Every CEO is writing press releases about how their company has rolled out AI. Enterprise spending is skyrocketing. But we’re seeing far fewer press releases about big AI wins.
Across hundreds of organizations we’ve worked with, the story is the same: big investments, low impact. Why? Because getting ROI from AI is directly tied to how often and deeply your people use it.
At Section, our data shows that to see meaningful returns, 80% of your workforce needs to use AI to its fullest capabilities every week. That’s the threshold. Below that, you don’t have an AI-powered company.
So what’s holding companies back from reaching the threshold? We see ten major barriers, but there are four that show up almost everywhere.
1. There’s no “Why of AI”
When we begin a partnership with a new organization, we always start by asking employees a simple question: What’s your company’s AI strategy?
More than half tell us they’re “developing one.” Another 15% aren’t sure if one even exists. And only 23% say there’s a formal strategy – even in organizations that have rolled out multiple large language models (LLMs).
That gap between leadership intent and employee understanding is where adoption dies.
When employees don’t know why they’re being asked to use AI, they fill in the blanks. We hear things like, “We shouldn’t be using AI just for the sake of using it.”
People hate change, which makes AI emotional. And without clear rationale, they’ll interpret your push for AI as a threat, not an opportunity.
The mandate: Make AI mission-critical
Companies that achieve true adoption reframe AI as imperative to their business – not a side project or a tech trend, but a requirement for survival.
At Section, we help clients create what we call an AI Manifesto: a living, breathing document that connects AI directly to your core business goals. Because “being more productive” isn’t enough. That’s the CFO’s why, not the workforce’s. Employees rally around purpose, not vague efficiencies.
Look at Duolingo. Their Why of AI isn’t productivity – it’s scale. Without AI, they can’t serve all the languages they need to in order to meet consumer demand. AI allows them to grow the business quickly and exponentially. Their customers are happier and the business is more valuable. That’s a reason employees can see the merit in.
Or take a hospital system we work with: “AI is the only way we can meet our mandate to treat every single patient with excellence.”
That’s the level of clarity it takes to drive adoption.
2. You haven’t achieved true access
Most executives believe they’ve cleared the access hurdle because they’ve approved an LLM plan. But when we ask employees, we get a different story.
Among companies that approve of AI use, only 29% of employees say they’ve been given access to an enterprise LLM.
What’s happening is messy: Some employees get Copilot licenses, others don’t. Some lose access without explanation. Others have access only because they personally sought it out and requested it.
We see two common scenarios:
- The hamstrung rollout: Everyone gets AI, but it’s a stripped-down version missing key capabilities like custom GPTs. Beginners think AI doesn’t work and bounce off. Experts get frustrated and use shadow AI to access more advanced features
- The restricted rollout: Only a few teams get access. Beginners get anxious because they’re not among them. Experts go underground and find their own tools. Either way, adoption stalls.
The mandate: Universal, powerful access
If you want ROI, half measures won’t work. Your mandate should be clear:
- Every knowledge worker gets access and coaching to an enterprise LLM
- Turn on the most advanced capabilities possible – especially custom GPTs and reasoning models.
- Enable fast, efficient approvals for department-specific tools that accelerate functional outcomes.
3. Your AI policy Is hampering experimentation
When we ask employees why they’re not using AI more, the number-one reason – above hallucinations, job fears, or lack of skills – is: “I’m worried about data security or privacy”.
This tells us most companies’ AI policies are built for compliance, not adoption.
Only 9% of employees say their company has a comprehensive AI policy that covers risks. Nearly 70% say they’re not sure if one even exists.
That uncertainty breeds silence. People experiment privately or not at all. They stay at beginner-level use cases – rewriting emails, summarizing documents – and never move up to transformative workflows.
People need guardrails in order to feel safe taking risks. If your AI policy is a CYA document that doesn’t acknowledge key considerations, employees aren’t going to take the chance.
The mandate: Write policies that encourage play
AI is a sandbox – and a blank page. You need to know what to use it for and you can’t discover high-value use cases without experimentation.
So publish simple, specific, easy-to-find guardrails that make people feel safe to play:
- Use plain English, not legalese
- Don’t just say what not to do – explain why
- Outline clear governance processes the enable experimentation
The best companies treat governance as an enabler, not a restriction.
4. Employees lack use cases
Even in advanced companies, 45% of employees say they don’t have a specific AI use case that meaningfully improves their work. And among those who do, most are surface-level: rewriting emails, quick fixes for code, using AI instead of Google etc.
These are fine starting points, but they won’t deliver ROI.
Compare that to a financial services client, where one analyst used Gemini to draft research briefs and query transcripts in GPT. What used to take four market analysts doing a deep dive could now be done in 48 hours. That’s the kind of workflow innovation that scales – and drives measurable returns.
But because AI is emotional – and a blank page – employees need to be guided to value. Their aha moments have to be orchestrated. And without structured training that uncovers these use cases, they’re going to continue using AI in ways that don't benefit the business and barely benefit them.
The mandate: Build a culture of sustained use
This journey takes time. It might take 6-12 months to move from “no use cases” to “advanced, repeatable workflows.”
Your job as an AI leader is to shorten that curve. Facilitate personalized training that surfaces aha moments for the employee (like ProfAI). Empower champions to model strong use. Showcase real examples of what’s working right now. Host lunch-and-learns and prompt contests.
Over time, your goal is to create the conditions where employees tell you the transformative AI use cases, not the other way around.
The rest of the iceberg
Beyond these four barriers, six more often stand in the way of enterprise ROI:
- Transformation fatigue – employees are still recovering from the last big change.
- You’re not talking about AI (and its failures) enough – silence tells people “return to your regular work.”
- You haven’t empowered middle managers – they’re your linchpins for translating strategy into action.
- You started with tools, not workflows – pilots fail because they’re not mapped to real business value.
- You haven’t defined acceptable accuracy – many prototypes stall because they can’t meet quality thresholds.
- You don’t understand employees’ real objections – from fear of errors to environmental concerns, hidden resistance runs deep and has to be acknowledged.
Basically, the companies truly seeing AI ROI didn’t get there by rolling out another piece of software. They got there by rethinking change management, access, policy, and purpose.
They didn’t ask, “How do we get people to use AI?” They asked, “How can AI radically change how our business works?”
I did a full presentation on this. You can get the slides here.
