The short answer: Before you buy AI training, ask what your team will do differently 30 days after it ends. Then ask three more things: which approved tools and tasks it teaches, how the vendor measures use rather than completion, and who supports people after day one. Only 15.9% of US workers say their employer offers any AI training.
This applies to any organization buying AI training; we cover it from the buyer’s side, for the CIO or IT director who signs the contract and answers for whether it worked.
Only 15.9% of US workers say their employer offers any AI training, according to the New York Fed’s analysis of its November 2025 Survey of Consumer Expectations. In the same survey, 39% say they’ve used AI tools in their job in the past year. So plenty of your people already use the tools. Most are figuring it out alone. That gap puts AI training on a lot of 2027 budget lists. Some of it will get bought in a hurry.
Hurried buying lands on what’s easiest to show. A course catalog. A completion dashboard. A certificate with your logo on it. Everyone finishes, the dashboard turns green, and the work looks exactly like it did before. Completion is easy to sell because it’s easy to count. It’s still the wrong number. The number that matters is what your people do on an ordinary Tuesday afternoon a month later.
This guide covers why completion misleads and the four questions to ask a vendor. Then it shows what real support looks like, how to get a usage measure into the contract, and a 30-day test of whether the money changed anything.
Why Course Completion Is the Wrong Number to Buy
Completion measures attendance, not change. In Gallup’s May 2026 reading, 47% of US employees say their organization has integrated AI tools, yet only 15% use AI daily in their role. Our point: a finished course is an input, and daily use is the outcome. Buy against the outcome.
The money is moving faster than the measurement. Training magazine’s 2025 industry report found 25% of US organizations expect AI training to get more funding next year, second only to management training at 30%. Yet AI made up just 2% of training delivery hours, up from 0.8% a year earlier. So more money is heading toward a category few buyers have a test for.
A course is also only one part of the picture. Economists Anders Humlum and Emilie Vestergaard surveyed 25,000 workers in 11 AI-exposed occupations in Denmark for a July 2026 study. Chatbot adoption was about 41% where employers took no active steps and 93% where they combined encouragement, enterprise tools and training. The study doesn’t split out training’s own share. Don’t expect a course to carry the other two parts.
Access isn’t learning, either. The New York Fed’s authors caution that offered training “does not necessarily mean that these workers have already received AI training.” A completion report hides that gap. So does a course with no link to your AI rollout plan. Nobody notices when it misses.
What Should You Ask an AI Training Vendor Before You Sign?
Ask four things before you sign: what changes in 30 days, which approved tools it uses, how use gets measured, and who supports people after. In Gallup’s data, users who say AI helps their productivity rise from 45% at one or two uses to 90% at seven or more. Aim for breadth of real use.
Good vendors answer all four in writing. Weak ones answer the first and change the subject.
1. Ask what people will do differently in 30 days
Here’s the detail most buyers skip: ask for the task list, not the syllabus. A syllabus says what gets taught. A task list names the real work each role should do with AI a month later. Think drafting a change request, summarizing a ticket queue or writing a test script. No tasks for your roles means the course was built for someone else.
2. Ask which approved tools and tasks it uses
The New York Fed survey found 37% of employed respondents say their workplace doesn’t offer AI tools, and another 11% say their employer actively prohibits them. Training people on a tool they can’t use at work is a cost with no return. Hand the vendor your approved-tools list and your written usage rules before the first call. Every module should map to one.
3. Ask how they’ll measure use, not completion
Most AI tools already count use. Microsoft’s Copilot usage report, for one, shows active users and an active-users rate over 7, 28, 90 or 180 days. Ask the vendor which number in your own systems will move, by how much, and by when. Then write the answer down.
4. Ask who supports people after day one
Close the loop before you sign. Who answers questions in week two? Who checks in at day 30? If the answer is “our help desk,” the support ends when the invoice clears. A good answer names a person, a schedule and the part your managers play.
Pull up the last training your team finished and ask each person to name one task it changed. Everyone should have one. If not, you paid for a transcript, not a capability. Not sure what to ask next? Ask what we’d check
What Real Follow-Up Support Looks Like After the Course
Real follow-up support sits where the work happens and lasts for weeks, not one session. Gallup found 78% of employees who strongly agree their manager supports AI use are frequent users, against 44% of those who don’t. Manager support tends to travel with frequent use. So buy support that equips managers first.
The table below compares weak and strong vendor answers to the four questions, and what each answer tells you.
| Question | Weak answer | Strong answer | What it tells you |
| What changes in 30 days? | “Everyone will be AI-literate” | Three named tasks per role | Whether the course fits your work |
| Which tools and tasks? | “We cover all the major tools” | Your approved tools, your written rules | Whether people can use it on Monday |
| How do you measure? | Completion rate and quiz scores | Active-use rate in your own systems | What you’re actually buying |
| Who supports people after? | “Our help desk” | Named coach, manager guide, day-30 check | Whether the habit survives week two |
Why does manager support beat a better course? Because managers decide what gets done on Tuesday. And most of them already use the tools. In the Real-Time Population Survey, published through the St. Louis Fed’s FRED database, 72% of US workers in management occupations used generative AI for their job in the second quarter of 2026. Across all employed adults it was 45%. Your managers can model the habit. They just need a job description for it.
The payoff is time. In the Danish study, nearly one in five users saved more than an hour a day. That’s about two and a half working days a month for every person who gets there.
The offensive play: put your managers in the first cohort, not the last. Give each one a 15-minute weekly check-in script for the month after the course. It costs nothing extra, and it turns a one-day course into a month of practice. It’s the same principle behind upskilling that sticks.
How Do You Get a Vendor to Commit to a Usage Measure?
Write the measure into the statement of work before the first session. Name the tool, the number and the date, such as the active-users rate in your Copilot usage report at 28 and 90 days. A vendor that won’t sign a number is selling attendance. One that will is selling a result, and you’ll know which before the kickoff call.
Three steps get it done:
- Pull the baseline before kickoff. Record the active-users rate for the 28 days before training starts. Without it, any number afterward looks good.
- Put the day-30 and day-90 readings in the SOW. Name who pulls them and who sees them. Agree up front what the vendor owes if a reading misses: a repeat session, a coaching block or a holdback.
- Book the reviews now. Put both reviews on calendars the week you sign, with the vendor’s lead and your managers in the room.
Automate the timing, never the human moment. A dashboard can flag who stopped using the tool after week two. A person should ask them why. The answer is usually a task the course skipped, and only a conversation finds it.
Larger firms use AI more, so the question lands on them first. Census Bureau data from December 2025 to May 2026 show 32% of US firms with 100 to 249 employees using AI. At 250 or more it was 37%, against 17% to 20% of all firms. Use the same baseline-first logic to measure AI ROI. The CFO will ask.
Can you say how many of last quarter’s trainees used the tool in the past 28 days? If most didn’t, the course taught a skill nobody kept. A trained team uses it every week. Not sure where the gap starts? Start with the assessment
The 30-Day Test: What Changed in the Work
Thirty days after the course, check three numbers: active-use rate, tasks per person and manager check-ins held. Set the targets before kickoff. The New York Fed found 39% of US workers have used AI tools in their job in the past year, so measure from your own baseline, never from zero. Then read the three numbers together.
Run the same three on your last training before you buy the next one. If you can’t, that’s your first finding.
- Active-use rate. Active users divided by trained users. Check how your tool’s report defines its own rate, or work it out yourself. Look for a clear rise over your baseline, and ask why if it slides back by day 90.
- Tasks per person. Ask each person to list the work tasks they now use AI for. Breadth seems to matter: in Gallup’s data, users who say AI helps their productivity climb from 45% at one or two uses to 90% at seven or more. Three tasks per person is a sensible first target.
- Manager check-ins held. Count the day-30 conversations that actually happened. The target is all of them.
These targets are a rule of thumb, not a benchmark. The diagnosis comes from reading the numbers together. If use is up but tasks per person stay at one, people treat the tool as a search box, so train on tasks, not features. If people list several tasks but the usage report stays flat, they may be using tools you haven’t approved, so check your rules before you retrain anyone. If the check-ins happened and both numbers are flat, the course didn’t fit the work. That’s the vendor’s problem to fix, and the SOW should already say so.
Judge it the way you’d judge any build-or-buy call on AI: against a baseline you recorded first. If you can’t name the tasks yet, an AI readiness check is the faster first step.
Frequently Asked Questions
Buy the Change in the Work, Not the Course
Picture your team 30 days after the next course. Only 15.9% of US workers say their employer offers any AI training. In the Danish study, adoption was about 41% with no active steps and 93% with training, encouragement and enterprise tools together. Don’t shop for a fancier course. Buy a measure, a support plan and managers who carry the habit.
The tools, the rules and the judgment about your work are yours to own. Building training around them and checking whether it took is what C4 Technical Services’ AI consultants do, with the experts who scope the work delivering it. If you want a second read on a training proposal before you sign, let’s talk it through.
References
- “Use of Gen AI in the Workplace and the Value of Access to Training.” Federal Reserve Bank of New York, Liberty Street Economics (Ali Hashim, Gizem Kosar and Wilbert van der Klaauw; November 2025 Survey of Consumer Expectations). Published April 14, 2026. Verified live October 20, 2026.
- “Indicator: Artificial Intelligence.” Gallup. Published January 2026 (updated with the May and August 2026 waves). Verified live October 20, 2026.
- “AI and Workplace Productivity: What Leaders Need to Know.” Gallup. Published July 27, 2026 (updated September 30, 2026). Verified live October 20, 2026.
- “2025 Training Industry Report.” Training magazine. Published November 10, 2025. Verified live October 20, 2026.
- “Still Waters, Rapid Currents.” Anders Humlum and Emilie Vestergaard, RFBerlin Research Insight No. 28/26 (survey of 25,000 Danish workers). Published July 2026. Verified live October 20, 2026.
- “Microsoft 365 Copilot usage.” Microsoft Learn, Microsoft 365 admin center documentation. Published August 18, 2026 (last updated date). Verified live October 20, 2026.
- “Generative Artificial Intelligence, Adoption Rate for Work: Management Occupations.” Federal Reserve Bank of St. Louis, FRED (Real-Time Population Survey, Alexander Bick, Adam Blandin and David Deming). Published August 2026 (Q2 2026 observation). Verified live October 20, 2026.
- “Generative Artificial Intelligence, Adoption Rate for Work: Employed Adults.” Federal Reserve Bank of St. Louis, FRED (Real-Time Population Survey, Alexander Bick, Adam Blandin and David Deming). Published August 4, 2026 (Q2 2026 observation). Verified live October 20, 2026.
- “Large Firms With at Least 20 Employees Biggest AI Users.” US Census Bureau, America Counts (Business Trends and Outlook Survey). Published May 26, 2026. Verified live October 20, 2026.
