EY CEO Outlook 2026: AI Costs Belong in the Business Units

The EY CEO Outlook of October 2026 shows that German companies manage their AI costs in the wrong place: at 47 percent of them primary responsibility lies with the CIO or IT, while only 8 percent place it in the business units where the benefit arises. According to an EY-Parthenon survey of 100 companies published on 15 October 2026 with the CEO Outlook Survey, only 18 percent see their AI costs comprehensively and in real time, although 88 percent expect costs for generative AI to rise over the next three years. At 50 million euros in revenue, a doubling costs 450,000 euros a year by our calculation, the full margin lead KfW Research measures for AI users among German SMEs. Executive teams should therefore place cost responsibility where the benefit is measured before spending rises.

The EY CEO Outlook shows a gap between budget discipline and AI cost transparency

Seven in ten German companies usually or almost always stay within their AI budgets, according to EY-Parthenon, although only 32 percent have introduced comprehensive cost monitoring processes and a further 26 percent have such processes in part. 18 percent see their AI costs comprehensively and in real time, compared with 16 percent worldwide according to EY. 24 percent of German companies have no reliable overview of their AI costs and return on investment. Martin Gerber, CFO of EY Germany, sees many companies in a deceptive comfort zone, because whoever does not measure AI costs cannot manage them.

At 47 percent of companies IT owns AI costs, at 8 percent the business unit does

In almost half of the companies, a function that cannot judge the benefit of the applications is responsible for AI costs. At 47 percent this is the CIO or IT, at 16 percent several functions share it and at 8 percent the business units own it, which EY names as one reason for the management gap.

In the release, Gerber calls it a natural reflex to hand cost control for AI to IT. Because AI is used in very different areas with different benefits, he concludes that cost responsibility belongs where the benefit arises, since a CIO often cannot judge this for the whole company.

In organisation design terms, Gerber describes a separation of decision rights and accountability. IT can cap spending without knowing which initiative loses its benefit, while the business unit can expand usage without feeling the costs. How decision and accountability can coincide is described in Clarifying decision rights.

For many companies a combination makes sense, because shared platforms are bought centrally while usage per application is charged to the business unit that gains the benefit.

Only one in five companies has a business case for all major AI initiatives

The majority of respondents lack a yardstick for rising AI costs. According to EY, 20 percent of companies run a sound calculation for all major AI initiatives, 29 percent do so for selected projects, 33 percent only occasionally and 14 percent not at all.

At 47 percent of companies a cost increase can therefore usually be neither justified nor refuted, so it either passes unchecked or is stopped across the board. Without monitoring, budgeting and clear business cases, Gerber says, every statement about the economics of AI remains speculation. How to calculate benefits and costs in advance is described in AI business case, how to prove the benefit after launch in Measuring the ROI of AI projects.

A business case also covers workflow change and training, because the benefit depends on them. 62 percent of German CEOs name process redesign as a central productivity lever, against 46 percent worldwide. Sandra Krusch, Managing Partner of EY-Parthenon in Germany, says that AI achieves little when it is placed on outdated processes. 53 percent of German CEOs see their company building skills too slowly, which is addressed in Building AI capabilities in the organisation.

More than half expect costs for generative AI to rise by over 50 percent

57 percent of the companies surveyed expect their costs for generative AI to rise by more than 50 percent within three years. Overall 88 percent expect rising costs, 27 percent more than double and 5 percent unchanged costs. These cumulative shares yield the distribution below.

Costs grow with usage, because language models are billed per token and assistants per user and month. A pilot can therefore cost many times as much in regular operation once more cases or agents with several model calls are added. The cost items involved are described in Controlling AI costs.

At 50 million euros in revenue, doubled AI costs consume the margin lead of AI users

At 50 million euros in revenue, doubled AI costs add 450,000 euros a year and thus 0.9 percentage points of net margin. This is the gap by which, according to KfW Research, the net margin of a typical German SME using AI exceeds that of one without AI, at 4.5 against 3.6 percent. If the benefit does not grow with the costs, the measured margin lead is used up.

Today AI costs in this example equal one fifth of profit, after a doubling two fifths, so the increase pays off only with at least 450,000 euros of additional benefit a year. Among the 24 percent without a reliable overview, a gap of this size shows up only in the annual accounts.

The possible benefit is set out in KfW study on AI in German SMEs, spending plans in Planning an AI budget.

FinOps for AI assigns every AI expense to an initiative and a cost per case

FinOps for AI applies cost management from the cloud to AI applications. The FinOps Foundation describes FinOps as collaboration between engineering, finance and the business in three recurring phases: making costs visible and allocating them (Inform), improving usage and rates (Optimize) and putting the measures into operation (Operate). If total costs rise at a stable cost per case, an initiative is growing as planned. A rising cost per case points to how the solution is built, such as longer inputs or extra model calls. The metrics management reporting needs for this are described in AI KPIs for management reporting.

The EY figures rest on 100 self-assessments and show a reliable order of magnitude

The figures are self-assessments and suit orders of magnitude more than fine comparisons. The release does not define “comprehensively and in real time” and states neither company sizes nor the survey period. EY provides the study only via an order form, so all figures here come from the press release. For 18 percent and 100 respondents, the 95 percent confidence interval is around plus or minus 7.5 percentage points, assuming a random sample. The difference from the 16 percent worldwide is therefore not reliable, while even the upper end of just over 25 percent describes a minority.

Executive teams in mid-sized companies should assign AI costs to the unit that gains the benefit

Executive teams can close the gap shown in the EY CEO Outlook within two quarters if they start with a complete overview and place responsibility for each initiative in the business unit:

1. Build the overview: on behalf of the CFO, controlling compiles all AI spending of the past twelve months within four weeks, including surcharges in existing software and expensed subscriptions. It is complete when every expense has an initiative and the share of revenue is set against the EY average of 0.9 percent.
2. Assign ownership: in the following four weeks the executive team names a business owner for every major initiative while IT keeps platform costs centrally. The target is a business owner for at least 90 percent of AI spending.
3. Catch up on business cases: within eight weeks every owner presents a business case with benefit, cost per case and the costs of training and workflow change. Success means every initiative above about 25,000 euros a year has one.
4. Make costs visible: by the end of the first quarter IT sets up cost centres or keys per application and an alert before the budget limit. The measure is that every business unit sees its AI costs monthly without asking.
5. Decide every quarter: from the second quarter, the executive team and heads of department review costs and benefits per initiative and decide on expansion, adjustment or ending, which applies in particular when the cost per case rises without the benefit per case.

Weighing AI initiatives in the portfolio is covered in Managing AI in the project portfolio.

Niehues Consulting links AI management with the management of cost and benefit

Niehues Consulting supports executive teams and heads of department in building organisations that deliver reliably. Organisation design creates structures and decision paths for day-to-day business and for transformation. In delivery management, a delivery manager or product owner steers the programme until the change has arrived in operations. Where AI enters the workflows, a business case and a prototype are added, followed by introduction in the business units.

The approach follows the Niehues Delivery Model with its phases Understand, Build and Introduce. For an AI initiative it asks three questions. In Understand: which benefit should the initiative deliver and who owns it? In Build: what cost per case does the solution cause and which metrics make it visible? In Introduce: are users enabled and workflows adapted? Founder Marin Niehues, a management consultant from the delivery side, builds organisations that deliver. His projects range from venture-backed scale-ups to DAX 40 groups, with assignments in more than 20 countries on six continents. In the free 30-minute initial consultation you assess together which AI spending in your company is matched by a proven benefit and who should own it: book a consultation.