If your company spent more on AI this year and you still can't point to a change in revenue, margins, or turnaround times, you're in the same position as many CEOs heading into 2027. The tools are paid for, and people are using them, but the results haven't shown up in the numbers you report to your board.
Here are three AI experiments CEOs can run in 2027 to find out whether AI is improving the work and to expand the parts that are.
In the first experiment, executives and middle managers answer the same four questions about your AI rollout, and you compare their answers. The second compares one recurring workflow before and after AI, so you have real results, not estimates. The third asks each team to show reusable AI work, such as a prompt, a documented workflow, or a repeatable process, instead of sending usage reports. All three use information you already have or can collect quickly, and each one gives you numbers you can bring to your board and your 2027 budget.
Why Your AI Spending Isn't Showing Up in Results
Hereās how this usually looks inside a company:
Usage reports go up while results stay flat. Revenue, margins, and turnaround times look the same as they did before the rollout. Logins show who opened the tool, not whether anything improved.
Executives and managers disagree about whether AI is paying off. Leadership may report strong early returns, while the managers running the work report extra steps, errors to fix, and no visible time savings. When managers see weaker results than executives expect, the rollout slows, and leadership often doesnāt learn why for months.
They also disagree about speed. Executives often believe the company is ahead of competitors on AI. Managers who see the day-to-day reality are usually less sure.
Successful pilots stay in one department. A team tested something useful last spring, and no other team has picked it up since.
Training is measured by attendance. Course completions get reported, and nobody checks whether the people who finished the course do their jobs differently.
No one can put a dollar figure on AI. Your leadership team can't say what AI is worth to the business today or what it could be worth next year.
Buying another tool wonāt fix these problems, because each one comes from missing information. You need to know what AI changed in the work, whether your leaders agree, and what your people can do with it. The three AI experiments below answer those questions with your own data.
1. Ask Your Executives and Managers the Same Four Questions
The first experiment is a short written survey. Give your executive team and your middle managers the same four questions about your AI rollout, collect their answers separately, and compare what each group wrote. It takes very little time to set up, and it tells you whether your leaders are working toward the same result.
Why run it?
Executives and managers usually see AI from different positions. Your executives probably use it for summaries, drafts, and planning, where the tools perform well. Managers work with AI in daily operations, where someone has to catch and fix a wrong answer before it reaches a customer. Both views are accurate. When they donāt match, decisions slow down and teams end up doing the same work twice.
How do you run it?
Send the same four questions to your executive team and to the managers who lead teams using AI. Ask each person to answer alone, in writing, without discussing it with anyone first:
What is the goal of our AI rollout?
How will we know itās working?
Who makes the final decisions about AI tools and processes?
What skills do our people need that they donāt have yet?
Keep the answers anonymous so managers feel free to be honest. A short online form works well, and most people can finish it in about 10 minutes.
What do the answers tell you?
Put the executive answers next to the manager answers and look for differences. For example, executives might say the goal is faster proposals, while sales managers say it is reducing headcount. That one difference would explain why some managers have been slow to introduce new tools to their teams.
Different answers to the third question usually mean nobody is sure who owns AI decisions. Mismatched answers to the fourth question tell you where training money should go first.
How do you continue it in 2027?
Ask the same four questions every quarter to see whether the answers are getting closer together. Share the results with both groups. Treat cautious answers and failed pilots as useful information, since they point to the parts of the rollout that need more support. Bring managers into AI planning meetings before decisions are final, so the next round of answers starts from a plan both groups helped build.
2. Measure One Workflow Before and After AI
Which workflow should you pick?
Choose one task your company repeats every week, with a clear owner and a cost you can see. Proposals, monthly reports, customer onboarding documents, and job descriptions all work well. Avoid tasks that happen only a few times a year, because you won't collect enough results to compare. Pick something important enough that a better result would matter to the business, and small enough that one team can test it without disrupting everyone else.
How do you set the baseline?
Before anyone changes how the work gets done, record how it works today. Track these four things for at least two or three rounds of the task:
Total hours from start to finish
The number of people who touch it along the way
Errors or rewrites that happen before it's approved
Time spent waiting between handoffs
Without this baseline, any improvement you report later will be a guess.
What makes the AI version reliable?
Give the team three things before they start using AI on the task. First, they need a standard format for the information they give the AI, so results don't depend on who wrote the request. A defined review step should check the output before it goes anywhere. The last piece is a named reviewer who signs off on the final version.
For example, imagine a sales team that took six days to finish a proposal before the test. With a standard input format and a review step, they finished the next five proposals in two days each, and the reviewer caught two pricing errors along the way. Those are numbers a CEO can take to the board.
Where does the saved time go?
Decide ahead of time how the team will use the hours they get back. If you don't, the time disappears into email and meetings. A sales team could use it to follow up on more leads, and an operations team could use it to clear a backlog.
How do you expand it?
Once the results hold up over several rounds, document the full process, including the input format, the review step, and the before and after numbers. Then hand that document to a second team with a similar task and measure their results the same way.
3. Ask Teams to Show Their AI Work Instead of Reporting Usage
Why isn't usage data enough?
A person can log in to an AI tool every day and still produce nothing the company can reuse. Usage reports count activity, so they reward people for opening the tool rather than for doing better work with it. Course certificates have the same limit. They prove someone finished the training, and they say nothing about whether that person's work changed afterward.
What should teams show you?
Ask each team lead to bring three kinds of examples to a short review meeting:
Prompts a coworker can pick up and use without extra explanation
A documented workflow that includes AI, with the review step written down
Processes other people on the team already rely on every week
Keep the meeting to about 30 minutes per team, and ask the person who built each example to walk through it. You'll learn more from watching someone explain their work than from any dashboard.
For example, imagine your finance team reported the highest usage numbers last quarter but brought nothing reusable to the meeting. Your customer service team logged in less often, yet they showed a shared set of prompts that reduced the time it took to draft customer replies. The usage report would have pointed you to the wrong team.
How do you assess skill levels?
Compare what each team shows you to a skill scale, so you can see where people are today. The 10 Levels of AI Mastery works well for this because it separates basic familiarity with AI from the ability to build workflows and systems that other people use. Place each team at the level its examples support. Once every team is placed, you'll see where most of your people are grouped and which teams are ahead.
What do you do with the results?
Focus training on the largest group and set a clear goal: move them up one level. Then repeat the review in 90 days with the same three kinds of examples. If the new examples are more reusable than the old ones, the training worked. If they look the same, you've found that out in one quarter instead of a full year.
Each of these experiments answers a question many CEOs can't answer today. The first shows whether your executives and managers agree on the goal of AI. Measuring a single workflow gives you a real before and after result you can take to your board. Asking teams to show their work reveals what your people can actually do with AI, beyond how often they log in. Together, they replace guesses about AI with numbers from your own company, and those numbers are what you need to decide where to spend in 2027.
You can start all three AI experiments on your own. If you'd like a clear starting point before you begin, we built the AI Impact Analysis for that purpose. It's a custom 6-part report that shows your company's current AI skill level, what that level is costing you, and what to change first.
You answer 35 questions about your company size, how your workforce uses AI, your industry, and your priorities, and most people finish in 5 to 10 minutes. Your answers are turned into a report built from your own headcount and employee costs, with no generic benchmarks.
Here's what the report gives you:
A profile of your workforce mapped against the 10 Levels of AI Mastery, showing where your people are grouped today
Between 8 and 15 specific AI tools your employees could build, based on your industry, departments, and actual problems
The dollar value of the time your workforce is recovering with AI, along with a projection of what changes when you train them up one level
One specific recommended next step, explained in plain language and tied to your data
Your 2027 AI budget should be based on what's actually happening inside your company. Get your AI Impact Analysis and start the new year with your own numbers.

