If your employees are busy all day but the business still feels slow, you're probably experiencing operational drag.
You can see it in a scene that plays out in nearly every company: a customer request arrives by email. Someone enters it into the CRM. Someone else copies it into the project management system. It gets discussed in a status meeting, then summarized in a report. At every step, information is transferred, reformatted, and verified by people whose expertise would be better spent solving the customer's problem. Nothing is broken. Everything is slow.
Most organizations respond by buying technology—and in 2025 and 2026, that has meant AI. The results have been humbling. MIT researchers studying enterprise AI deployments found that roughly 95% of pilots produced no measurable financial return, despite tens of billions of dollars in investment. The tools weren't the problem. The processes underneath them were.
Most companies don't have an AI problem. They have an operational drag problem. AI creates measurable value—but only after the sources of friction have been identified and understood. In many cases, they're removed without any technology at all.
What Is Operational Drag?
Operational drag is the gap between how fast your people work and how fast your business moves.
It isn't failure. Failures get noticed, escalated, and fixed. Drag gets absorbed. It accumulates through dozens of small inefficiencies—searching for information, updating multiple systems with the same data, attending meetings to clarify status, waiting days for approvals that should take minutes—until it becomes "just how things work."
Asana's Anatomy of Work Index, drawn from surveys of more than 10,000 knowledge workers, found that roughly 60% of the workday goes to "work about work"—coordinating, searching, app-switching, and chasing status—rather than the skilled work people were hired to do. Harvard Business Review found the average digital worker toggles between applications nearly 1,200 times per day and spends about four hours a week simply reorienting after switches. Over a year, that's roughly five working weeks lost to changing contexts before a single unit of productive work is done.
The most damaging thing about drag is that it's invisible to leadership. It never appears on a financial statement or an operational dashboard. It shows up instead as delayed decisions, duplicated effort, inconsistent customer experiences, and teams that are perpetually busy but never able to move strategic initiatives forward.
→ What Is Operational Drag? A Complete Definition
Why AI Initiatives Fail: Drag Meets Automation
When MIT's NANDA initiative published its GenAI Divide report, the headline number—95% of enterprise AI pilots delivering no measurable P&L impact—dominated the coverage. The mechanisms behind the number received far less attention.
The researchers found the divide had little to do with model quality or regulation. It came down to integration: tools that don't adapt to real workflows stall, no matter how capable the underlying model is. They also found budgets were pointed at the wrong targets. More than half of enterprise GenAI spending flowed to sales and marketing—the visible, demo-friendly use cases—while the strongest returns came from unglamorous back-office operations: eliminating outsourced process work, cutting agency costs, and streamlining administration.
In other words, companies fund AI where it's visible, not where the drag lives.
The exact percentage is open to debate. The threshold for measurable financial impact was high, and many successful projects required 12 to 18 months to show results. That doesn't weaken the conclusion. It reinforces it. The failures were organizational, not technological. Automating a process nobody has bothered to understand doesn't remove the drag. It accelerates it.
→ Why 95% of AI Pilots Fail — And What the 5% Do Differently
The Verification Tax: When AI Creates Its Own Drag
Here's the uncomfortable finding that most AI vendors won't put in a deck: sometimes AI makes skilled people slower.
A randomized controlled trial by METR gave experienced open-source developers access to frontier AI coding tools. The developers completed tasks more slowly while believing the AI was making them meaningfully faster. The gap came from invisible overhead: validating, correcting, and integrating AI output takes time that never gets attributed to the tool.
Telemetry from Faros AI across more than 10,000 developers tells the organizational version of the same story. Teams with high AI adoption completed 21% more tasks and merged nearly double the pull requests, but code review time rose 91% and company-level delivery metrics stayed flat. The AI accelerated one node in the process, and the drag simply pooled at the next constraint.
AI also creates its own drag. Researchers at BetterUp Labs and Stanford documented what they call "workslop"—polished-looking, low-quality AI output that shifts the real work downstream. Forty-one percent of workers reported receiving it, costing roughly two hours of rework per instance while eroding trust between colleagues. Deploy AI on top of a slow process and you don't remove friction. You breed a new species of it.
→ Coming Soon: The Verification Tax: The Hidden Cost of Managing AI Output
The Operational Drag Audit: Diagnose Before You Automate
If drag is invisible on dashboards, you have to go looking for it deliberately. That's what an Operational Drag Audit does: it maps where work slows down, where information gets stuck, and where skilled people repeatedly perform low-value tasks.
The audit examines questions like:
-
Where does the same information get entered more than once?
-
Which approvals take days but require minutes of actual judgment?
-
What meetings exist primarily to transfer status that a system should surface automatically?
-
Where do employees maintain private workarounds—spreadsheets, personal AI tools, side channels—because the official process is slower than the unofficial one?
That last signal deserves special attention. MIT found that while only about 40% of companies had official AI subscriptions, roughly 90% of employees were already using personal AI tools daily. This "shadow AI economy" is your workforce telling you where the drag is. They've already found it. They're routing around it every day.
The output of the audit isn't a technology recommendation. It's a friction map—a prioritized inventory of where the organization loses time, ranked by cost and fixability. Only then does the question of AI become answerable.
→ Download the Operational Drag Audit: A Step-by-Step Guide
Remove, Simplify, Standardize, Automate
Once the friction map exists, the fix follows a sequence. The sequence is the point.
Remove. The cheapest process to improve is the one you stop doing. Every friction map contains reports nobody reads, approvals that exist for historical reasons, and meetings that have outlived their purpose. No technology required.
Simplify. For work that must exist, reduce the steps, handoffs, and decision points before touching any tool. A five-step process automated is better than a twelve-step process automated—and dramatically cheaper to maintain.
Standardize. Automation requires predictability. If the same request is handled three different ways by three different people, no system—AI or otherwise—can reliably take it over. Standardization is the unglamorous work that makes everything after it possible.
Automate. Only now does AI enter. And at this stage, it works because it's being applied to processes that are understood, streamlined, and consistent. This is what the successful organizations in MIT's research had in common: they automated operations they had already designed instead of hoping technology would impose design on chaos.
Skip the sequence and you get the familiar outcome: expensive tools layered on top of confused processes, producing faster confusion.
→ Coming Soon: Remove, Simplify, Standardize, Automate: The Operational Design Sequence
Where AI Creates Business Value
None of this is an argument against AI. It's an argument about order of operations. When the drag has been mapped and the sequence followed, AI delivers exactly where the research says it does:
-
Back-office and administrative work—the highest-return category in MIT's data and the least funded.
-
Information retrieval and synthesis—attacking the hour per day workers lose searching across fragmented systems.
-
Structured, repeatable processes—the ones that survived Remove, Simplify, and Standardize.
-
Drafting and first-pass work with clear review standards—where verification is designed in, not discovered later.
The pattern across every success story is the same: the organization understood its own operations first and pointed AI at friction it had already named. That's operational design—human-led, AI-amplified. The technology amplifies whatever it's given. Give it drag, and it amplifies drag. Give it a well-designed operation, and it compounds the advantage.
Start With the Friction, Not the Tool
The question most leadership teams are asking—how can AI make us more productive?—is the wrong first question.
The right one is: Where is work slowing down? Answer that, and the AI decisions largely make themselves.
Download the Operational Drag Audit Checklist, a practical framework for identifying where work slows down before deciding where AI belongs.
Need another set of eyes? Talk with Infernal Machinery.
FAQ
What is operational drag?
Operational drag is the accumulated friction of small inefficiencies—duplicate data entry, information searches, status meetings, and slow approvals—that consumes employee time without appearing on a financial statement. It's the gap between how fast your people work and how fast your business moves.
Why do most AI pilots fail?
MIT research attributes the high failure rate primarily to integration and process problems rather than model quality. Organizations automate processes they haven't first understood, simplified, or standardized.
What is "work about work"?
Asana's term for coordination overhead—communicating about work, searching for information, switching apps, and chasing status—which its research across more than 10,000 knowledge workers found consumes roughly 60% of the workday.
What is workslop?
Polished-looking, low-quality AI output that shifts real work downstream to whoever receives it. BetterUp Labs and Stanford researchers found 41% of workers have encountered it, costing roughly two hours of rework per instance.
Should we fix processes before implementing AI?
Yes. Remove. Simplify. Standardize. Automate. AI applied to a well-designed process compounds its advantages. Applied to a broken one, it compounds the problem.