If your company has more AI trials, tools and agent experiments running than results to show for them, it isn't just you. We read fourteen studies published in 2026, most of them between May and September, covering more than 68,000 workers and executives, to see what other companies are running into right now and what the ones that scale do differently.
Almost everyone is using AI. Few have scaled it.
The adoption question is settled. Nearly nine in ten organizations in McKinsey's 2026 survey use AI regularly in at least one function, and 86% of middle-market companies in RSM's survey have partially or fully integrated it.23 What has not moved at the same speed is the step from use to results.
Gartner surveyed 1,303 functional leaders between January and April 2026: only 22% have scaled AI across multiple business units or adopted an AI-first approach. McKinsey, surveying 1,719 executives in May and June, found 44% scaling AI somewhere in the enterprise, but only 37% attribute any EBIT impact to it, essentially unchanged from 2025. The group McKinsey calls high performers is still about 6% of companies.12
Sources: McKinsey, The State of AI in 2026 (n=1,719, fieldwork May–Jun 2026), rows 1–3 and 5; Gartner functional leaders survey (n=1,303, Jan–Apr 2026), row 4. Different surveys and definitions; read as a pattern, not a funnel.
Company size matters. Among organizations with $1 billion or more in revenue, 54% are scaling AI; among smaller ones, about a third. For AI agents the gap is wider: 40% of large organizations are scaling them, up from 27% last year, while smaller organizations stayed at 22%.2
Five things other companies are running into right now
1. More tools than anyone is tracking
The tool count is growing from two directions. From the bottom up, 66% of office professionals at companies with $500 million or more in revenue have used AI tools at work that they believed were not allowed. At companies with 1,500 or more employees, the figure is 72%. Forty-four percent did it to work around the limits of the tools their company approved, and 34% have entered customer data into public AI tools.6
From the top down, 70% of the 2,000 CIOs and CTOs IBM surveyed say teams are deploying technology faster than IT can track it, and 77% say AI adoption is already outpacing their governance.7 Agents add to the count: Microsoft reports 15 times more active agents in Microsoft 365 than a year ago.9
2. The bill arrives before the value
Zylo, which tracks $75 billion in software spend, found spending on AI-native applications up 108% in a year, and 78% of IT leaders reported unexpected charges tied to AI features or usage-based pricing.10 85% of technology executives lack full visibility into real-time AI spend, and Gartner found 11% of organizations do not know what their function spent on AI in 2025.71 Usage-based pricing adds a cost line many budgets did not plan for: 82% of US executives investing in AI told EY they are concerned about token costs.11
Spending is still going up. 85% of functional leaders plan to increase AI spend in 2026, and 84% of middle-market companies expect theirs to rise.13 But only 12% consistently compare what AI delivers with what it costs across the organization.4
3. Employees feel more work, not less
BCG's survey of 11,749 workers found that 47% now spend more time managing and directing AI than doing the work itself, and 41% report a higher cognitive load.5 In a separate study of 2,500 knowledge workers, 42% said they spend more time checking AI output than the tool saves them, and 52% regularly correct colleagues' AI-generated work.12
Use is broad but shallow. In Gallup's Q2 2026 survey of 22,573 US employees, 52% use AI in their role, 30% use it a few times a week or more, and 15% use it daily.8 Even where AI does save time, it is often not redirected: among frontline users who save a full workday a week, 66% get limited or no guidance on what to do with it.5
Source: Gallup Q2 2026 workforce study, n=22,573, fieldwork May 6–20, 2026, ±0.9 pts.
4. Leadership and teams are not on the same page
85% of middle-market leaders agree that their executives are more enthusiastic about AI than their employees.3 Only 26% of AI users in Microsoft's 2026 Work Trend Index say leadership is clearly and consistently aligned on AI.9 CEOs in IBM's 2026 study estimate that only 25% of their workforce uses AI regularly, yet 86% believe their people have the skills to do so.13 And 81% of office professionals believe leadership plays by different AI rules than everyone else.6
5. Pilots stall on data and governance, not on the model
Asked what holds back deployment, middle-market companies put data quality first (34%), followed by security and privacy (30%), legacy systems integration (28%) and skills (28%). Among manufacturers whose pilots fell short, 51% pointed to data quality and 43% to integration.3 Agents raise the stakes: IBM's respondents averaged 54 AI agent incidents last year that needed human correction, and only 11% say they are fully ready for the scale of agent deployment they expect.7
What the companies that scale do differently
Across the studies, the companies that get AI into production share a short list of habits. None of them depends on buying a specific tool.
Sources: McKinsey, Aug 2026; KPMG Global AI Pulse Q3 2026 (n=2,131); Gallup, Jul 2026.
- They redesign the work, not just add a tool. In BCG's data, companies that redesign workflows are 24 points more likely to see measurable business improvement. A clear strategy adds 25 points of impact; better tools without strategy and redesign add about 5.5
- They run AI as a portfolio and stop what doesn't pay. Gartner's high performers track ROI continuously and reallocate or stop underperforming projects; they report positive returns on 81% of their AI initiatives. Low performers did not know the return on 29% of theirs.1
- They build controls into the system. Organizations that embed control in their AI systems, rather than relying on manual review, report 25% fewer incidents, deploy 16 times more agents and report 18% higher operating margins.7
- Managers go first. Where employees say their manager actively supports AI use, 78% use it frequently; without that support, 44%. Only 36% of employees in organizations integrating AI strongly agree they have that support.14
- They start where returns already show up. The use cases leaders pursue most are cybersecurity threat detection and IT service desk automation (54% each). The ones most often reported with positive returns are IT asset and cost optimization (40%), synthetic data (28%) and code generation (23%).1
What this means for a 100 to 1,000 person company
Mid-market companies do not have a transformation office or a dedicated AI team, and they don't need one to act on this. The studies point to a few decisions that fit in one leadership meeting:
- Count what you have. List the AI tools the company pays for, including expensed subscriptions and features inside existing licenses, with an owner and a monthly cost next to each. Most companies find more than they expected; that is where our first article started.
- Choose one or two workflows, not ten pilots. Pick work the team does every week, with a cost you can see today (hours, tickets waiting, a process that depends on one person), and a business owner who wants the result.
- Measure the process, not the logins. An "active user" count tells you people opened the tool. Decide before you start which number in the process should change.
- Settle governance before rollout. Data quality, access and oversharing are where pilots stall. Agree with security on what the AI may see before it is switched on, not after.
- Tell people what to do with the time. If a workflow saves hours, decide in advance where those hours go. Otherwise the saving is real but invisible.
There is also a cost to leaving things as they are. Microsoft raised commercial Microsoft 365 prices on July 1, 2026; Microsoft 365 E3, for example, went from $36 to $39 per user per month, with more Copilot capability included in the suites.15 Capability that is paid for but not running a business process costs more every month it stays switched off.
This is the second article in our Switch It On series. Next: the four levels of AI use, from asking to production, and a three-minute check to see where your company stands.
Want to know what this looks like in your company?
Tell us who you are and what you'd like: more information about the Switch It On program, or a meeting with the Bravent team to look at your Microsoft 365 tenant together. Someone from Bravent will get back to you.
Asking AI questions is not putting AI to work →
The four levels of AI use, from asking to production, and five questions to see where your company stands.