Only 22% of organisations have scaled AI across business units. What the rest are missing

What Gartner found

Scaling is still the exception. Gartner surveyed 1,303 leaders at organisations with annual revenue of $50 million or more, between January and April 2026; the release does not state the geography. Of those, 22% had scaled AI across multiple business units or adopted an AI-first approach.¹

Three other findings explain much of the gap. About 11% of respondents did not know their function's AI spend in 2025.¹ Productivity was a target for 75% of functional leaders and took roughly 30% of functional AI spend, nearly double the next objective.¹ And high performers reported positive returns on 81% of their AI initiatives, while low performers could not determine returns for 29% of theirs.¹

Spending is still rising regardless. In the same survey, 85% of functional leaders planned to increase AI spending in 2026.¹ Gartner's recommendation is to track returns by outcome category, such as productivity, revenue growth, risk mitigation and innovation.

Gartner, January–April 2026

Spending rises; measurement lags

22%have scaled AI across multiple business unitsGartner, Jan–Apr 2026 ¹

About 11%did not know their function's 2025 AI spendGartner, Jan–Apr 2026 ¹

75%of functional leaders targeted productivityGartner, Jan–Apr 2026 ¹

How it fits with McKinsey and KPMG

The three big surveys this summer point the same way. In McKinsey's State of AI 2026 (1,719 respondents in 97 countries, 4 May to 8 June 2026), 37% attributed any EBIT impact to AI.² In the same survey, 40% of respondents at organisations with more than $1 billion in revenue reported scaling AI agents, against 22% below that size.³

KPMG's Q2 2026 Pulse adds the cost side. Of 204 US leaders at firms with $1bn or more in revenue (28 April to 25 May 2026), 53% were deploying AI agents and 26% reported full, real-time visibility of what their AI systems cost to operate.⁴

The definitions differ. Gartner counts AI scaled across business units, McKinsey counts agents being scaled and KPMG counts agents deployed. Do not set the percentages against each other. Read the pattern instead: use is widespread, spend is rising, and the ability to show what it returns lags behind. Our note on McKinsey's survey covers the EBIT gap in more detail.

What are the rest missing?

Our view: the gap is less about models and more about bookkeeping. An organisation that cannot state what one agent costs and returns has no basis for a second. Gartner's findings map onto three missing habits.

A known cost

About one respondent in nine did not know their function's AI spend at all.¹ Below that sits a larger group who know the licence bill but not the usage, the people time or the cost of exceptions. Cost per completed action is the figure that lets two agents be compared.

An outcome beyond productivity

Productivity targets are reasonable for a first agent, but they are easy to claim and hard to bank. Hours saved only become money when work is redeployed or volume grows without hiring. A second agent should name its outcome category before it is built, as Gartner recommends.

An owner

Scaling across business units means each unit has someone who owns the agent's exceptions and results. Without that, the second agent is an orphan, and the first one often turns into one as well.

Three checks before you add a second agent

Run these against your first agent. If it fails any of them, fix that before you build again.

Before agent number two

1. Cost per action

   Usage, platform and people time, divided by actions completed without a person, for the last full month.

2. Signed result

   The outcome agreed before the build, measured and signed by the process owner, in one outcome category.

3. Named owner

   One person who owns exceptions, approves changes and will own the next agent's results too.

If the first agent passes, the second can borrow its log design, approval pattern and cost reporting. That reuse is most of what scaling means in practice. Our guide to pilot length and budget covers the first agent's measurement period, and our scoring model helps pick which process comes next.

Limits of the survey

This is a press release on a survey, not the full report, and it does not state the geography. "Scaled across multiple business units" is the respondent's own judgement. The sample starts at $50 million in revenue,¹ so it says little about smaller companies, where one agent in one team may be the right scale for years.

Our view: for a mid-sized business the lesson is not to chase scale. It is to run the first agent like something you intend to scale, with the numbers to prove it. The AI automation page sets out how we scope one workflow with its acceptance criteria agreed up front.