AI Agent ROI in 2026: What the Research Actually Says, Minus the Hype

The short answer

The credible research on AI agent ROI in 2026 says individual productivity gains are common, but company-level profit gains are still rare. In McKinsey's 2026 survey, 80% of respondents who use AI said it improved their own productivity, yet only 37% attributed any EBIT impact to AI and about 6% saw significant impact. Returns come from narrow, redesigned workflows, not from adding agents everywhere.

Vendor decks quote eye-catching returns. Headlines swing between "agents will replace departments" and "95% of AI fails." Neither helps an owner decide where to spend. This piece pulls together what large, named surveys actually found in 2025 and 2026, with the caveats that usually get cut.

What the big surveys found about AI agent ROI

McKinsey: productivity yes, earnings mostly not yet

McKinsey's "The state of AI in 2026: On the road to ROI," published August 25, 2026, surveyed 1,719 respondents. As reported by The Register, 80% of respondents who use AI said it improved their individual productivity, but 37% attributed at least some EBIT impact to AI, about the same as the year before. "High performers," who attribute at least 5% of EBIT to AI and call the impact significant, stayed at about 6%. Among organisations with more than US$1 billion in revenue, 40% said they were scaling AI agents, up from 27%. And 20% said AI operating costs, including token costs, had constrained their use.

Deloitte: most companies are layering agents, not redesigning

Deloitte's August 2026 survey of 501 US leaders at organisations at least piloting agentic AI found that 42% had tested or deployed agents but only 15% had scaled, orchestrated, cross-functional multi-agent adoption. Only 5% said their business processes were highly prepared for agents. The blockers were practical: 72% lacked "unified, accessible data," 70% didn't feel they could "trust and govern agents," and 67% called integration "too costly and complex."

Gartner: expect many agent projects to be cancelled

In June 2025, Gartner predicted that over 40% of agentic AI projects will be cancelled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. That's a forecast, not a measurement, but it lines up with the survey data above.

Statistics Canada: adoption is rising fast among Canadian firms

For Canadian context, Statistics Canada's analysis of AI use in the second quarter of 2026 found that 19.2% of businesses used AI to produce goods or deliver services in the prior 12 months, up from 12.2% a year earlier. Among AI users, 28.2% used virtual agents or chatbots. StatCan doesn't measure profit outcomes, and notes that the employment impact "remains unclear."

Separating hype from evidence

Some popular numbers deserve care. The most famous is the 2025 MIT NANDA report claiming that 95% of generative AI pilots deliver little or no measurable P&L impact. Fortune's coverage describes the basis as 150 interviews with leaders, a survey of 350 employees and an analysis of 300 public deployments. It's a real signal about pilots stalling, but it is a modest sample with a narrow definition of success, not a measured failure rate for all AI. The same report found that buying from specialised vendors succeeded about 67% of the time, versus roughly a third as often for internal builds.

A few rules for reading any AI ROI statistic:

• Who was asked? Most surveys poll large companies. Small-business evidence is thinner.
• Self-reported or measured? Almost all of these figures are what respondents believe, not audited results.
• Who paid for it? Vendor surveys aren't worthless, but they aren't neutral.
• Productivity or profit? Time saved only becomes return if it is redeployed into revenue or removes a cost.

What the evidence means for a small business

Read together, the research points to a few practical conclusions:

1. Start narrow. The returns show up in specific workflows with clear volume, such as enquiry handling, invoice processing and appointment booking, not in "AI transformation." Our list of what to automate first is built on that idea.
2. Fix data and integration first. Deloitte's top blockers are data and integration, not model quality. If your customer data is spread across five tools, connecting them is step one.
3. Budget for running costs. One in five McKinsey respondents said operating costs limited their AI use. Price the per-task cost before scaling, as covered in what AI automation really costs.
4. Redesign, don't bolt on. Agents layered onto messy processes deliver small, local gains. Rethinking the process around the agent is where bigger returns come from.
5. Kill pilots quickly. If a pilot can't show a measurable result in a set period, stop it. The patterns are covered in why AI pilots fail.

How to measure ROI on your own agent

Don't rely on surveys for your decision. Measure your own baseline for four weeks before launch, then compare:

Measure	Example
Time per task	Minutes to process one invoice or enquiry
Volume handled without a person	Share of calls or emails fully resolved
Speed	Time from enquiry to first reply
Quality	Error rate, rework, complaints
Running cost	Software, usage and review time per task
Revenue effect	Bookings or won jobs attributable to faster response

Our one-week automation audit helps you gather the baseline.

Key takeaways

• AI agent ROI is real at the task level but still rare at the company level, according to McKinsey's 2026 survey.
• Only 15% of US organisations in Deloitte's 2026 survey had scaled multi-agent adoption; data and integration are the main blockers.
• Treat single headline numbers, including "95% fail," with care and check the method.
• Measure your own baseline and costs; start with one narrow, high-volume workflow.

Frequently asked questions

What is a realistic ROI from an AI agent?

There's no reliable average. Credible surveys show most organisations see productivity gains but few see significant profit impact yet. Your return depends on task volume, the cost of the current process and how well the agent is integrated.

Is the "95% of AI pilots fail" statistic true?

It comes from a 2025 MIT NANDA report based on interviews, a survey and public deployment analysis, with a narrow definition of success. It signals that many pilots stall, not that AI fails 95% of the time.

How quickly should an AI agent pay back?

Set a clear test period and a target before you start, based on your own baseline. If the agent can't show measurable time saved, faster responses or more completed work by then, stop or redesign it.

Sources

1. McKinsey says enterprise AI is finally 'on the road to ROI' — The Register, August 25, 2026
2. AI Agents are Only the Beginning: Deloitte Survey Examines the AI Readiness Gap and Reveals How Enterprises Can Prepare for Agentic Success — Deloitte, August 12, 2026
3. Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 — Gartner, June 25, 2025
4. Analysis on artificial intelligence use by businesses in Canada, second quarter of 2026 — Statistics Canada, June 11, 2026
5. MIT report: 95% of generative AI pilots at companies are failing — Fortune (via Yahoo Finance), August 18, 2025

How Flowmatix can help

We put this into practice for businesses in Toronto and around the world through our AI Agents & AI Solutions, digital strategy, business process automation services. Want to know what it would take for your business? Get a personalised quote in about 3 minutes.