AI ROI in 2026: How Enterprises Measure Returns on AI Investments

By Elton Chan, Co-Founder October 7, 2026 10 min read

How Enterprises Are Measuring ROI on AI. How Enterprises Are Measuring ROI on AI Investments in 2026, by Second Talent.

TL;DR: Most large companies now track AI ROI, but few can show it: 72% of big US firms formally measure it, yet only 37% of companies in McKinsey's 2026 survey see any EBIT impact from AI, and 56% of CEOs told PwC they got neither higher revenue nor lower costs. A typical AI use case takes two to four years to pay back, against 7 to 12 months for other technology, Deloitte found.

In a 2025 randomized trial by the research group METR, experienced developers took 19% longer to finish real tasks when allowed to use AI tools. Afterwards they estimated AI had made them 20% faster.

That gap between felt and measured gains runs through enterprise AI ROI. Surveys that ask whether AI is paying off get a yes from most executives, and surveys that ask for the profit figure mostly do not.

Four numbers that frame AI ROI

1Only 7% of 2,145 leaders in KPMG's mid-2026 AI Pulse said they had established ROI on AI.
2Early adopters that put a number on it report a 49% return, up from 41% a year earlier, Snowflake's surveys show.
3Employee productivity and usage is the most common ROI metric, tracked by 47% of large US firms that measure gen AI returns.
4Gartner expects over 40% of agentic AI projects to be canceled by the end of 2027 over costs, unclear value or weak risk controls.

What ROI Are Enterprises Getting From AI?

About 6% of companies in McKinsey's survey get large, proven returns. That is McKinsey's count of AI "high performers", firms that attribute at least 5% of EBIT to AI and report significant value, and the share has not moved since 2025.

The State of AI 2026 survey of 1,719 respondents found most of the rest either see no EBIT effect or one below 5%.

CEOs give a similar answer. In PwC's 2026 Global CEO Survey, 30% reported higher revenue from AI over the past year and 26% lower costs, with 12% getting both.

Waffle chart of what CEOs got from AI in the past year, PwC 2026 Global CEO Survey of 4,454 CEOs: 18% higher revenue only, 12% both higher revenue and lower costs, 14% lower costs only, 56% neither.

BCG's split is the starkest. In its survey of 1,250 companies, 60% reported minimal revenue and cost gains from AI, while the 5% it calls "future-built" posted 1.7 times the revenue growth and 1.6 times the EBIT margin of the laggards.

Why the surveys disagree

Ask whether AI pays off and most executives say yes. Ask whether it moved profit and far fewer do. The share reporting returns ranges from 92% to 7% depending on who is asked and how "return" is defined.

Bar chart of the share reporting AI returns by survey: Snowflake early adopters with a positive return 92%, Wharton large US firms with positive ROI 74%, McKinsey any EBIT impact 37%, PwC CEOs seeing higher revenue 30%, IBM AI initiatives meeting expected ROI 25%, Deloitte significant ROI from gen AI 15%, KPMG leaders who have established ROI 7%.

Snowflake's 92% comes from 2,050 early adopters, companies already committed to AI, asked whether their investment shows a positive return. KPMG's 7% asks whether ROI has been established.

The two figures measure different things and should never sit side by side as if they were one trend.

The most quoted figure of all, MIT NANDA's "95% of organizations are getting zero return", needs the same care.

The report drew on 52 interviews, 153 conference survey responses and a review of 300 public projects, and its authors call the numbers "directionally accurate" rather than company reporting.

How Do Enterprises Measure AI ROI?

Mostly through productivity, with profit a close second. The Wharton School and GBK Collective's 2025 survey of 801 US firms with 1,000+ staff found firms that measure gen AI ROI use 3.6 metrics on average.

HR (84%) and finance (80%) teams were the most likely to track ROI formally.

Dot plot of metrics large US firms use to measure gen AI ROI, Wharton and GBK 2025: employee productivity and usage 47%, profit or loss from gen AI 46%, performance change after training 42%, throughput and efficiency 42%, time to productivity for new hires 37%, spend tied to business KPIs 36%.

Finance chiefs lean the same way. In a Gartner survey of 204 finance leaders in March 2026, 45% said their AI spending leans toward productivity and 20% toward better decisions.

Some finance teams are approaching that integration by testing Claude for financial reporting directly, using it to fold AI usage metrics into the reports they already produce each month.

The trouble is that productivity rarely shows up in the accounts. Deloitte's State of AI in the Enterprise 2026, a survey of 3,235 leaders, shows how far the soft gains outrun the hard ones.

Operational gains reported

66% productivity and efficiency
53% better insights and decisions

Financial gains reported

40% cost reduction
20% revenue increase

Share of leaders reporting each benefit from AI. Source: Deloitte, State of AI in the Enterprise 2026 (n=3,235).

The basic AI ROI formula

AI ROI is the value an AI project creates, minus its full cost, divided by that cost. Snowflake's 2025 respondents who quantified returns got back $1.41 for every $1 spent, a 41% ROI.

Presenting these figures visually using a chart maker can make it easier for a board to compare projects.

The formula is simple. The hard parts are the inputs: which gains count, which costs belong to the project, and what would have happened without AI.

For a more granular view of AI returns, Larridin connects AI spend and adoption to the work delivered across departments.

Its measurement platform tracks capacity added, cost per AI hour, engineering output, code quality, delivery metrics, and workflow automation opportunities.

Frameworks in use

Forrester Total Economic Impact (TEI): scores cost, benefits, flexibility and risk. Most published TEI studies are commissioned by the vendor being assessed, and Forrester marks them that way.
Gartner's three outcomes: Gartner pairs classic ROI with return on employee and return on the future, so a project that builds skills or options is not judged on cost savings alone.
Levelized cost of AI: total capital and operating spend per unit of useful AI output, modeled on the energy sector's levelized cost of electricity. It comes from a 2025 paper by Eliseo Curcio and is not yet a common enterprise standard.

How Long Does AI Take to Pay Back?

Longer than most technology. Only 6% of the 1,854 executives in Deloitte's AI ROI study saw payback in under a year. Even among their most successful AI projects, 13% returned the investment within 12 months.

Range chart of typical payback periods from a Deloitte survey of 1,854 executives: a typical technology investment pays back in 7 to 12 months, a typical AI use case in 24 to 48 months.

Newer forms of AI are further back. Deloitte found 15% getting significant, measurable ROI from generative AI today and 38% expecting it within a year.

For agentic AI, 10% see significant returns now, and most expect them in three to five years.

US executives are more patient than their results. Four in five in the Wharton survey expect positive ROI within two to three years, and 88% expect to spend more on AI.

What Controlled Studies Show About AI Returns

Randomized and field studies are the closest thing to a measured return, and they show gains that are real but narrower than survey answers suggest. The size depends on the task and on who does it.

Customer support5,179 agents

+14%

issues resolved per hour; +34% for novice agents

BCG consultants758 people

+12.2%

more tasks, 25.1% faster; 19% less likely correct on tasks beyond AI's reach

Senior developers246 tasks

19% slower

with AI tools, on their own codebases

Danish workers25,000 people

2.8%

of work hours saved, with no measurable effect on pay or hours

Sources: Brynjolfsson, Li and Raymond (QJE 2025); Dell'Acqua et al. (Organization Science 2026); METR (July 2025); Humlum and Vestergaard (NBER 2025).

The customer support study found the biggest gains among the least experienced staff and almost none among the best.

The METR trial ran the other way: developers who already knew their codebases well were slowed down, in part by time spent reviewing AI output.

Both results matter for ROI, because the same tool can pay back on one team and cost money on another.

Workplace-wide effects are small so far. A randomized rollout of Microsoft 365 Copilot across 66 firms and 7,137 workers cut about two hours a week from email but did not change how much work got done.

The St. Louis Fed estimates generative AI users save 5.4% of their work hours, or 1.4% of all hours across the workforce. For developer tools specifically, see our AI coding assistant statistics.

Why the Cost Side of AI ROI Is Hard to Measure

More than 90% of CIOs told Gartner that managing cost limits the value they get from AI. Gartner warns that a CIO who does not understand how generative AI costs scale "could make a 500%-1,000% error" in cost calculations.

Usage-priced models are the reason: the bill rises with every user and query, and Gartner put some generative AI deployment approaches at $5 million to $20 million.

Seeing those costs is what separates firms that can prove ROI from those that cannot. In KPMG's Q2 2026 AI Pulse, leaders with full visibility of their AI costs were five times as likely to have established ROI.

Grouped column chart from KPMG's Q2 2026 Global AI Pulse: leaders with full AI cost visibility 15% have established ROI against 3% without it; with the CEO accountable for AI 14% against 4% without.

Tracking is catching up. The FinOps Foundation found 98% of cloud cost teams now manage AI spend, up from 63% in 2025 and 31% in 2024. Their top tooling request is granular AI cost monitoring by tokens, requests and GPU hours.

KPMG also found 49% of organizations had delayed or scaled back AI agent deployments when costs started to outweigh benefits.

Measuring ROI on AI Agents

Agents are where expectations run furthest ahead of results. In BCG's AI Radar 2026, 90% of CEOs expected agents to deliver measurable ROI this year, and 94% said they would keep investing even if the investments do not pay off.

17% to 29%

agents' share of total AI value, 2025 to 2028 (BCG)

47%

return executives expect from agents over 12 months (Snowflake)

88%

of agentic early adopters report ROI (Google Cloud)

Sources: BCG, The Widening AI Value Gap (Sep 2025); Snowflake, The ROI of Gen AI and Agents 2026; Google Cloud, The ROI of AI 2025.

Agents also need different metrics from chat tools: tasks completed end to end, how often a human has to step in, and the cost per completed task rather than per query. Our AI agents statistics track adoption and spending.

What Firms That Prove AI ROI Do Differently

They set up measurement before they scale. McKinsey's high performers are twice as likely as other firms to have defined processes for measuring AI's impact.

KPMG found leaders whose CEO is accountable for AI outcomes were three and a half times as likely to have established ROI, 14% against 4%.

They also aim at growth, not only savings. Gartner's CFO survey found finance functions that funded AI for new products, markets or business models were more than twice as likely to report high realized value.

For how adoption and returns vary by sector, see AI adoption by industry and our enterprise AI adoption statistics.

Building a Team That Can Deliver AI ROI

Measuring AI ROI depends on the people who wire usage, cost and outcome data together. Second Talent matches companies with vetted AI engineers from Asia, data engineers and agentic AI specialists at 50 to 70% below US hiring costs.

Tell us the role you need to fill and we will come back with candidates.

Frequently Asked Questions

What is a good ROI on AI?

Early adopters that quantified returns reported 49% in Snowflake's 2026 survey, meaning about $1.49 back for every $1 spent. McKinsey sets a higher bar for its high performers: at least 5% of company EBIT attributable to AI.

Why do so many AI projects fail to show ROI?

Gartner's reasons for abandoned generative AI projects are poor data quality, weak risk controls, rising costs and unclear business value.

In infrastructure and operations teams, Gartner found only 28% of AI use cases fully succeed and meet ROI expectations, while 20% fail outright.

Which AI benefits are easiest to measure?

Cost reduction in defined processes, such as support tickets resolved or invoices processed per hour, because there is a clear before and after.

Time savings for knowledge workers are harder, since saved time only becomes money if it goes into other work.

Get developers who ship with AI agents.

Starts at $23/Hour.

Start Hiring

$0 cost until you hire

Written by

Elton Chan

Elton Chan is the Co-Founder of Second Talent, a solution that connects global tech leaders with top-tier tech talent across Asia. He specializes in talent solutions and has led Second Talent’s rapid growth since 2024, helping scale its network to over 100,000 pre-vetted developers and earning industry recognition as the #1 in the Global Hiring category on G2. A long-time entrepreneur with deep roots in digital transformation, Elton previously co-founded Branch8, a Y Combinator–backed e-commerce technology firm, and served as the Founding Chairman of HKEBA, a leading Asia-focused business association driving innovation, digital education, and cross-border collaboration. His work bridges technology, talent, and business strategy to shape how companies scale in an increasingly remote and digital world.