McKinsey's 2026 AI Survey: Why 94% of Companies Still Can't Show ROI

McKinsey's 2026 State of AI survey defines an AI high performer as an organization that attributes at least 5% of its EBIT to AI use and describes AI's impact on the business as significant. The survey ran online from May 4 to June 8, 2026 and drew 1,719 participants in 97 nations; about 6% met both criteria, the same share as the prior year (McKinsey & Company, The State of AI: Global Survey 2026).

The 94% Problem

McKinsey's survey arrived at a number the AI industry has been reluctant to state plainly: record AI spending is not moving the earnings needle for 94% of organizations (TechTimes, August 2026).

The high-performer share held flat year over year at about 6% in both 2025 and 2026 (McKinsey & Company, The State of AI: Global Survey 2026). Two years of better tooling moved it nowhere, which is what sends the question toward how organizations are structured around AI.

Ownership Is the Variable

High performers are three times more likely than their peers to strongly agree that senior leaders at their organizations demonstrate ownership of and commitment to their AI initiatives (McKinsey & Company, The State of AI: Global Survey 2026). That is a perception measure rather than an audit of what those leaders actually did, so read it as a signal about where conviction sits inside a company.

What McKinsey measured here is conviction at the top, reported by the people below it, and that is a different variable from budget size or model access.

McKinsey does not break down what that ownership produces in practice. From the deployments we run, two practices follow from it. The first is focused deployment: AI pointed at one specific workflow instead of rolled out as a company-wide productivity experiment. The second is measurability: a defined function produces a countable output, so ROI can be calculated instead of assumed.

Companies spending on AI without both properties tend to run the same experiment in a loop. Buy a tool, give it to the team, hope something improves. The improvement rarely shows up, and when it does it is rarely traceable to the spending that supposedly caused it.

Company B defines a function: qualify inbound leads and route them to the right technician within 15 minutes of a service request. They build an agent to own that function end to end, running inside their existing scheduling tool. The output is countable, in leads routed per day, response time per inquiry, and jobs booked per week. The ROI calculation follows from the counter.

A Three-Question Filter

Before renewing any AI tool subscription or scoping a new AI project, three questions separate productive spending from sunk cost.

1. Which specific business function is this agent running? Name a bounded function: dispatch routing, quote follow-up, client intake, invoice matching. If the answer runs longer than one sentence, the deployment is not focused enough to measure.

2. Who owns the outcome? If the answer is "the team" or "everyone," nobody owns it. A named person has to be accountable for whether the agent is working, because without that accountability there is no feedback loop and the function drifts.

3. What is the measurable output? Leads qualified per day. Tickets closed without escalation. Hours of manual work eliminated per week. If the output cannot be counted, the ROI cannot be calculated, and the investment stays in the 94%.

The Limitation Worth Naming

McKinsey's 2026 survey draws on 1,719 participants across 97 nations (McKinsey & Company, The State of AI: Global Survey 2026), and it reports no separate cut for firms under 50 staff, so we read most Canadian SMBs as outside the sample. Applying its findings to a small Canadian firm requires a caveat: the structural patterns of ownership, specificity and measurability hold, while the scale and the implementation path look nothing alike.

For an SMB, senior ownership usually means the founder deciding this matters, without an AI committee. Focused deployment means one agent for one function, without a cross-departmental program. The principle holds at both scales, and the smaller implementation is the easier one to measure, because a small firm can see the whole workflow at once.

One barrier in McKinsey's companion trust research applies regardless of company size. In its 2026 AI Trust survey of roughly 500 organizations, nearly two thirds named security and risk concerns, ahead of regulation or technical limits, as the top barrier to scaling agentic AI, and average responsible-AI maturity sat at 2.3 out of 4 (McKinsey, State of AI Trust in 2026). The same survey ties EBIT impact above 5% to responsible-AI spending over $25M (McKinsey, State of AI Trust in 2026), an enterprise threshold that does not transfer to a small firm. What transfers is the barrier: the thing that stops an agent going from pilot to production is usually confidence about what it is allowed to do. For an SMB that means guardrails built into the agent from the start, not bolted on after an incident. Our write-up on the governance gaps most SMBs carry covers what those guardrails look like in practice.

What This Means for Your Spending

What the data shows is narrow and worth stating precisely: the high-performer share has not moved in two years while the tools improved (McKinsey & Company, The State of AI: Global Survey 2026). It does not measure whether the gap between the 6% and everyone else is widening, and we have not seen evidence either way. What it does rule out is the idea that better tools alone will move a company across the line, because two years of better tools did not.

Start from the function. Name the business function an agent should run end to end, decide in advance how its output will be counted, and put one person's name against the result. That sequence is ours, not McKinsey's, and it is built to make the three things the survey associates with high performers visible before you spend rather than after. It costs nothing to apply to a purchase order you are already holding. If you want a structured version of it, our checklist on whether a Toronto SMB is ready for AI walks the same ground.