AI ROI: The Real 2026 Numbers
Direct answer
Most AI projects still do not move the needle: only 37% of companies report any EBIT impact attributable to AI, flat versus the prior year, and just 6% reach "high performance" (at least 5% of EBIT plus significant impact), according to McKinsey's State of AI 2026 (1,719 surveyed, via The Register). But there is another side: AI agents that reach production at scale deliver 171% ROI globally and 192% in the United States (IDC/Microsoft, September 2026). The difference is not the technology: it is reaching production. About 88% of enterprise AI pilots never get there.
The broken funnel: where ROI dies
The failure numbers are consistent across sources. Between 86% and 88% of AI agent pilots never graduate to production (Forrester/Anaconda). MIT (2025) found that 95% of generative AI pilots generate little or no P&L return. And PwC reports that only 12% of CEOs see benefits in costs and revenue at the same time.
The pattern is clear: companies do not lose by picking the wrong technology. They lose because the project never leaves the lab. A perpetual pilot has no possible ROI.
Those that make it: 171% ROI
When the agent reaches production at scale, the numbers change completely. The IDC/Microsoft study (September 2026, via ayautomate.com) measures 171% return globally and 192% in the United States for agents deployed at scale. IDC/Microsoft's classic measurement already pointed that way: $3.50 back for every dollar invested, with payback in about 14 months.
The gap between McKinsey's 6% and IDC's 171% is the same story told from two ends: the average includes everyone who never finished; the 171% measures those who crossed the finish line.
Why graduates still fail: no success criteria
Reaching production does not guarantee returns. Of the pilots that do graduate, 41% show negative ROI at 12 months (Forrester/Anaconda), and the most common cause is not having defined success criteria before starting. Without a written goal, nobody knows whether the project worked, and without that answer there is no way to improve it.
The rule is simple: before building, write down what success means in numbers (hours saved, leads answered, cost per process) and over what timeframe. To apply this when choosing who builds your project, read how to choose an AI vendor: 7 questions before you sign.
How to measure your AI project's ROI
- Measure the before: what the process costs today in hours and money. Without a baseline there is no ROI.
- One process at a time: the most painful one. Projects that try to automate everything at once are the ones that never leave pilot.
- Written criteria: which number the agent has to move, and in how many months.
- A 12-month horizon: the timeframe where studies measure real returns, not first-month enthusiasm.
To understand which agent to deploy first, start with what an AI agent is for business and how many AI agents your business needs: start with one, measure it, and scale what works.
So does AI deliver ROI or not?
It depends on whether it reaches production. The average is low because most projects never leave pilot (37% with some EBIT impact, McKinsey 2026). Agents that do reach scale deliver 171% ROI globally (IDC/Microsoft, September 2026). The right question is not whether AI works, but whether your project will finish.
How long until the return shows up?
IDC/Microsoft's classic measurement puts payback at around 14 months, and serious studies measure ROI at 12 months. Distrust anyone promising returns in weeks: real value appears when the agent works every day on real data.
What is the most common mistake when measuring AI ROI?
Not defining success criteria before starting. 41% of graduated pilots show negative ROI at 12 months, mostly for that reason (Forrester/Anaconda). Without a written goal, any result can be declared a success, and that is not measuring.
My business is small: do these numbers apply?
They apply even better. A small business measures a single process (calls answered, appointments booked, hours saved) with a clear baseline, without the bureaucracy that kills large pilots. Start with one agent and one number.
At inicIA we turn artificial intelligence into business results: automations and agents that serve, qualify, and book while your team focuses on closing. If you want to see which process in your business gets automated first, let's talk on WhatsApp.
Chat on WhatsApp