this year's research: companies that clearly assign responsibility for AI outcomes report demonstrable ROI three times as often. Not better models, not bigger budgets — a named owner.
The effect repeats with cost transparency. Organisations that know what their AI operations actually cost reach a measurable result five times more often. That is not an accounting detail. If you cannot state the running cost, you cannot compute a return, and the project will be defended with anecdotes instead of numbers.
Capgemini names the same set of levers from the other end: board attention, partner ecosystem, governance and data. Three of those four are organisational. Only one is technical, and it is the one most companies start with.
Data Integration Is the Number One Failure Cause
More than 60 percent of companies name weak data integration as the main reason their AI projects fail. That is the highest-ranked cause in the KPMG data, and it matches what I see in practice in midsize companies almost every week.
The typical picture: offers live in the CRM, contracts in a document management system, project notes on a network drive, and the actual knowledge in three long-serving employees. An assistant with access to one of those four sources will answer fluently and incorrectly.
The fix does not require a data lake. It requires deciding, for one concrete use case, which sources the answer depends on and making exactly those reachable and current. That is a two-week project, not a two-year programme.
What Scaled AI Is Actually Worth
Where AI does run across core processes, the KPMG report measures an average productivity increase of about 20 percent in those processes. Capgemini finds 66 percent of companies reporting better performance in both productivity and decision quality.
The EY European AI Barometer adds the commercial side: broad AI use shows up in profit, while a narrow focus on cost cutting leaves most of the potential untouched. Savings have a floor. Better decisions and faster throughput do not.
The Indian market offers a useful contrast. EY reports 47 percent of surveyed companies scaling several use cases simultaneously, while more than 95 percent still spend less than a fifth of their IT budget on AI. Scaling is evidently not primarily a spending question.
Budget: From Three to Five Percent
Capgemini observes AI budgets growing from roughly 3 to 5 percent of the annual budget. Accenture recommends exactly that move explicitly: fix AI spending as a defined share of the annual budget instead of funding it from project leftovers.
A fixed share changes behaviour more than the amount does. It forces a portfolio decision — which use cases get funded this year and which do not — and it ends the pattern where every initiative competes as a special case against the operating business.
Accenture's second recommendation follows from the first: tie AI goals directly to company KPIs. An initiative that cannot name the KPI it moves is research. Research is legitimate, but it should be labelled and budgeted as such.
The Perception Gap Between Leadership and Team
EY points at a cost factor that appears in no tool comparison: the gap between how management assesses AI use and how teams actually experience it. Leadership sees a rollout. The team sees a licence and a login.
That gap is expensive because it hides the real blockers. If nobody reports that the assistant cannot reach the price list, the management dashboard will keep showing a successful rollout while the tool sits unused.
Closing it is cheap: ask the people doing the work, at a fixed interval, what works and what does not. Twenty minutes per team per month surfaces more actionable detail than the next platform evaluation.
Your Next Steps
- Name one person accountable for AI results — for the business number, not for the tool.
- Write down what your current AI use actually costs per month.
- Pick one use case and make exactly the data it needs reachable and current.
- Fix the AI budget as a share of the annual budget instead of funding it from leftovers.
- Tie every funded initiative to one named company KPI.
- Ask the teams monthly what is blocking them, and act on the answers.
Mentioned Tools & Resources
- KPMG Global AI Pulse Q2 2026: Productive AI use, accountability and cost transparency. kpmg.com
- KPMG Global AI Pulse Q1 2026: From adoption to orchestration — strategy versus realised value. kpmg.com
- KPMG Global Tech Report 2026: Value perception, multi-use-case ROI and data integration. kpmg.com
- Accenture, From Early Impact to Enduring Advantage: AI budgeting and KPI alignment. accenture.com
- Capgemini Research Institute, AI Perspectives 2026: Performance gains and growing AI budgets. capgemini.com
- EY European AI Barometer 2025: Financial results of broad AI use and perception gaps. ey.com
Frequently Asked Questions
Why do only 22 percent of companies use AI productively?
Most AI initiatives stay in the pilot stage because nobody owns the result, the target metric was never defined, and the data the use case depends on is scattered across systems. The technology is rarely the limiting factor.
What is the biggest mistake in AI projects?
Weak data integration, named by more than 60 percent of companies. An assistant that cannot reach the right source produces fluent answers that are wrong, and user trust rarely survives more than a few of those.
How much productivity does scaled AI deliver?
Where AI runs across core processes, productivity in those processes rises by about 20 percent on average. The gain is concentrated in the processes actually covered — it is not a company-wide multiplier.
What does AI governance mean in practice?
A named owner for the business result, documented rules for data and privacy, transparent operating costs and a fixed review date per use case. Companies with that setup report a measurable return three times as often.
How much budget should go into AI?
Current research shows AI budgets moving from roughly 3 to 5 percent of the annual budget. The decisive factor is not the percentage but that it is fixed in advance and allocated as a portfolio rather than case by case.
Why does a documented AI strategy not guarantee results?
Because 95 percent of companies have a strategy and only 8 percent report a return. A strategy describes intent. Results require ownership, usable data, a named KPI and the discipline to stop what does not work.
Conclusion
Seven studies, one message: the distance between an AI strategy and an AI result is organisational, not technical. Accountability, cost transparency, reachable data and a budget that is decided rather than scraped together — those are what separate the 22 percent from everyone else.
None of that requires a transformation programme. It requires one owner, one use case, one number and one honest review date. Think first, then pick the tool — the order matters more than the choice.
