McKinsey's 2026 State of AI Survey Shows Why Scouting AI Tools Isn't Enough — You Have to Govern the Pilot
McKinsey's newest State of AI survey — 1,719 respondents, 97 countries, fielded May–June 2026 — surfaces a finding that innovation and IT leaders can't afford to read past: the organizations pulling ahead on AI aren't the ones finding the most tools. They're the ones with the discipline to evaluate and govern what they pilot before they scale it.
AI pilot governance is the set of practices an organization uses to structure, track, and de-risk an AI tool's evaluation period — defined success criteria, owner accountability, risk mitigation, and a documented outcome — before a decision is made to scale, kill, or keep testing. Without it, "piloting AI" is really just quietly adopting AI, one team and one unmanaged trial at a time.
The gap McKinsey found
Eighty percent of respondents say AI has made them personally more productive. Only 37 percent of organizations report any EBIT impact from AI — flat versus last year. The share of true AI high performers (organizations attributing at least 5 percent of EBIT to AI, with self-described significant impact) has held at roughly 6 percent for two years running, even as the share of organizations scaling AI overall climbed from 38 to 44 percent.
More tools, more functions, more agents — and no more proof it's working. That's the gap.
What separates the 6% who are actually seeing ROI
McKinsey's data points to two habits that separate high performers from everyone else, and both point straight at how AI tools get evaluated before they scale.
First, high performers redesign workflows instead of bolting AI onto old ones. Nearly three-quarters of them report fundamentally rebuilding how work gets done because of AI, up from about half last year — versus just a quarter of everyone else.
Second, and more overlooked: high performers manage far more of the risk that comes with AI deployment. They're substantially more likely than others to report actively working to mitigate AI-driven technical vulnerabilities and unauthorized or unintended agent actions. They're also more than twice as likely to report defined processes for measuring the impact of their AI initiatives in the first place.
Put together, that's a governance story. High performers aren't just trying more AI tools — they're running structured evaluations with defined risk controls and measurement built in from day one, not bolted on after something breaks or a budget gets questioned.
The cost-and-chaos problem hiding underneath
Two more numbers make the case for governance even sharper. About one in five organizations say AI operating costs, including token spend, are already constraining their AI use — and among high performers specifically, cost constraints hit software coding agents about three times more often than any other tool. At the same time, 32 percent of all organizations have decided against buying a software product because agentic coding tools let them build the functionality in-house instead, a share that jumps to nearly half among high performers.
Read together, that's an organization running dozens of ungoverned AI experiments at once — some vendor tools, some internally built, each with its own cost profile and risk exposure, evaluated by whichever team got there first. Nobody has a clean before-and-after, so nobody can say with confidence which pilots earned a place in the budget and which ones should have been killed months ago.
That's not an AI-capability gap. It's an evaluation-and-governance gap.
Scout, then evaluate: the workflow high performers are effectively running
McKinsey's high performers are, in effect, doing two things most organizations do separately or not at all: they know what's out there, and they run a disciplined process to decide what earns a place in the stack. That's the exact sequence Traction is built around.
- AI Scouting surfaces the emerging AI tools, agents, and vendors relevant to a defined brief — pulled from a database of 1M+ verified, enterprise-ready companies — so teams aren't discovering options ad hoc through vendor outreach or a coworker's Slack message
- Company Snapshots and SWOT reports give every candidate tool a consistent evaluation format, so a coding agent under review in engineering is being judged against the same criteria as a chatbot under review in customer service
- Pilot Management with governance turns "we're trying it" into a structured evaluation: defined success criteria, an accountable owner, a fixed evaluation window, and risk checkpoints — the same discipline McKinsey's high performers apply to managing AI-driven vulnerabilities and unintended actions
- Portfolio-level dashboards roll pilots up into a scorecard, stage funnel, and owner breakdown, so leadership can see which AI tools are actually earning their keep across the whole organization — not just within the one team that piloted them
Scouting without governed evaluation just produces a longer list of tools nobody has actually tested rigorously. Governance without scouting means evaluating whatever tool happened to land on someone's desk. High performers — and the platforms built to support them — need both, in sequence.
The takeaway for innovation and IT leaders
McKinsey's data makes the diagnosis clear: the AI ROI gap isn't a discovery problem anymore. Most organizations can find AI tools just fine. What separates the 6% seeing real financial impact is a governed process for deciding what to pilot, how to measure it, and when to scale it — versus letting adoption happen team by team, unmanaged.
FAQ
What is AI pilot governance?
It's the structured process an organization uses to evaluate an AI tool during its trial period — defined success metrics, an accountable owner, risk mitigation steps, and a documented outcome — before deciding to scale, kill, or extend the pilot.
Why does McKinsey's 2026 survey point to a governance gap rather than an adoption gap?
Because adoption is already high: 80 percent of respondents report personal productivity gains from AI, and 44 percent of organizations say AI is scaling enterprise-wide. What hasn't moved is EBIT impact (37 percent, flat) — and McKinsey's own high-performer data shows the differentiator is risk management and defined measurement, not more tools.
How are AI high performers different from everyone else in how they manage risk?
McKinsey found that high performers are substantially more likely than other organizations to actively work to mitigate AI-driven technical vulnerabilities and unauthorized or unintended agent actions, and more than twice as likely to have defined processes for measuring AI's impact.
Why are AI operating costs a governance issue, not just a budgeting issue?
One in five organizations report AI costs, including tokens, are already constraining use — and for high performers, that constraint hits software coding agents nearly three times as often as other tools. Without a governed pilot process that tracks cost against outcome, that spend is difficult to evaluate or justify.
How does scouting connect to pilot governance?
Scouting surfaces the tools worth evaluating in the first place; governance is the structured process that decides whether a scouted tool earns a permanent place in the stack. Without scouting, governance evaluates whatever happens to land on a team's desk. Without governance, scouting just produces a longer list of untested options.
What does Traction's Pilot Management feature track?
Pilot Management tracks each AI or technology pilot's stage, owner, success criteria, and outcome, rolling them up into a portfolio-level dashboard — scorecard, stage funnel, and owner breakdown — so leadership can see which pilots across the organization are actually delivering.
Is this different from just running an internal AI pilot program informally?
Yes — an informal pilot has no shared record, no consistent evaluation criteria across teams, and no way to compare outcomes. A governed pilot process run through a system of record means the evaluation that happened in one function is visible and reusable across the rest of the organization.
About Traction
Traction Technology is an innovation management platform that helps enterprise teams discover, evaluate, and govern emerging technologies — from AI-powered scouting and trend reports to RFI management and Pilot Management with built-in governance. Featured in the Gartner Market Guide for AI-Enabled Innovation Management Platforms, February 2026, and SOC 2 Type II certified, Traction gives enterprise innovation and technology teams one system of record for the full innovation lifecycle. See Traction's pricing to learn more.
Neal Silverman is Co-founder and CEO of Traction Technology. He spent 15 years as a senior executive at IDG, running conferences, councils, data services, and consulting businesses connecting enterprises with emerging technologies. Connect with him on LinkedIn.
.webp)