Tal Dagan, CPO at Atera, an AI platform for IT management.
A kitchen can plate a beautiful dish in four minutes. If the order takes nine minutes to reach the line, that dish does not make the restaurant faster.
Enterprise AI has the same problem. McKinsey found that 80% of respondents say AI has improved their own productivity. Yet only 37% of organizations report any measurable impact on company-level earnings, essentially unchanged from a year earlier.
AI is making individuals faster. It’s not making organizations faster. I see this constantly in enterprise IT. A team deploys AI and then technicians resolve tickets faster. Then someone asks what changed at the business level, and nobody can produce an answer. There were gains, but they didn’t survive the trip from individual to organization.
Something is absorbing the value along the way. That something is friction: the stalled systems, broken handoffs and workarounds that sit between a person and the work they are trying to do. Most organizations aren’t measuring it, which means most organizations can’t see where their AI gains are going.
The Friction Tax Nobody Budgeted For
Research recently commissioned, based on a survey of 1,000 full-time employees and 500 C-suite and senior enterprise leaders at large U.S. organizations, found that 88% of employees lose time every day to stalled systems. That looks like a password reset that blocks off an entire morning. Access to a system you needed yesterday, still pending after two requests.
There’s the work that never gets done—61% of employees say they have delayed or avoided a key work action because the process was too broken to bother with.
That avoidance is a cost nobody planned for. The friction tax compounds as a stalled system creates an interruption, the interruption creates a workaround and the workaround becomes the new way of working. Eventually, the original problem disappears from view. Nobody fixes it because nobody can see it anymore.
The Problems That Never Reach You
In our survey, 37% of employees told us they hit a technology issue in the past three months and chose not to file a ticket. Half of them simply worked around it.
Leaders may sense the gap without being able to size it. Most believe their IT teams miss at least a quarter of what employees run into. The problem is what that means for AI. When over one-third of your friction never enters a system, any AI initiative built on ticket data is optimizing against an incomplete map.
Falling ticket volume can look like progress when it is actually a sign of disengagement. Fewer tickets might mean friction has been resolved. It might also mean employees have stopped reporting problems and are simply working around them.
AI can create the same illusion. If an automated fix keeps resolving the same failure without surfacing the pattern, the dashboard shows success while the underlying fault goes untouched.
Catch It Early Or Pay For It Twice
Picture two versions of the same Monday. In the first, a certificate expires, 40 people cannot log in and your team spends the morning working a queue that should never have formed. In the second, the expiry was flagged on Friday, someone caught it and the issue never disrupted employees. It’s the same technology and same team, but a different place in the lifecycle.
Most IT teams handle friction after it lands. The higher-value opportunity sits on both sides: catching recurring patterns before employees hit them and understanding afterward why they keep returning.
McKinsey found that nearly three-quarters of the organizations seeing the strongest financial returns from AI redesigned workflows rather than inserting AI into existing ones, compared with a quarter of everyone else. Our own research agrees. The most-cited barrier holding AI back was workflows not being ready for it.
AI can accelerate a workflow, but speed alone isn’t evidence that the organization is working better. What does matter is what happens to the friction around that workflow.
Activity Is Not Evidence
Once friction becomes the test, evaluation gets sharper. Three measures matter: prevention, resolution and recurrence.
Did the system catch a problem before employees ran into it? Did it shrink the gap between something breaking and someone fixing it? Did it stop the same issue from resurfacing?
Those questions tell you whether AI is changing the experience of work. Then ask what that reduction creates: how much work shifts from employees to AI, and how much operational capacity that frees up.
Adoption and utilization tell you whether people are using the technology. They don’t tell you how much work it is performing, how much workload has shifted from employees to AI or how much operational capacity it has created.
Not all friction is waste. Security controls, compliance requirements and multi-cloud environments introduce legitimate meta-work; an organization with none of it would be insecure or ruinously expensive. The key is separating purposeful friction from friction that persists because nobody addressed the process.
Make Friction The Test
Here are five moves that can make a difference:
Measure More Than Ticket Volume
Categorize the last three to six months of tickets and name your top 10 recurring issues.
Ask The People Who Stopped Filing
Friction that never becomes a ticket appears in no dashboard, but technicians and employees can name it. Make that conversation a regular review.
Find The Approval Chokepoints
Half the leaders we surveyed named approval delays the top drag on workflow.
Fix The Workflow Before You Layer AI On It
Readiness is a prerequisite.
Require Automation To Surface Patterns
If the same fault is fixed repeatedly, the system should make the pattern visible.
The enterprise AI conversation has moved past proving AI can do things. The question now is whether it is making the organization work better. The organizations proving AI value are deploying it where the work actually breaks.
Because the four-minute dish wasn’t the problem. It was everything happening around it.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?