Most AI adoption research treats "struggling" as a fixable technology problem. In regulated industries, it's a workflow design problem — and the fix looks completely different.
A new report says 79% of enterprises face significant challenges with AI adoption in 2026. That number is getting passed around in slide decks and LinkedIn posts like it explains something.
It doesn't. Not on its own.
The implicit read is: companies are trying to adopt AI and running into friction. Fix the friction, get the adoption. That's the vendor narrative, and it's too clean.
For operators in regulated industries — medical device, financial services, legal, manufacturing — the challenge isn't friction in the conventional sense. It's that AI gets deployed into environments where the cost of a bad output isn't just a bad user experience. It's a 483, a consent order, a product liability claim, a malpractice exposure.
The 21% who aren't struggling? A lot of them work in industries where the blast radius of an AI error is manageable. That's not most of our clients.
Here's what we see consistently: companies that have "failed" at AI adoption didn't fail because their staff couldn't learn a new tool. They failed because they deployed AI into workflows that weren't designed to support it.
Consider a common scenario in medical device. A quality team wants to use an LLM to speed up complaint triage — summarizing incoming customer feedback and flagging potential MDR-reportable events. Sounds straightforward. But the workflow around that task involves:
You can have the best AI tool in the world. If the workflow isn't rebuilt around it — with all of that accounted for — you get shelfware. Or worse, you get a tool that's in production but operating outside your quality system.
That's not an AI problem. That's a workflow design problem.
The report distinguishes between companies achieving measurable outcomes and those stuck in what it calls "performative AI" — visible activity, no real return. Pilots that never scale. Demos that impress leadership but don't change how work actually gets done.
We'd go further: in regulated industries, performative AI carries a specific risk that general enterprises don't face. When a compliance-adjacent workflow has an AI layer bolted onto it — without proper validation, change control, or governance documentation — you now have a liability that isn't just operational. It's regulatory.
The audit question isn't "did you use AI?" It's "can you demonstrate that this AI-assisted output met your documented quality standards, and who was responsible for that determination?" Most companies running performative AI cannot answer that question.
Based on what we see in the field, the companies making AI work in regulated environments share a few characteristics. They're not necessarily the most sophisticated technically.
They started with workflow design, not tool selection. Before picking software, they mapped the existing process in detail — every handoff, every approval, every documentation requirement. Then they identified where AI could remove friction without introducing new risk.
They involved compliance and quality early. Not as a gate at the end, but as design partners. The companies that do this get slower starts and much faster scale. The companies that skip this step build things twice.
They defined "good output" before deployment. In a regulated context, this means establishing acceptance criteria for AI-generated content — the same way you'd validate any process output. What does a correct AI-assisted triage flag look like? What's the human review step? Where does the decision authority actually sit?
They treated AI like any other process change. That means change control. That means training records. That means documented rationale for why the workflow was modified and what risks were assessed. It's slower. It's also the only way to build something that survives an inspection.
If you're evaluating whether an AI workflow is ready for production in a regulated environment, start here:
1. Can you document the decision logic? If a regulator asked you to explain how an AI-assisted decision was made, could you produce that documentation? If not, you're not ready.
2. Is there a defined human checkpoint? Autonomous AI decisions in regulated workflows are almost always a mistake. Where is the human review, and is that reviewer equipped to actually evaluate the AI's output — not just rubber-stamp it?
3. Does your QMS or compliance infrastructure support this workflow as designed? If the answer is no, you have two choices: redesign the workflow to fit your infrastructure, or upgrade the infrastructure first. Deploying around it is how you build technical and regulatory debt simultaneously.
The 79% struggling with AI adoption aren't failing because the technology is immature. Most of them are failing because they're trying to layer new capability onto workflows that were never designed to support it — and in regulated industries, that's not a recoverable mistake you can patch later.
Build the workflow first. The AI fills in where it actually fits.
Dealing with a similar challenge?
We work with mid-market companies in regulated industries to build AI workflows that actually hold up.
Let's TalkSean Cummings
Founder of Laminar Flow Analytics. Specializes in AI workflow automation for regulated industries — medical device, financial services, and complex logistics operations.