AI agent ROI 2026:
all the numbers in one place

Deloitte says 171% average ROI. McKinsey says impact is flat. PwC says 56% of CEOs can't see any return. All three are right — this page puts every grounded 2026 number on one page and explains why they don't contradict.

The headline numbers

Deloitte reports an average enterprise AI ROI of 171%, with US companies achieving 192% — roughly three times the return on traditional automation. The same research carries the warning that matters: 40% of AI agent projects are at risk of shutdown due to unclear value. A high average with a fat failure tail is the defining shape of 2026 agent economics.

On the supply side, Gartner reports that 80% of enterprise applications shipped or updated in Q1 2026 now embed at least one AI agent— up from 33% two years ago. Google Cloud and NRG's survey of 3,466 senior leaders found 52% of executives report active agent use and 39% have more than ten agents in production. Deployment is no longer the bottleneck.

Payback by function: 3.4 to 11.2 months

BCG and Forrester 2026 data shows 41% of agent deployments reach positive payback within 12 months, 18% within 6 months, with a median time-to-value of 5.1 months. The spread by function is where planning actually happens:

The pattern is consistent: high-volume, well-scoped, measurable work pays back fastest. Legal pays back slowest — not because agents fail there, but because the work is low-volume, high-stakes, and demands governance infrastructure that itself takes time to build.

The adoption-impact gap

Now the numbers that seem to contradict the 171%. McKinsey and Stanford HAI: 88% of enterprises use AI in at least one function, but only 29% report significant impact. McKinsey's 2026 State of AI: 40% of large firms scale agents (up from 27%), yet EBIT impact is flat at 37% — and only 6% are high performers. PwC: 56% of CEOs can't see AI's return— not "returns below expectations," can't see them.

These are not contradictions of the 171% — they are its sampling frame. The average is pulled up by deployments that cleared the value-definition bar; the gap numbers describe the whole population, most of which never did. BCG's finding that AI leaders achieve 3.6x shareholder returns describes the same skew from the top of the distribution.

Why the winners win

Across the surveys, the differentiator is not model choice or infrastructure spend — the gap between winners and losers there is minimal. It is organizational clarity: a CEO- and board-level AI strategy, workflow redesign instead of tool purchase, and measurement that names a baseline before deployment. McKinsey's 6% high performers redesign workflows; everyone else buys licenses and waits.

The cost side is now part of the same discipline. SemiAnalysis measured AI subscription subsidy at 40-70x API-equivalent compute cost — subscriptions are radically underpriced relative to usage, which makes them the cheapest capacity an enterprise will buy this year. IDC found production agent workflows average $117,558 per month with 67% of enterprises blowing their budget. Cheap capacity without meters is how the 40% at-risk projects got there.

What to measure

Three numbers per deployment, each with a published benchmark to compare against: time-to-value against the 5.1-month median; payback against the function benchmark (3.4 to 11.2 months); and share of deployments positive within 12 months against the 41% benchmark. A project that cannot state its baseline before launch is already on the path to the 40%.

The deeper pattern across every source: ROI is a governance output, not a technology output. The enterprises seeing returns decided in advance what would count as a return, who owns the number, and when the project gets killed. The ones that can't see ROI — the 56% — are usually missing the measurement, not the model.

Deep dives

Full context and primary sourcing for every number above, in these six analyses:

Sources: Deloitte enterprise AI ROI research (171% average, function paybacks); BCG/Forrester 2026 deployment data; Google Cloud & NRG ROI of AI 2025 (3,466 executives); Gartner Q1 2026 embedding data; McKinsey 2026 State of AI and Stanford HAI Index; PwC Global CEO Survey; BCG AI Value Report; SemiAnalysis subscription analysis (June 2026); IDC agent cost governance survey. Each figure is covered in depth in the linked articles below.