48% of Executives Say AI Has Been a Massive Disappointment. Here’s What the Other 52% Are Doing Differently

The number nobody wants to put in a slide deck

In May 2026, software company Writer published its annual enterprise AI adoption survey. The headline finding was uncomfortable: 48% of executives described their AI adoption as a “massive disappointment.” Only 29% reported significant ROI from generative AI, and only 23% from AI agents. ^[1]

The Deloitte State of AI in the Enterprise 2026 report landed in parallel with a related finding: a widening gap between the companies extracting real value from AI and those who have invested the same money and gotten back a fancy chatbot nobody uses. ^[2]

The honest reading: two years in, the average AI rollout is underperforming. But “average” is hiding two very different stories. There is a 52% who are quietly getting it right, and the 48% who are not. The differences are not about how much they spent or which model they picked. They are about how they actually rolled it out.

Habit 1: They pick one workflow at a time, not “AI everywhere”

The disappointed 48% almost all share one origin story. They started with “AI strategy” instead of “AI workflow.” Strategy decks, vendor evaluations, enterprise licences for thousands of seats, internal launches with all-hands meetings. Then six months later, the dashboard says 12% of employees ever logged in.

The 52% who got it right started with one workflow. One. Usually something painful and specific. “Our procurement team takes four hours to summarise each supplier proposal. We are going to use AI to get that down to 40 minutes.”

Then they actually measured whether it worked. Then they did the next one. ^[3]

What this looks like in practice: a marketing director picks “draft first-pass brief responses” as their team’s one workflow, builds a Claude or ChatGPT project around it, trains the four people who do briefs, and watches the metric for a month. Once that’s working, they add the next workflow. Slow looks slow until you compare it to the alternative, where six months in nobody has changed how they actually work.

The reason this works is mathematical, not philosophical. Real adoption is a series of habit changes. Habits take time. Layering a second habit change on top of the first one before the first has fully formed is how you get a workforce that has tried five AI tools and stuck with none of them. The companies that look “behind” in tool count are usually ahead in actual changed behaviour.

I have seen a 12-person sales team out-deliver a 200-person enterprise rollout on AI-assisted outreach, simply because the smaller team did one workflow well for three months before adding the second. The bigger team rolled out four workflows on day one. By month three the smaller team was generating real meetings. The bigger team was running a survey to find out why adoption was flat. ^[5]

Habit 2: They train people before they buy tools

The biggest barrier to enterprise AI integration, according to nearly every survey published in the last year, is the skills gap. ^[4] The disappointed companies treat this as a problem to solve later. The successful ones treat it as a prerequisite.

Here is the pattern that keeps showing up. Companies seeing real returns spend roughly 30% of their AI budget on training and change management, not tools. ^[2] The disappointed ones spend less than 5% and assume people will “figure it out.” People do not figure it out. They use AI to draft a single email, decide it is not as useful as everyone said, and stop opening the tab.

What training actually means here is not a 90-minute webinar. It is hands-on, role-specific practice with the workflows that team actually has to do. A finance team needs different practice than a sales team. Generic AI training is theatre. Specific AI training changes behaviour.

If you are leading a team and you have a Claude or Copilot licence sitting underused, the fix is not a better tool. The fix is two hours a week, for a month, of supervised practice on real work. We unpack this approach in detail in our guide to why corporate AI training fails.

One specific format that works: pair a confident AI user with one or two skeptics for a weekly 60-minute working session. They bring real tasks. The skeptics drive the AI. The confident user coaches. After four sessions, the skeptics are usually no longer skeptics. After eight, they are training the next pair. Cost: zero. Impact: more than most training budgets.

Habit 3: They measure changed behaviour, not usage

The most common metric in disappointed AI rollouts: “AI usage.” How many people logged in this week. How many prompts were sent. How many tokens consumed.

This metric is useless. It rewards activity, not outcomes. You can have 90% of your team using AI daily and still see zero impact on revenue, cycle time, or quality. You can also have only 20% of your team using it heavily and see dramatic gains.

The companies seeing ROI measure something different. They pick a specific workflow, define what “better” looks like, and track that. Here are real examples from the Deloitte report and adjacent research: ^[2,5]

• “Time from RFP received to first-draft response” (procurement)
• “Average customer support resolution time” (CX)
• “Number of qualified leads created per SDR per week” (sales)
• “Time spent on monthly close” (finance)
• “Documents reviewed per lawyer per day” (legal)

If you cannot point at a metric that has moved because of your AI rollout, you are still in the disappointment camp regardless of how high your usage numbers look.

Honest aside: the executives I have talked to who run the most successful AI programs all say the same thing about metrics. They picked one. They committed to it for a quarter. They did not change it even when the early numbers looked weird. Discipline on a single metric beats a dashboard with 15. If you cannot defend the metric you chose in one sentence, you have not really chosen one.

Habit 4: They let teams own their own AI use, with guardrails

One of the more surprising findings in the 2026 research is how often top-down AI rollouts underperform bottom-up ones. ^[6]

When the CIO picks the tool, defines the workflows, and pushes it down, adoption stays low. When individual teams identify their own pain points, pick the workflow, and ask for the tool to support it, adoption is much higher and outcomes are better.

The companies getting this right do something specific. They set clear guardrails at the top (which data can be used, which tools are approved, what compliance requires) and then they let individual managers run AI experiments inside those rails without asking permission for every one. The result is dozens of small wins instead of one large rollout that nobody loves.

This is the opposite of what most enterprises did in 2024 and 2025. The early playbook was “central team picks Microsoft Copilot, deploys to everyone.” Two years later, the data is clear: that approach left value on the table.

Habit 5: They treat AI like a colleague who needs onboarding

The disappointed companies talk about AI as “a tool you press.” The companies seeing ROI talk about it as “a colleague you bring up to speed.” This is more than a vibe shift. It changes what you actually do.

When you onboard a new colleague, you tell them what good work looks like. You share the brand voice document, the customer list, the company history, the templates. You explain the why behind the rules. You give them sample outputs. You give them feedback when their first draft misses.

The 52% who get AI working do exactly this. Every Claude Project, every custom GPT, every Copilot agent inside their company has a clear “role description,” a context library (brand guidelines, examples, past work), and a feedback loop. The first draft is rarely perfect. By draft five, it is.

The 48% who are disappointed treat AI like a vending machine. Insert prompt, expect result. The result is bland, generic, and not on-brand. They blame the AI. The AI is fine. The onboarding is missing.

If you want a concrete starting point, we walk through the practical version of this in our 2026 prompt engineering patterns guide.

What to do this week if you suspect you are in the 48%

Three honest questions to put on a Monday morning meeting agenda.

Question 1: name the one workflow your team uses AI for, and the metric that has moved because of it. If nobody can answer this in 30 seconds, you do not have an AI strategy yet. You have an AI subscription.

Question 2: how much of your last year’s AI budget went to training your people versus buying tools? If it is under 20%, that is where the disappointment is coming from. Reallocate now, not next budget cycle.

Question 3: who on your team has shipped a real AI workflow they are proud of? If the answer is nobody, your starting point is finding the one person on your team who already loves AI and giving them official time to lead. Quiet champions exist on almost every team. Find yours.

The companies in the 52% did not get there because they are smarter or better funded. They got there because they did fewer things, taught their people properly, measured what mattered, and stopped treating AI like a magic button. Every one of those is something you can start this week. None require approval from finance.

One last thought. The narrative around AI in 2026 is still dominated by big platform announcements and benchmark numbers. The story that actually matters for your career is much more boring: the people who quietly learn to integrate AI into the way they already work are going to spend the next five years pulling ahead of the people who treat it as a side project. The disappointment data is loud right now. The compounding gains are quiet. Choose accordingly.

Sana Mian, Co-Founder, Future Factors AI

Sana is an AI educator and learning designer specialising in making complex ideas stick for non-technical professionals. She has trained 2,000+ learners across corporate teams, bootcamps, and keynote stages. Future Factors offers AI Bootcamps, Corporate Workshops, and Speaking & Consulting for businesses ready to adopt AI without the overwhelm.

More about Sana →