The short answer: A year of AI adoption taught mid-market IT leaders three lessons. Use spread but stayed shallow, value followed redesigned workflows rather than licenses, and the headcount cuts leaders expected mostly never came. A Census Bureau study found 57% of AI-using firms apply it in three or fewer business functions. McKinsey found 14% of respondents at AI-using organizations reported workforce declines.
This applies to any mid-size organization planning its second year of AI; we cover it from the buyer’s side, for the CIO, CTO or IT director who owns the AI budget and next year’s plan.
Nine in ten middle market companies now use AI, according to the National Center for the Middle Market, which surveyed 1,000 executives at firms with $10 million to $1 billion in revenue in June 2026. Census data reads narrower: 32% of US firms with 100 to 249 employees reported using AI by early May. The two surveys ask differently, so read them as a range. Either way, the question in most IT shops has moved. It isn’t whether to start. It’s what a year of use actually changed, and what that means for the 2027 plan.
Here’s the pattern the numbers keep pointing to. Year one put AI tools in front of a lot of people and changed how work moved for very few of them. The firms that pulled ahead didn’t buy more seats. They rebuilt a handful of workflows end to end and measured what changed. Meanwhile the layoffs leaders braced for mostly didn’t arrive. That leaves a capacity question most plans haven’t asked.
This guide covers what changed in year one, the three lessons and the number behind each, why headcount held, and the three numbers to settle before you plan year two. Every figure comes from federal data, Federal Reserve research, a university research center or McKinsey’s annual global survey, not from anyone selling a model.
What Did a Year of AI Adoption Actually Change in Mid-Market IT?
Less than the launch demos promised, and more than the skeptics expected. Among service firms in the New York Fed’s region, AI use more than doubled in two years, from 25% in 2024 to 61% in 2026. But for most firms adoption meant access. How work moved changed in far fewer places, and that gap is where year two starts.
The New York Fed’s August surveys of firms in New York and northern New Jersey show the shape of it. Three-quarters of service firms described their AI investment as minimal to modest. Only about 5% called it a major strategic investment. That’s a lot of firms using AI and very few betting on it.
Inside IT, the reach is real. A Census Bureau working paper on how AI spreads inside firms ranked IT third among business functions using AI, at 41% among adopting firms. Only sales and marketing (52%) and strategy (45%) ranked higher. So your team is probably a user already, whether it’s AI in IT operations or help with code and tickets. The harder question is what changed. Often, not much.
The misconception to retire before planning season: adoption isn’t the outcome. A firm can report AI use and still run every process as it did in 2025. Year two is decided by change, not use.
Lesson One: AI Use Spread Wide and Stayed Shallow
Most firms that adopted AI kept it in a small corner of the business. The Census working paper found 57% of AI-using firms apply it in three or fewer business functions, and 65% of firms in the paper limit AI use to three or fewer tasks. Wide adoption, narrow use. That’s the first lesson.
Shallow use shows up three ways, and each one has its own fix.
1. A few functions, not the whole business
The same Census paper found a positive correlation between a firm’s commercial performance and how broadly it integrates AI. It’s a correlation in a working paper, not a measured return. But it points the same way as everything else here. Narrow use caps what AI can do for you.
2. A few people, even where it’s adopted
Among New York Fed region service firms that use AI, the median share of workers using it was just 17%. Roughly one in six. A small group carries the use while everyone else works the way they did before. The rollout reached the enthusiasts and stopped.
3. A few tasks, mostly to help rather than replace
Two-thirds of AI users, 66% in the Census paper, rely on it only to augment tasks. That’s not a failure. It’s where most firms sensibly start. The counterintuitive part is that shallow use isn’t a skills problem you fix with more licenses. It’s a workflow problem, and an enterprise AI rollout that skips the workflow tends to stall in exactly this spot.
In Census data, 57% of AI-using firms stop at three business functions or fewer. Count yours by weekly use, not licenses. Wider use tracks with stronger performance. Three or fewer means tools bought, work unchanged. Not sure where to widen? Score your AI maturity
Lesson Two: Value Followed Workflow Redesign, Not Licenses
The firms getting real value from AI changed how the work moved, not just which tools people could open. In McKinsey’s 2026 survey, nearly three-quarters of high performers fundamentally redesigned workflows because of AI, against about a quarter of other respondents. Redesign is the dividing line. A license is the entry fee.
That high-performer share was 55% a year earlier, so the leaders are pulling further ahead. The table below sets where year one usually went against what the leaders did instead.
| Where year one went | What it bought | What the leaders did instead |
| Licenses for everyone | Access, not habits | Rebuilt a few workflows end to end |
| A pilot per team | Scattered wins | Picked workflows with an owner and a baseline |
| Counting seats | Spend, not results | Defined how impact gets measured |
| Waiting for headcount savings | A plan built on cuts | Planned for the capacity freed |
Why does redesign beat access? Because time saved only turns into value when the work around it changes. In survey data from the second half of 2024, the St. Louis Fed found generative AI users saved about 5.4% of their work hours, roughly 2.2 hours a week. A third of daily users saved at least four hours a week, against about one in ten people who used it one day a week. Same tools, very different returns. The habit came from the workflow.
Those are self-reported numbers, and the St. Louis Fed notes that saved time may go to easing up rather than the next task. Hours saved inside an unchanged workflow tend to leak away. Hours saved inside a redesigned one have somewhere to go.
High performers are also about twice as likely to have defined processes for measuring the impact of their AI initiatives, McKinsey found. They knew what each workflow was supposed to change, and they checked. Our guide to measuring AI ROI covers the workflow metrics a CFO will accept.
The offensive play: fund workflows, not tools. Put each AI line in the 2027 budget against a named workflow with an owner and a before-and-after number. A line that names only a tool is a license renewal, not a plan.
Why Didn’t AI Shrink IT Teams the Way Leaders Expected?
Because most firms used AI to help people do the work, not to remove them from it. In McKinsey’s 2026 survey, 14% of respondents from organizations using AI said it had reduced their workforce, less than half the 32% who expected cuts a year earlier. The cut keeps slipping a year out.
The forecast hasn’t gone away, either. In the same survey, 39% expect AI to reduce headcount over the next year. A number that missed once is still sitting in a lot of 2027 decks.
Federal data points the same way. The Census paper found AI-related employment decreases at only 2% of firms. Among New York Fed region service firms that use AI, just 4% laid off workers because of AI over six months, while just over a third retrained workers. About 15% said they’d hired fewer people than they otherwise would have, and about 13% said AI led them to hire more. When a third of AI-using firms are retraining and almost none are cutting, the job is changing shape. It isn’t disappearing.
Three moves keep a 2027 plan honest about headcount:
- Replace every assumed cut with a named task. If a budget line says AI saves a role, write down which tasks move and who picks up what’s left.
- Budget the time to retrain, not just the tool. Plan the hours, or the rollout eats them.
- Re-forecast at mid-year, not at renewal. Check whether freed hours went to new work or quietly disappeared.
Automate the timing, never the human moment. Let a dashboard flag when a workflow’s numbers drift. Keep the conversation about what someone’s job becomes between people, under clear AI governance that says who decides.
If your 2027 plan banks on AI cutting headcount, check each line for the task behind it. Every cut should name the work that moves. One with no task repeats a forecast that missed this year. Found one? Let’s talk it through
Three Numbers to Settle Before You Plan Year Two
Track three numbers before the 2027 budget locks: business functions with weekly AI use, workflows redesigned with a before-and-after measure, and freed hours that went to named new work. Census data puts most adopters at three functions or fewer. Set your baseline now. Without it, year two repeats year one.
Take the baseline before anything new is bought, then read the three together at mid-year.
- Functions with weekly use. Count the business functions where people use AI every week. Three or fewer puts you with the 57% in the Census data. Four or more is wider than most.
- Redesigned workflows with a measure. Count the workflows changed end to end because of AI, each with a before-and-after number. Zero is a common answer, and it’s the one to fix first.
- Freed hours with a destination. For each workflow, estimate the hours saved and name where they went. Hours with no destination aren’t savings yet.
These thresholds are a rule of thumb, not a benchmark. The diagnosis comes from the combinations. If use is wide but no workflow is redesigned, you’ve bought access: pick two workflows and rebuild them. If a workflow is redesigned but use stays in one function, you have a proof without a path, so take the pattern to the next function. If hours are freed but none has a destination, the saving exists on paper only. Plan the work it pays for before you plan any cut.
Our piece on starting with one workflow shows how to begin small. Whether you rebuild in-house or with help is its own call, and build vs. buy lays out how to decide.
Frequently Asked Questions
How much should a mid-size company budget for AI next year?
We found no reliable public benchmark by company size, so budget by workflow instead. Price the two or three workflows you plan to rebuild, including the people time to redesign and retrain, then add the tools. A budget that starts from license counts usually funds access and little else.
What is an AI maturity assessment?
An AI maturity assessment scores how prepared an organization is to use AI well, across areas such as strategy, data, governance and staff skills. It shows where the gaps are before money is spent. A useful one ends in a short list of priorities, not a grade.
Who should own AI in a mid-size company?
One executive should own the AI plan and its budget, usually the CIO or another senior technology leader. Each redesigned workflow also needs a business owner who answers for its numbers. Ownership split across every department tends to produce scattered pilots and no measurement.
Is generative AI the same as agentic AI?
No. Generative AI produces content, such as text, code or summaries, in response to a prompt. Agentic AI takes actions across several steps toward a goal, with less human input along the way. Agentic tools raise bigger questions about permissions, oversight and who checks the result.
How long does it take to see results from AI?
It depends on the workflow, not the tool. A narrow, well-measured workflow can show a before-and-after within a quarter, which is a reasonable first read. Broad rollouts with no baseline can run a year without showing anything you can defend.
Year One Bought Access. Year Two Has to Buy Change.
Look at your own year one against the numbers. Census found 57% of AI-using firms at three functions or fewer. McKinsey found 14% of respondents at AI-using organizations reported workforce declines, against 32% expecting them a year earlier. If yours looks like that, you’re typical. The leaders redesigned a few workflows, measured them, and planned for capacity, not cuts.
The plan, the budget and the final call are yours. Working alongside your team to rebuild the workflows that matter, with the same senior people who scoped the work, is what C4 Technical Services’ consulting team does. It leaves your team able to run what was built. If you’re writing the 2027 AI plan now, start the conversation.
References
- “Middle Market Indicator, Mid-Year 2026.” National Center for the Middle Market, The Ohio State University Fisher College of Business, survey of 1,000 executives fielded June 2026. Published July 29, 2026. Verified live October 20, 2026.
- “Large Firms With at Least 20 Employees Biggest AI Users.” US Census Bureau, America Counts story on the Business Trends and Outlook Survey (Grundy, Breaux and Khatiwoda). Published May 26, 2026. Verified live October 20, 2026.
- “The Microstructure of AI Diffusion: Evidence from Firms, Business Functions, and Worker Tasks.” US Census Bureau, Center for Economic Studies working paper CES-WP-26-25 (Bonney, Breaux, Dinlersoz, Foster, Haltiwanger and Pande). Published April 15, 2026. Verified live October 20, 2026.
- “Businesses Are Using AI to Transform Work, Not Cut Jobs.” Federal Reserve Bank of New York, Liberty Street Economics (Abel, Deitz, Emanuel and Montalbano), from its August regional business surveys. Published September 1, 2026. Verified live October 20, 2026.
- “The state of AI in 2026: On the road to ROI.” McKinsey & Company, global survey of 1,719 participants. Published August 25, 2026. Verified live October 20, 2026.
- “Generative AI, Productivity and the Future of Work.” Federal Reserve Bank of St. Louis, Open Vault (research by Bick, Blandin and Deming). Published October 8, 2025. Verified live October 20, 2026.