KPMG finds 49% cut AI agent rollouts when costs outran value

# KPMG finds 49% cut AI agent rollouts when costs outran value Source: [https://ppc.land/kpmg-finds-49-cut-ai-agent-rollouts-when-costs-outran-value/](https://ppc.land/kpmg-finds-49-cut-ai-agent-rollouts-when-costs-outran-value/) *Nearly half of large organisations have narrowed, delayed or paused their AI agent deployments after operating costs began to exceed the value produced, according to survey findings*[*KPMG*](https://ppc.land/?s=KPMG)*published on Wednesday, June 24, 2026\. Only 7 percent of the 2,145 senior leaders polled said their organisation had reached established return on investment\.* The[Global AI Pulse Q2 2026,](https://kpmg.com/xx/en/our-insights/ai-and-technology/ai-pulse.html?ref=ppc.land)released from London, describes a market where adoption is still climbing and the ability to prove a financial return is not\. Planned AI investment held at a weighted average of US$188 million over the next twelve months, effectively flat against the US$186 million recorded in the first quarter\. Confidence rose on every measure the study tracks\. What did not move was the share of organisations able to say the money came back\. The findings resurfaced this month in wider financial commentary\. Reuters Breakingviews columnist Sebastian Pellejero drew on the 7 percent figure in a piece filed from Las Vegas on August 7, 2026, reporting from the Ai4 conference and describing corporate decision\-makers caught between technological progress and demands for financial discipline\. ## Deployment decisions turn selective The most concrete number in the report concerns behaviour rather than sentiment\. Asked whether their organisation had questioned, delayed or scaled back deployment of AI agents because expected costs began to outweigh the value generated, 24 percent of respondents said they had scaled back or narrowed a deployment and 25 percent said they had delayed or paused further rollout\. Those two categories combine into the 49 percent headline\. A further 22 percent said they had questioned the decision without changing course\. Only 24 percent reported that costs and value remained aligned, and 5 percent had not yet deployed agents at all\. According to KPMG, these actions do not indicate falling confidence in the technology\. The report frames them as a growing willingness to distinguish between deployments that create value and deployments that do not, with investment concentrating where expected returns are strongest\. That reading is defensible, but the operational effect on vendors is the same either way\. A narrowed deployment is a smaller contract\. A paused rollout is deferred revenue\. For advertising technology firms selling agentic capability into enterprise marketing departments, the survey quantifies a purchasing pattern that has been visible anecdotally for months\. ## Adoption climbs while returns stay flat The AI maturity curve KPMG uses runs across six phases, from research and development through to established ROI\. The driving\-adoption phase, defined as embedding AI across the organisation, rose from 13 percent in the first quarter to 22 percent in the second, a nine\-point jump described in the report as the largest movement observed anywhere on the curve\. Every other stage either fell or held\. Research and development dropped two points to 9 percent\. Experimentation fell three points to 19 percent\. Strategic planning stayed at 21 percent\. Scaling the technology declined four points to 22 percent\. Established ROI, the terminal phase, slipped one point to 7 percent\. The shape is unusual\. Organisations moved out of the earlier phases and stacked up in the second\-to\-last one, without the final phase absorbing them\. Confidence measures rose in parallel: 76 percent of leaders said AI is currently delivering meaningful business value, up twelve points from the first quarter; 78 percent said they can future\-proof their AI strategy, up eight points; and 79 percent said AI would remain a top investment priority even in a recession, up five points\. **Meaningful business value**and**established ROI**are separate constructs in the survey instrument\. The first captures tangible outcomes such as productivity gains, cost savings, revenue growth and improved decision\-making\. The second requires that those outcomes demonstrably justify the investment\. The distance between 76 percent and 7 percent is the distance between the two definitions\. Simon Benson, Regional AI Lead for ASPAC at KPMG Australia, characterised the shift in terms of what leaders are asking\. "One of the clearest indicators of AI maturity is that the questions change," he said in the report\. "Early adopters ask whether they should move forward\. Leaders ask how to scale\." ## Token economics reaches the board The report attributes much of the cost pressure to**token economics**, which KPMG defines in a footnote as the way AI is priced, consumed and managed at scale, with models charging by units of text, data, reasoning and interaction processed during a task\. Visibility into that spending is thin\. Asked how visible the operating costs of their AI systems are today, 35 percent of respondents said fully visible and actively monitored\. Another 42 percent described costs as somewhat visible, 13 percent said costs became visible only after billing, 8 percent called them largely invisible and 3 percent were unsure\. Roughly two\-thirds of the sample therefore operates without a complete real\-time view of what its AI systems cost to run\. The associated outcome gap is the study's sharpest correlation\. Organisations with full visibility into AI operating costs reported established ROI at 15 percent, against 3 percent for those without it, a fivefold difference\. Controls are being installed unevenly\. Cost review as part of AI approval processes was reported by 54 percent of organisations and AI cost\-monitoring dashboards by 53 percent\. Usage or token budgets reached 40 percent, and architecture or prompt design standards 39 percent\. Five percent had none of these in place\. Access to lower\-cost, high\-fidelity models was the fastest\-rising influence on AI strategy, climbing seven points from 15 percent to 22 percent quarter over quarter\. Energy use, carbon concerns and macroeconomic conditions each rose six points\. Pressure to demonstrate value to investors or the board rose from 19 percent to 24 percent\. Data security, privacy and risk held at 33 percent, still the largest single concern\. Rob Fisher, Global Head of Advisory at KPMG International, framed the exposure in financial terms\. "AI is now as much a financial management priority as it is a technology one," he said\. "The real risk isn't investing in AI but doing so without cost visibility and an understanding of the economics of AI\." ## Accountability separates the results Executive sponsorship is close to universal in the sample\. Around three\-quarters of organisations, 75 percent, said their chief executive actively owns AI as a strategic priority\. Formal accountability is a different matter\. Asked who is ultimately accountable for business decisions informed or executed using AI outputs, 24 percent named the CEO or executive committee and 29 percent named a specific C\-suite executive such as a chief operating officer, chief AI officer or chief information officer\. Business unit or function leaders accounted for 15 percent\. Centralised AI governance or risk committees and shared accountability across multiple roles each took 13 percent\. Three percent said accountability is unclear or not formally defined, 1 percent said AI systems operate autonomously within approved limits, and 2 percent were unsure\. Where the CEO is accountable, outcomes diverge sharply\. Confidence in the ability to future\-proof AI strategy ran at 60 percent against 22 percent\. Reported meaningful business value ran at 57 percent against 21 percent\. Established ROI ran at 14 percent against 4 percent, a difference of more than three times\. Operational governance lags the executive layer\. Across four practices measured in the report, only around a third of organisations described their arrangements as very clear and well managed: 35 percent on accountability for AI data quality and refresh cycles, 35 percent on where humans are expected to override or correct AI outputs, 35 percent on how easily staff can intervene or pause AI\-driven decisions, and 34 percent on understanding the ongoing operating costs of AI once embedded in workflows\. Between a quarter and 28 percent reported unclear arrangements or outright gaps in each category\. Samantha Gloede, Global Head of Risk Services and Global Trusted AI Leader at KPMG International, tied the two layers together in the report, arguing that governance can no longer sit on the sidelines and that the organisations creating the most value are building accountability into how decisions get made\. ## Workforce data cuts against the trend in one market Employee adoption of AI agents rose from 25 percent to 28 percent globally, while reported resistance declined to 14 percent\. The United States moved the other way, with resistance rising from 5 percent to 20 percent\. KPMG notes the contrast without offering an explanation for it\. Elsewhere the workforce indicators point upward\. Seventy\-one percent of organisations reported good progress toward a fully integrated AI\-human workforce, up eleven points from the first quarter, and 78 percent expected AI fluency to become more important, with roles changing for employees who do not develop those capabilities\. Priorities shifted in a matching direction\. Productivity gains fell seven points to 35 percent as a stated AI priority and faster, better decisions fell five points to 36 percent\. Human\-AI collaboration and fluency rose to 30 percent, responsible AI and governance to 28 percent, adaptability and resilience to 20 percent, and ecosystem and partnerships to 16 percent\. Benedikt Höck, Regional AI Lead for EMEA at KPMG in Germany, put the emphasis on work design rather than deployment\. "The next challenge is not deploying AI\. It is redesigning work for a world where humans and AI operate together," he said in the report\. ## Where the source documents diverge Three inconsistencies across the published materials are worth recording\. The press release describes the sample as "more than 2,000\+ business leaders across 20 countries at organizations with annual revenues exceeding US $50\+ million\." The KPMG web page for the report says more than 2,100 senior leaders\. The methodology section of the report itself specifies n=2,145 and sets eligibility at US$50 million or more in revenue for the global sample, with the US tracking sample using US$1 billion or more\. The press release footnotes add a further tier, citing a US$100 million threshold for Canada, the United States, the United Kingdom, China, Germany, India, Japan, Singapore, Korea and Saudi Arabia\. On accountability, the press release states that 24 percent of leaders say the CEO is accountable for AI\-driven business outcomes "while 29 percent point to the broader C\-suite\." The report chart defines that second category more narrowly as a named C\-suite executive accountable for AI outcomes\. On cost literacy, the press release highlights cite 33 percent of leaders reporting limited understanding of usage costs as a deployment challenge for AI agents\. The report body attributes the same one\-third figure to AI cost and economic literacy skills as a challenge, with a separate 29 percent reporting difficulty understanding and controlling operating costs as systems scale\. The press release also introduces a 23 percent figure for leaders struggling with usage\-based costs and a 42 percent figure for partial visibility into AI spending, neither of which appears in the report deck\. The survey was fielded online between April 28 and May 25, 2026, across 20 countries, territories and jurisdictions, covering respondents at managing director level or above, with more than 40 percent drawn from the C\-suite\. The KPMG data lands on a question the advertising industry has been circling for most of 2026: who pays for inference, and how is it billed\. Agency holding companies have started[treating AI tokens as a margin business](https://ppc.land/agencies-turn-ai-tokens-into-a-margin-business-as-agentic-spend-stalls/), negotiating wholesale rates with model providers and reselling capacity to clients at a markup, in a structure that mirrors principal media buying\.[Omnicom](https://ppc.land/?s=Omnicom)chief executive John Wren told analysts on July 29, 2026 that the market has not yet seen what AI actually costs, with the group reporting third\-party service costs of almost $2\.9 billion in the first half of 2026 against $1\.7 billion a year earlier\. PMG has deployed a tool across its business under a[$50\-a\-day token cap per user](https://ppc.land/omnicoms-phd-gains-just-10-20-productivity-from-ai-in-apac/)\. Engineering teams are attacking the same bill from the architecture side\. A Dutch agency published a diagram in August arguing that agentic media buying inherited the wrong chain of command,[collapsing twelve separate protocol calls into a single buyer agent](https://ppc.land/draft-digital-cuts-12-mcp-calls-to-one-buyer-agent-to-stop-token-burn/)to hold context load down\. The revenue side of that equation remains modest: Magnite chief executive Michael Barrett[placed 2027 protocol\-based advertising spend somewhere below $700 million](https://ppc.land/magnite-ceo-caps-2027-agentic-ad-spend-near-700m-as-ai-cuts-contractor-tasks/), a figure he characterised as small against total programmatic volume\. Buy\-side research has been converging on the same ratio KPMG measured\.[TransUnion](https://ppc.land/?s=TransUnion)research published on August 5, 2026 found[only 53 percent of senior US marketing leaders reporting meaningful ROI from AI](https://ppc.land/only-53-of-marketers-get-meaningful-roi-from-ai-transunion-finds/)while 89 percent expect to increase AI spending over the next 12 to 24 months\.[Gartner](https://ppc.land/?s=Gartner)forecast in June 2025 that[more than 40 percent of agentic AI projects would be cancelled by the end of 2027](https://ppc.land/nearly-half-of-agentic-ai-projects-may-fail-by-2027-warns-research-firm/), naming escalating costs and unclear business value among the causes\. Typeface data published in June 2026 found[AI had made campaign cycles slower rather than faster](https://ppc.land/ai-made-campaigns-slower-not-faster-new-typeface-data-shows/)for marketing operations teams\. The governance half of the picture has been measured too\. DigiCert research published on July 7, 2026 found[78 percent of organisations had experienced an AI\-related security incident or vulnerability](https://ppc.land/digicert-78-of-firms-face-ai-incidents-half-lack-governance/)while roughly half lacked formal governance\. Sector\-level returns vary: Snowflake data published in March 2026 showed[advertising and media firms outperforming every other industry on generative AI returns](https://ppc.land/ad-and-media-firms-beat-every-industry-on-gen-ai-roi-snowflake-data-shows/), which suggests the 7 percent global figure understates the position of some marketing organisations while flattering others\. At the Ai4 conference in Las Vegas, held from August 4 to August 6, 2026 and drawing more than 12,000 attendees from over 90 countries according to organisers, software group Dataiku described an audit that found more than 400 self\-directed AI agents inside a client organisation whose chief information officer had expected roughly 40, according to the Reuters Breakingviews account\. That gap between assumed and actual agent counts is the operational version of the visibility problem KPMG measured\. Where budget owners cannot see how many agents are running, the token bill arrives as a surprise\. Steve Chase, Global Head of AI and Digital Innovation at KPMG International, summarised the study's central division\. "We're seeing a clear divide between organizations with leadership accountability at the top and those without," he said\. "These companies are seeing materially better results across the board\." ## Timeline - June 25, 2025 \- Gartner forecasts that[more than 40 percent of agentic AI projects will be cancelled by the end of 2027](https://ppc.land/nearly-half-of-agentic-ai-projects-may-fail-by-2027-warns-research-firm/)on cost and unclear value grounds - March 10, 2026 \- Snowflake publishes data showing[advertising and media firms leading all industries on generative AI returns](https://ppc.land/ad-and-media-firms-beat-every-industry-on-gen-ai-roi-snowflake-data-shows/) - April 2026 \- KPMG publishes Global AI Pulse Q1 2026, based on 2,110 respondents - April 28 to May 25, 2026 \- Global AI Pulse Q2 2026 fieldwork runs online across 20 countries - June 23, 2026 \- Typeface data indicates[AI has made marketing campaign cycles slower rather than faster](https://ppc.land/ai-made-campaigns-slower-not-faster-new-typeface-data-shows/) - June 24, 2026 \- KPMG publishes Global AI Pulse Q2 2026 from London - July 7, 2026 \- DigiCert reports that[78 percent of firms have faced an AI\-related incident while half lack governance](https://ppc.land/digicert-78-of-firms-face-ai-incidents-half-lack-governance/) - July 29, 2026 \- Omnicom chief executive John Wren tells analysts[the market has not yet seen what AI actually costs](https://ppc.land/agencies-turn-ai-tokens-into-a-margin-business-as-agentic-spend-stalls/) - August 2, 2026 \- Magnite chief executive Michael Barrett[caps 2027 agentic ad spend expectations near $700 million](https://ppc.land/magnite-ceo-caps-2027-agentic-ad-spend-near-700m-as-ai-cuts-contractor-tasks/) - August 3, 2026 \- Reporting documents[agency holding companies turning AI tokens into a margin business](https://ppc.land/agencies-turn-ai-tokens-into-a-margin-business-as-agentic-spend-stalls/) - August 4 to August 6, 2026 \- Ai4 conference runs in Las Vegas with more than 12,000 attendees - August 5, 2026 \- TransUnion publishes research finding[53 percent of US marketing leaders report meaningful AI ROI](https://ppc.land/only-53-of-marketers-get-meaningful-roi-from-ai-transunion-finds/) - August 7, 2026 \- Reuters Breakingviews cites the KPMG 7 percent ROI figure in a column filed from Las Vegas - [Only 53% of marketers get meaningful ROI from AI, TransUnion finds](https://ppc.land/only-53-of-marketers-get-meaningful-roi-from-ai-transunion-finds/)\- Survey of 100 senior US marketing leaders documenting an AI confidence and readiness gap alongside rising budgets\. - [Agencies turn AI tokens into a margin business as agentic spend stalls](https://ppc.land/agencies-turn-ai-tokens-into-a-margin-business-as-agentic-spend-stalls/)\- How holding companies are pricing model capacity to clients while nobody has settled who pays for compute\. - [Omnicom's PHD gains just 10\-20% productivity from AI in APAC](https://ppc.land/omnicoms-phd-gains-just-10-20-productivity-from-ai-in-apac/)\- Agency\-level productivity figures and the token caps being applied to control per\-user spending\. - [Draft Digital cuts 12 MCP calls to one buyer agent to stop token burn](https://ppc.land/draft-digital-cuts-12-mcp-calls-to-one-buyer-agent-to-stop-token-burn/)\- An architecture argument for collapsing protocol calls to reduce duplicated context in agentic media buying\. - [Magnite CEO caps 2027 agentic ad spend near $700m as AI cuts contractor tasks](https://ppc.land/magnite-ceo-caps-2027-agentic-ad-spend-near-700m-as-ai-cuts-contractor-tasks/)\- Sell\-side forecasting on when protocol\-based buying reaches meaningful scale\. - [Nearly half of agentic AI projects may fail by 2027, warns research firm](https://ppc.land/nearly-half-of-agentic-ai-projects-may-fail-by-2027-warns-research-firm/)\- Gartner's cancellation forecast and the cost and governance factors behind it\. - [DigiCert: 78% of firms face AI incidents, half lack governance](https://ppc.land/digicert-78-of-firms-face-ai-incidents-half-lack-governance/)\- Survey of 1,001 IT and security decision makers quantifying the oversight gap in enterprise AI\. - [AI made campaigns slower, not faster, new Typeface data shows](https://ppc.land/ai-made-campaigns-slower-not-faster-new-typeface-data-shows/)\- Practitioner\-side evidence that AI tooling has added operational friction to marketing workflows\. - [Ad and media firms beat every industry on gen AI ROI, Snowflake data shows](https://ppc.land/ad-and-media-firms-beat-every-industry-on-gen-ai-roi-snowflake-data-shows/)\- Sector comparison placing advertising ahead of other industries on reported generative AI returns\. - [Google Cloud faces 48% revenue reliance on OpenAI and Anthropic in 2027](https://ppc.land/google-cloud-faces-48-revenue-reliance-on-openai-and-anthropic-in-2027/)\- The financing structure underneath the infrastructure that enterprise AI bills run through\. ## Summary **Who:**KPMG International surveyed 2,145 C\-suite and senior business leaders with direct knowledge of AI use at their organisations\. Findings were presented by Steve Chase, Global Head of AI and Digital Innovation, and Rob Fisher, Global Head of Advisory, with commentary from regional AI leads Simon Benson, Benedikt Höck and Priya Emmanuel, and from Samantha Gloede, Global Head of Risk Services\. **What:**The Global AI Pulse Q2 2026 found 49 percent of organisations had scaled back, narrowed, delayed or paused AI agent deployments after expected costs began to outweigh anticipated value, while only 7 percent reported established ROI\. Planned AI spending held at a weighted average of US$188 million over twelve months\. Organisations with full visibility into AI operating costs reported established ROI at 15 percent against 3 percent for those without\. **When:**Fieldwork ran online from April 28 to May 25, 2026\. The report and accompanying press release were published on June 24, 2026\. Reuters Breakingviews cited the 7 percent ROI figure on August 7, 2026 in coverage of the Ai4 conference, held in Las Vegas from August 4 to August 6, 2026\. **Where:**The survey covered 20 countries, territories and jurisdictions: Australia, Brazil, Canada, China including Hong Kong SAR and Taiwan, France, Germany, India, Ireland, Italy, Japan, Korea, Mexico, Netherlands, Saudi Arabia, Singapore, South Africa, Spain, Switzerland, the United Kingdom and the United States\. The press release was issued from London\. **Why:**The findings matter to marketing and advertising organisations because they put a number on a purchasing pattern that has been reshaping agentic ad tech procurement through 2026\. Enterprise buyers are not abandoning AI, with 79 percent saying it would remain a priority even in a recession, but they are narrowing deployments where the operating cost cannot be traced or justified\. For vendors selling agent\-based campaign management, retail media assistants and generative creative tooling, cost transparency and demonstrable return have become gating conditions rather than differentiators\. For agencies now reselling model capacity, the survey indicates that the client\-side scrutiny of those charges is already under way\.