When the people at the top aren't on the same page about AI, the bill shows up everywhere else: unused licenses, unfinished training, failed pilots, and hours your team never gets back.
In a Harvard Business Review Analytic Services survey, 76% of small and midsize business respondents said they expect their organization to increase its use of AI in the next 12 months, yet only 19% feel their organization is highly prepared to recruit or develop the AI skills it needs. The distance between those two numbers is where the money disappears.
In our work with small businesses, that cost tends to show up in four places, and each one has a fix you can start on this quarter:
A definition of AI adoption that nobody agreed on, so nobody knows when the company gets there
A pilot project funded before anyone inside the company could support it
Training that employees start and never finish
AI committees staffed with the wrong mix of people
Nobody Agreed on What AI Adoption Means
If your leadership team can't describe what "adopted" looks like, your employees have no finish line to reach
Ask three people on your leadership team what AI adoption means, and you'll probably hear three different answers. One says everyone has a login. Another says people finished the course. A third says employees are building things with AI, though nobody can say what those things should be or how many would be enough.
Each answer sounds reasonable, and that's the trouble. When leaders never settle on one definition, every employee is chasing a finish line that doesn't exist. People can't tell whether they're behind or ahead. Managers can't tell who needs help. Six months in, nobody can say whether the company is any closer, because nobody ever decided where "there" was.
The cost keeps building, and without a target, most people stop at the first comfortable place, which usually means asking AI questions the same way they'd type them into Google. In our assessments over the past 3.5 years, 96% of AI users never get past Level 3 on the 10 Levels of AI Mastery. At Level 3, AI is a faster search engine that writes longer answers.
Your people aren't lazy, and they aren't failing you. The software they've used for decades had menus and buttons that showed them what was possible. A blank chat box shows them nothing, so they have no way to see what's above Level 3. Left on their own, it takes a person 2-3 years to reach Level 6 in our experience, if they get there at all.
A shared definition gives everyone something to aim for. Here is the one we use with clients: anyone in the organization who uses a computer for more than 30% of the workday reaches Level 5 or 6.
Once leadership agrees on that definition, three groups finally know what they're working toward:
Employees get a personal target. "You're at Level 3, and we need you at Level 6 by the end of the quarter" is an expectation a person can act on.
Managers get a way to see who is on track. They can spot the person who has stalled and step in before another month passes.
Leadership gets a result it can report. The board hears how many people have reached the target instead of how many licenses were purchased.
The definition only works if every leader repeats it the same way. Agree on it before the next rollout announcement, and put it in writing where every employee can see it.
The Pilot That Failed Before It Started
Funding an AI project before your people understand AI puts your money in someone else's hands
Here's a story we hear often. A leadership team feels pressure to show progress on AI, so it picks one impressive project, hires a developer, and writes a check. Maybe it's a customer service assistant or a tool that drafts proposals. Everyone is excited at the kickoff meeting.
Then the questions start. The developer asks what data the tool should draw from, what rules it has to follow, and who will test it. Nobody inside the company knows how to answer, because nobody inside the company has used AI beyond asking it questions. The leaders who approved the budget can't tell whether the developer's estimates make sense or whether the promised results are realistic. They're paying for something they can't evaluate.
The real damage arrives later. AI models now change every few weeks, and when one shifts, a tool that worked on Monday can behave differently by Friday. When that happens in a company with little internal AI knowledge, the same painful chain of events tends to follow:
The tool starts producing strange or inconsistent results.
Employees blame the vendor, because they don't understand what changed.
The vendor has nobody inside the company who can help diagnose the problem.
Trust breaks down, the project stalls, and leadership decides AI "didn't work for us."
The technology usually was never the problem. The company simply didn't have enough people who understood AI to support the project. We call this a density-of-knowledge problem, and we've found it in nearly every organization we've assessed.
That's why we recommend putting pilots near the end of an AI rollout instead of the beginning. First, leaders and employees learn to use AI in their own work. Once enough people reach Level 5 or 6, they can tell a developer exactly what they need, spot a promise that doesn't hold up, and adjust a tool when a model changes. Pilots built on that foundation are far more likely to succeed, and the best ideas for them often come from employees who started by solving small problems in their own jobs.
The change starts with the leadership team. In our executive workshops, every leader builds a small AI tool for their own work, using ideas tailored to their role. We're regularly surprised by what a CEO builds when they're the one doing it instead of delegating it. Leaders who have built something themselves ask better questions before they sign the next check.
The Training Nobody Finished
Buying courses is easy, but without someone responsible for follow-through, most employees never reach the last lesson
Most AI rollouts follow the same script: leadership buys the tools, adds a training course, sends the link to every employee, and waits for results. It feels like a complete plan, and that's exactly why it hurts so much when nothing changes.
We've seen this in our own work. One company bought 10 seats in our training. After 12 months, 2 people had started and nobody had finished. That isn't unusual. In our experience, when there's no process behind the training, around 80% of employees won't complete it, and the reasons are nearly always the same:
Their current work comes with deadlines and customers attached.
Many are suspicious of what AI means for their jobs.
The course feels like one more task on an already full plate.
Some assume AI is just a better version of Google, so a course seems unnecessary.
Even the employees who finish may not be able to use what they learned. A completion certificate means someone passed a quiz. It doesn't mean they can prepare a report faster or take on work they couldn't handle before.
Here's where leadership misalignment does its damage. When every leader assumes training is someone else's responsibility, nobody sets expectations, checks progress, or notices when an employee quietly stops. The company paid for the course, and the course stays unopened in people's inboxes.
A training process fixes this, and it needs a few specific pieces:
A starting point for each person. An assessment shows where every employee stands on the 10 Levels of AI Mastery.
A target and a deadline. Each person knows the level they need to reach and when.
Training mapped to that target. Your existing courses are matched to the levels they're meant to build.
A project to apply it to. Employees build a tool that saves them time on a task they already do, with role-specific ideas for people who don't know where to start.
A manager who follows up. Someone checks who has stalled and helps them get going again.
When each of those pieces has an owner, employees finally have a reason to finish the training, because they can see how it will change their own workday.
The Wrong People in the Room
How your team members prefer to work decides whether your AI committee gets anything done
When it's time to form an AI committee, most leadership teams reach for the obvious picks. The IT manager goes on it. So does whoever seems good with technology, plus a couple of department heads who had room on their calendars. Then the meetings start, and months pass without much to show for them.
The people on that committee are usually smart and well-meaning, so the stall comes from somewhere else: the mix of how they prefer to work.
Everyone has a default way of responding to new work, and we describe four preferred working styles:
Doers want to get it done. Hand them a clear task and the steps, and they move fast.
Administrators want structure. They put rules and quality standards in place before anything goes out the door.
Innovators look for a better way. They're often experimenting with AI already, with or without permission.
Connectors bring people together. Once they believe in something, they get everyone else on board.
Each person uses all four at different times but leans toward one. These styles describe how someone works, never how smart or valuable they are.
Now picture a committee missing one of them. A group made up mostly of Administrators will write careful policies and debate them for months. Put people without Innovators in front of an assignment like "build a tool that saves three hours a week," and they'll struggle to come up with a single idea. And a committee with no Connectors may build something useful that nobody else in the company ever hears about.
We saw the effect of the mix at a community college bootcamp. Of the 32 participants who completed the assessment, 13 were Administrators, 13 were Connectors, 4 were Doers, and only 2 were Innovators. Knowing that ahead of time, we gave each person a sheet of 5-7 project ideas tailored to their job. About 85% built from the list, and most of the rest said it sparked an idea of their own.
Over two days of 4-hour sessions, 91% of participants built something, and the group completed 22 approved projects. Based on the time savings participants recorded and the college's reviewers approved, those projects add up to an estimated 3,811 hours a year. The biggest skeptic in the room on day one showed up at 7:30 the next morning for an 8:00 start and finished at the top of the progress board.
Before you choose your next AI committee, find out how each candidate prefers to work. Make sure the group includes someone to generate ideas, organize them, get the work done, and bring the rest of the company along.
Getting Your Leadership Team on the Same Page About AI
The four costs of leadership misalignment all trace back to decisions your leaders can make together
Remember the question from the start of this post: what is the business getting from AI? When leaders aren't on the same page, that question has no good answer, and the cost spreads into every corner of the company. Employees chase a goal nobody defined. A pilot drains the budget because no one inside can support it. Courses go unfinished, and committees meet for months without producing anything.
None of these problems requires new technology to fix. Each one comes down to a decision your leadership team can make in the same room:
Agree on a definition of AI adoption and a target level for every employee.
Build knowledge across the company before funding a pilot.
Give training an owner, a deadline, and a project to apply it to.
Staff your AI committee based on how people prefer to work, not only on title or availability.
Once those decisions are made, the answer to the owner's question gets specific. Leaders can point to tasks that take less time, employees who reached their targets, and hours that people have put back into serving customers and taking on new work.
When you're ready to see what's at stake in your own company, start with an AI Impact Analysis. You answer 35 questions about your company, your workforce, and how your people use AI today. Within 24 hours, you’ll receive a 6-part report showing where your employees stand on the 10 Levels of AI Mastery, the dollar value of the time they're recapturing now, and what changes when they're trained up one level. The math uses your own headcount and employee costs, so the result reflects your business rather than a generic benchmark, and the report ends with one specific recommendation for what to do first.