‍The headline return figures from KPMG and IDC are real — but they describe a narrow group of organisations that deployed AI with discipline. Here is what the full data reveals, and what it takes to be in that group.

Published: March 2026   |   Category: AI & Finance strategy   |   Reading Time: ~9 minutes

When KPMG published its analysis of agentic AI returns in mid-2025, the number that travelled furthest was this one: on average, companies earn $3.50 for every dollar they invest in agentic AI. The top 5% of organisations globally earn $8 per dollar. IDC's parallel research confirmed that organisations can achieve a 2.3x return on agentic AI investments within thirteen months.

These figures are not projections. They are derived from documented outcomes across thousands of enterprise deployments. And yet, for most CFOs attempting to build a credible internal business case for agentic AI, they feel frustratingly out of reach. The median reported return on AI investment in finance remains a modest 10%. Only one in ten organisations using agentic AI today says it is generating significant returns right now. Just 13% of even the most successful AI projects deliver payback within twelve months.

Both sets of numbers are true simultaneously. The resolution lies not in the technology itself but in how it is deployed, governed, and measured. This article examines the full picture of agentic AI ROI in finance — the data behind the headline figures, the specific use cases generating returns, the performance gap between leaders and laggards, and the execution characteristics that separate the two.

"We've reached a tipping point where boards and CEOs are done with AI experiments and expecting real results. AI for AI's sake is a waste." — Jason Kurtz, CEO, Basware

1. The ROI landscape: what the data actually shows

Understanding the true financial picture of agentic AI in finance requires holding multiple data sources in view at once rather than relying on any single headline figure. The studies are consistent in their broad direction — returns are real, achievable, and growing — but they diverge significantly on how quickly and reliably those returns materialise.

$3.50 average return per $1 invested in agentic AI, with top 5% earning $8 per $1 (KPMG, 2025)

2.3x average ROI on agentic AI investments within 13 months (IDC, 2025)

$3T in projected global corporate productivity gains from agentic AI, with 5.4% EBITDA improvement annually (KPMG)

The divergence between the KPMG headline and the Deloitte ground-level survey data is not a contradiction. It reflects the reality that agentic AI ROI is not an average experience — it is a performance distribution. At one end sit the frontier organisations: those that have moved beyond pilots, redesigned core workflows, invested in data infrastructure, and built governance frameworks that support speed without sacrificing reliability. These organisations are generating the returns that drive the headline figures.

At the other end are organisations that have deployed AI tools into unchanged processes, treated implementation as an IT project rather than a business transformation, and measured outcomes — if at all — against activity metrics rather than financial ones. IDC's findings are unambiguous on this divide: frontier firms leading in AI adoption achieve returns of 2.84x on their investments, compared to just 0.84x for laggards. The gap between leading well and lagging is not incremental. It is the difference between value creation and value destruction.

ROI spectrum: agentic AI returns by adoption maturity

Top 5% globally

$8.00 / $1

Frontier firms (IDC)

2.84x

Average (KPMG)

$3.50 / $1

Average (IDC, 13 months)

2.3x

Finance median

~10%

Laggards (IDC)

0.84x

2. Where finance teams are generating the highest returns

The performance gap between leaders and laggards is not randomly distributed. It concentrates in specific finance functions where agentic AI creates the greatest structural advantage: high-volume, rules-proximate processes that have historically absorbed significant manual effort with limited strategic value creation.

Finance use caseWhat agentic AI doesDocumented returnAccounts payableAutomates invoice capture, validation, matching against POs and contracts, duplicate detection, and payment booking80% average ROI vs. 67% for general AI; 60% reduction in exception-handling time; 72% of finance leaders identify AP as the primary agentic deployment targetFinancial close and consolidationMonth-end close agents identify anomalies, prepare journal entries, cross-reference invoices, and flag discrepancies autonomously30 to 50% reduction in close-related manual workload; pathway to zero-day close; 60 to 80% increase in forecast accuracy in full agentic deployment scenariosCredit risk and underwritingAI agents redesign credit risk memo creation, automate data gathering, and accelerate underwriting decisionsUS bank case study: 20 to 60% productivity improvement; 30% improvement in credit turnaround time (McKinsey, 2025)FP&A and forecastingAgents forecast in real time, optimise cash allocation, model scenarios, and surface variance explanations without analyst interventionConsumer goods case study: 50% reduction in report generation time; dynamic forecasting replacing static quarterly cyclesCompliance and fraud detectionContinuous monitoring, automated surveillance, policy updates, and anomaly detection across transaction flows35 to 45% cost savings in compliance operations; 2 to 4x improvement in fraud detection accuracy; 60% reduction in false positivesWealth management operationsAutomates prospecting, KYC and AML onboarding, portfolio rebalancing, tax optimisation, and client reporting40 to 50% reduction in advisor prospecting time; 30 to 40% increase in net new AUM; 50% acceleration in client onboarding; 40 to 50% reduction in portfolio management operational costs

Two observations follow from this data. First, the highest-ROI use cases share a common characteristic: they are not AI tools added to existing processes. They are autonomous agents that operate within redefined processes — handling the entire workflow, not just individual steps within it. Second, the returns compound. An accounts payable agent that reduces exception-handling time by 60% simultaneously improves cash conversion cycle performance, strengthens audit trail integrity, and frees controllers to focus on strategic cash management. The value does not sit in a single metric.

3. The performance divide: what separates the $8 earners from the 10% median

The most important question in agentic AI ROI is not which technology to deploy. It is what organisational characteristics determine whether that technology generates frontier returns or median ones. The research is specific on this point.

• Direction before deploymentArtificial Intelligence News research found that 71% of finance teams with weak agentic AI returns had acted under pressure without clear direction, compared to only 13% of teams achieving strong ROI. The sequence matters more than most organisations acknowledge: strategic intent must precede technology selection, not follow it. Teams that deployed AI in response to competitive pressure, without first defining the specific financial outcomes they were targeting, consistently underperformed against those that began with outcome definition.
• Workflow redesign rather than overlayKPMG's analysis of its own client deployments and broader market data is consistent on this point: overlaying agentic AI on existing processes produces incremental efficiency. Redesigning those processes around agentic capability produces transformational returns. The distinction is not subtle — it is the primary variable separating the ROI distribution. Finance teams achieving 20x productivity gains are those where human workers have shifted from executing processes to designing and supervising the systems that execute them.
• Data quality as a precondition, not a follow-onIDC data shows that 48% of organisations cite data governance concerns as their primary implementation challenge, and 20% acknowledge their own data is not yet ready for agentic deployment. These organisations are attempting to extract value from a system whose output quality is a direct function of its input quality. Basware's accounts payable platform, which trains on a dataset of over two billion processed invoices, illustrates what data-ready deployment looks like. The system differentiates between legitimate anomalies and errors without human oversight precisely because the data foundation is sufficient. Without it, no governance framework compensates.
• Measurement discipline established at the outsetKPMG identifies measurement as a core differentiator between AI ROI leaders and followers: specifically, the practice of tying AI initiatives directly to financial outcomes before deployment begins, not after. Finance teams that defined their ROI metrics upfront — cost reduction, forecasting accuracy, cycle time — were consistently better positioned to demonstrate returns, attract continued investment, and iterate toward higher-value use cases. Those that measured activity (tools deployed, queries processed) rather than outcomes found it difficult to build a credible return story even when real value was being generated.
• C-suite ownership of the AI agendaAmong organisations that Deloitte categorises as AI ROI leaders, 95% allocate more than 10% of their technology budget to AI. These organisations treat AI as an enterprise transformation, not a technology upgrade. That framing comes from the top. Deloitte found that organisations increasingly led by CEOs or CFOs on AI adoption consistently outperformed those where the AI agenda sat with technology leadership alone. The accountability structure determines the ambition level — and the ambition level determines the returns.

4. The honest picture: why most finance teams are not there yet

Important context

Despite the headline ROI figures, Deloitte's survey of 1,854 executives found that most organisations achieve satisfactory returns on AI within two to four years — significantly longer than the seven-to-twelve-month payback period typical of conventional technology investments. Only 6% of organisations see payback in under a year. Even among the most successful implementations, just 13% deliver returns within twelve months.

More significantly, 42% of companies abandoned most of their AI projects in 2025. Among the organisations already using agentic AI, just 10% report currently realising significant ROI. These are not numbers to dismiss — they are the honest baseline against which the frontier figures should be measured.

The implication is not that agentic AI does not work. It is that the gap between deploying agentic AI and earning compelling returns from it is larger than most technology vendors acknowledge, and closing that gap requires organisational transformation that most teams are still in the early stages of executing.

The Deloitte data also surfaces a structural misalignment that underlies much of the ROI shortfall: AI rarely delivers value in isolation. It is typically introduced alongside parallel efforts — improving data quality, reconfiguring teams, streamlining operations — that make it difficult to isolate AI's specific contribution. Finance leaders attempting to attribute returns to agentic AI in environments where multiple transformation initiatives are running simultaneously face a genuine measurement challenge. The solution is not to abandon measurement but to design the measurement framework in advance, before the change programmes begin, so that attribution is defensible from the start.

5. The compounding effect: why the 13-month figure understates long-term value

The IDC 2.3x return in 13 months figure is significant. What it does not capture is what happens after month 13. KPMG's analysis of agentic AI systems describes a productivity structure that compounds with scale in ways that conventional automation does not.

A single agent operating autonomously on a defined task delivers measurable efficiency gains. Multiple agents operating simultaneously on related tasks, coordinated by an orchestrator agent, can deliver nine times more work per 24-hour period than a human team covering the same workflows in standard working hours. KPMG's analysis of full-scale agentic finance deployment scenarios — where the finance function operates as a digital command centre overseeing networks of autonomous agents — projects 40 to 60% reductions in finance costs, 30 to 50% reductions in cash conversion cycles, 70 to 90% reductions in decision latency, and 60 to 80% increases in forecast accuracy.

These are not linear returns. They are the product of compounding: each additional agent, each additional use case, each additional layer of automation adds value on top of the existing base, rather than beside it. Finance teams that understand this dynamic invest early and scale aggressively, accepting that the returns in the first year are a fraction of what the system will generate by year three.

"AI's ability to automate tasks is doubling every three to seven months. As of mid-2025, the acceleration in agent capability is outpacing the pace of enterprise deployment across most industries." — KPMG, 2025

6. A practical framework for measuring agentic AI ROI in finance

Finance leaders building internal business cases for agentic AI deployment face a specific challenge: the metrics that matter most in AI ROI are not always the metrics that existing financial tracking systems capture. A practical measurement framework needs to encompass four dimensions simultaneously.

CFO measurement framework: four dimensions of agentic AI ROI

Efficiency gains: Reduction in hours, headcount requirements, or cycle time for target processes. This is the most straightforward dimension — measurable in FTE reduction, overtime elimination, or close cycle compression. Track at the process level, not the department level, to isolate AI's specific contribution.

Revenue enablement: Capacity created for higher-value activities as routine tasks are automated. When FP&A analysts are freed from manual data assembly, what do they do with that time? If it goes to strategic scenario modelling that improves capital allocation decisions, the revenue value of that redeployed capacity can dwarf the direct efficiency saving.

Risk mitigation: Reduction in compliance failures, audit exceptions, fraud losses, and error correction costs. These are real financial values, often invisible in traditional ROI calculations. A 60% reduction in compliance false positives has a dollar value — in analyst hours recaptured and in regulatory exposure reduced.

Business agility: The ability to respond to market changes, regulatory updates, or operational shifts faster than the competition. This is the hardest dimension to quantify but, according to KPMG, the most durable source of long-term competitive advantage from agentic deployment.

Frequently asked questions: agentic AI ROI in finance

What is the average ROI from agentic AI for finance teams?

KPMG reports that on average, companies earn $3.50 for every $1 invested in agentic AI, with the top 5% globally earning $8 per dollar. IDC reports a 2.3x average return within 13 months. However, the median reported return in finance is a more modest 10%, reflecting the large gap between organisations that have deployed agentic AI with strategic discipline and those still running early-stage experiments without clear outcome frameworks.

How long does it take to see ROI from agentic AI in finance?

IDC's research points to an average payback of 2.3x within 13 months for organisations that deploy agentic AI with the right foundational conditions. However, Deloitte's survey of 1,854 executives found that most organisations take two to four years to achieve satisfactory ROI on a typical AI use case — significantly longer than the seven-to-twelve-month payback expected from conventional technology investments. The timeline shortens dramatically when data quality is high, processes are redesigned around agentic capability, and measurement frameworks are defined before deployment begins.

Which finance processes deliver the best agentic AI ROI?

Accounts payable automation consistently leads as the highest-confidence starting point, with 72% of finance leaders citing it as the obvious first use case. Credit risk memo production, financial close automation, FP&A forecasting, and compliance monitoring all demonstrate documented high returns. The common characteristic of high-ROI use cases is that they involve structured data, high transaction volume, and clear rules-based decision criteria — conditions under which autonomous agents operate with the highest reliability and lowest governance risk.

Why do so many finance AI projects fail to generate expected returns?

The primary failure modes are consistent across multiple independent studies: AI deployed into unchanged processes without workflow redesign; insufficient data quality creating unreliable agent outputs; measurement frameworks defined retrospectively rather than upfront; and deployment driven by competitive pressure rather than specific financial outcome targets. Deloitte found that 71% of finance teams with weak AI returns had acted without clear strategic direction. The technology is not the constraint — the organisational preparation around it is.

How does agentic AI ROI differ from generative AI ROI in finance?

Generative AI typically delivers value within specific, bounded tasks — drafting documents, summarising data, answering queries. Its ROI is measurable but relatively contained. Agentic AI operates across entire workflows, executing multi-step processes autonomously without continuous human instruction. This creates a fundamentally different ROI profile: the value compounds as agents coordinate, scale, and operate around the clock. KPMG's full-scale agentic deployment scenarios project 20x productivity gains relative to baseline — a ceiling that generative AI alone cannot approach.

Closing perspective

The $3.50 return figure and the 10% median are both accurate. They describe the same technology deployed under very different conditions. The finance teams earning $8 per dollar did not stumble into those returns. They made a series of deliberate decisions: about which processes to target first, how to redesign those processes around autonomous agents, what data infrastructure was required, and how to measure outcomes against financial metrics that mattered to the board.

The organisations that have not yet reached those returns are, in most cases, not facing a technology problem. They are facing an execution and governance challenge that is within their control to address. McKinsey's analysis of the first-mover advantage in agentic finance is explicit: pioneers are set to gain a 4% return on tangible equity advantage over slow movers, while laggards risk being left with an uncompetitive cost base. That gap, once established, is structurally difficult to close.

The window to be on the right side of that divide is present now. The conditions for agentic AI ROI in finance are well understood, the use cases are proven, and the measurement frameworks exist. What remains is the organisational will to move from experimentation to transformation — and to hold that programme to the same financial accountability standards applied to every other strategic investment.

Sources: KPMG — The Agentic AI Advantage (June 2025) · KPMG — Q4 AI Pulse Survey (January 2026) · KPMG — Agentic AI in Wealth Management (November 2025) · IDC — AI ROI and Agentic Investment Returns (2025) · Deloitte — AI ROI: the paradox of rising investment and elusive returns (October 2025) · McKinsey — Agentic AI in Financial Services (2025) · Neurons Lab — Agentic AI in Financial Services Research Roundup (2026) · Artificial Intelligence News — Agentic AI drives finance ROI in accounts payable automation (February 2026) · Basware / FT Longitude — CFO AI Adoption Survey (2025) · Financial Executives International (FEI) — AI and Agentic AI in Finance: from hype to high-impact execution (December 2025) · Wolters Kluwer — Finance AI Adoption Survey (2025)

‍