As executives and their enterprises navigate the dynamic world of AI - we're here to provide usable insights that can help you succeed. This edition is our view in September 2026 of the current state of AI in the enterprise. The information below is sourced from over 65 studies and our own work with clients. It is meant to be a pulse on AI - and the major reads we see as advisors and practitioners.
We see 4 major themes in the current environment.
AI usage continues to escalate.
Ask how many companies use AI and the answer runs from 18% to 88%.1 Depending on who you ask and their methodology, the data varies significantly. However, the key insight is that usage is increasing, across all types of firms and organizations.
View as table
| Source | What was asked | % |
|---|---|---|
| US Census Bureau | Used AI in any business function, past two weeks | 18 |
| Eurostat | Enterprises (10+ employees) using any AI technology | 20 |
| US Census, by employment | Same survey, employment-weighted | 32 |
| UK ONS | Businesses (10+ employees) using AI | 35 |
| New York Fed | Regional service firms using AI | 61 |
| Central-bank executive panels | Firm actively uses AI | 69 |
| RSM | Mid-market firms with AI partially or fully integrated | 86 |
| McKinsey | AI used in at least one function | 88 |
What none of these surveys measures is depth. Among American adopters, 57% use AI in three or fewer business functions.2 There is a big dropoff between "trying" something and "running" something. Most companies are still testing or piloting. Very few are using AI in production.
AI, mid-market companiesRSM, n=1,030
AI agents, large companiesCapgemini, n=1,500, revenue above $1B
Uber stood up sixteen small teams across sixteen functions in two months: one AI engineer beside one domain expert, two weeks from watching the work to shipping. Capital allocation across 150 cities fell from fifteen hours to thirty minutes.3
At one distributor, a system went from first meeting to sending real customer email in five weeks. Linking various systems, connecting to the right customers and informing the right humans was mostly process work, not technology-limited.
For a mid-sized companyAdoption is likely already in motion, but not yet seeing results. Data quality is a major barrier, named by 53% of mid-market firms whose pilots did not meet their goals.
1The Federal Reserve Board traces the spread to sampling, question wording and weighting, and notes that executives tend to give the answer that sounds best. FEDS Notes, March and April 2026.
2Bonney and others, The Microstructure of AI Diffusion, US Census Bureau and NBER, April 2026.
3Praveen Neppalli Naga, chief technology officer, Uber, July 2026.
Governance is the accelerator.
The companies reporting returns are disproportionately the ones that have invested in governance and foundational work first. Chief executives whose companies had a responsible-AI framework, and an environment that lets AI reach across the business, were three times as likely to report meaningful financial returns.4 In addition, understanding the costs and accountability for AI has helped drive ROI.
5×Full visibility into AI operating costs
3×Explicit accountability for AI outcomes
Without governance, people do not stop using AI. Where companies provide no approved tools, employees bring their own,5 and one in four malicious breaches now involves AI.6 The gap is widest with agents: 74% of companies plan to deploy them within two years, yet only 21% have mature governance for them.7
JPMorgan Chase gave staff a secure internal AI platform in the summer of 2024. Within eight months about 200,000 employees had it. Chase's chief information officer describes the rule as moving quickly but responsibly, with a person in the middle wherever the output has to be correct. With people's money, that is often.8
A multi-billion dollar supply company's analysts had already found myriad ways to automate processes in their day-to-day work, but had no systematic way to integrate them into their corporate workflows. Governance was used to say "no", not to enable the team.
For a mid-sized companyGovernance should be built around enablement and controls - allowing for breakthroughs, while limiting breakdowns. 67% of mid-market firms now set controls before they pilot.
4PwC, 29th Global CEO Survey, January 2026; 4,454 chief executives in 95 countries.
5SmarterX, 2026 State of AI for Business; n=2,109, weighted toward small businesses and marketing roles.
6IBM and Ponemon Institute, Cost of a Data Breach 2026.
7Deloitte, State of AI in the Enterprise 2026; n=3,235.
8Gill Haus, interviewed by ZDNET, September 2026; JPMorgan Chase disclosures.
AI is an executive function that needs ownership.
Most chief executives now sponsor AI but fewer leadership teams are answering for it.9 They feel the gap: 72% of chief executives told BCG they are now the main AI decision-maker, and half believe their job depends on AI paying off.10
Depending on which survey you read, a quarter of companies have a chief AI officer, or three-quarters do.
Employees whose managers support AI use are 1.7 times as likely to use it every week,11 and executives name middle management as their largest AI capability gap.12
"I think we should have chief roles for outcomes."
Tilak Mandadi, CVS Health, arguing against the chief AI officer title at the WSJ Technology Council Summit, September 2026
Jack Dorsey's company, Block, has a named individual that owns one problem, such as merchant churn in a single segment, for ninety days, with authority to pull whatever resources it takes.13
Many mid-market firms cannot afford full-time AI leadership. Fractional and embedded AI roles can jumpstart activity and set strategy and accountability quickly. These models can also lead to sustainable, accountable outcomes without significant organizational overhaul.
For a mid-sized companyThe enterprise chief AI officer is likely overkill - one senior operator or an embedded external resource who owns the work, the roadmap and the guardrails will likely create quicker results, learnings and an effective organizational approach for the future.
9KPMG, Global AI Pulse, Q2 2026. Where accountability is explicit, established ROI is three times as common (Exhibit 3).
10BCG, AI Radar 2026; 2,360 executives, including 640 chief executives.
11Gallup, AI in the workplace indicator, May 2026.
12Accenture, Pulse of Change, July 2026; 3,000 executives.
13Jack Dorsey and Roelof Botha, "From Hierarchy to Intelligence," Block, March 2026.
Most returns are in productivity.
In controlled trials the gains are real. Support agents resolved 15% more issues an hour, and developers completed 26% more tasks.14 Four in five survey respondents say AI has made them personally more productive. This doesn't always result in bottom-line gains, depending on how the resources are re-engineered.
Planned AI budgets are doubling, to about 1.7% of revenue.15
Klarna said in 2024 that its AI assistant was doing the work of 700 agents. In May 2025 its chief executive said cost had weighed too heavily and quality had suffered, and the company resumed hiring people. JPMorgan reports about $2 billion a year of AI spending returning about the same.
The first return is often visibility. At one wholesaler, the first full pull of receivables showed past-due balances 44% larger than the owners' estimate. Teams were spending significant time following up on the wrong data and the wrong customers.
For a mid-sized company58% plan to spend $1 million or more on AI this year. No published study measures what mid-sized companies get back. Set the baseline before the build.
14Brynjolfsson, Li and Raymond, Quarterly Journal of Economics, 2025; Cui and others, Management Science. The same literature finds reversals: experienced developers in one trial were slower with AI while believing they were faster (METR, 2025).
15BCG, AI Radar 2026. 94% say they will keep investing even without near-term payoff.
What we take from it
We have five main takeaways from what we have seen ourselves and what the market research shows:
- Start inside the business.The best opportunities are rarely visible from outside, and require an understanding of the business, from the people inside the business. Every engagement we run begins with a senior operator embedded in the client's business.
- Put every idea on one map to create alignment.Two questions: is it easy or hard to build, and does it make you different or merely current? We call the result the AI Leverage Map.
Strategic quick wins
Beats peers and is doable today. The window closes.
Document-heavy work · pricing integrity · knowledge captureWhere winners play
Hard to build and genuinely strategic. The prizes that compound.
Proprietary data · redesigned workflowsTable stakes
What every company does. You have to. Nobody wins because of it.
Assistants · meeting notes · general trainingDistractions
Expensive projects that leave you no different. Most write-offs live here.
The internal language model nobody uses- Proof before production.Utilize POCs to determine what is working and its impact. Today's tools allow for quick prototyping on live data with real results. This is the key to really understanding ROI and value creation.
- Utilize external, embedded resources.The quickest and most productive way to build in this environment. AI talent is expensive and hard to find. Start with flexible models that still provide expertise and ownership. Use this to drive a future more sustainable model.
- Build the guardrails first.Governance creates speed and security, and increases the likelihood of success. Create rules that help your teams win.
What this means for your business
If any of this is on your agenda, we'd be glad to talk it through.