Founder & Head of Design
Why HR's AI pilots keep stalling
You've run the pilots. You've sat through the demos. So why does leadership still ask what AI actually changed? Here's what's really going on, and what to do about it.
If you're in HR, you've probably had this exact moment: leadership asks for an AI update, and you genuinely have a lot to say. Three tools tested. A pilot with the recruiting team. A new policy doc. And yet, when you try to summarize what actually changed about how the function operates, the sentence trails off.
This happens because nobody told you that experimenting with AI and operating differently because of it are two separate jobs, and almost every framework, vendor pitch, and LinkedIn post only ever talks about the first one.
We've sat in enough of these conversations with HR leaders to recognize the pattern fast. We understand the frustration of doing what was asked and still seeing no results. That's a fair thing to be frustrated about. The good news is that it's also fixable, but not by running another pilot.
Here’s how to fix the problem.
The 3 problems you’re dealing with
A Gartner survey of nearly 3,000 employees, run in July 2025, found that 37% of people who could use AI in their work simply don't, because their co-workers aren't using it either. Adoption, it turns out, is social before it's technical.
Gartner's research points to where that hesitation usually comes from: executive urgency. Leadership pushes for fast rollout, the rollout happens before anyone's worked through what it actually means for how people do their jobs day to day, and the result is a tool that's technically available and practically ignored.
So if your team's adoption looks patchy, some people in deep, others barely touching the tools, that's not a sign you're doing it wrong. What's actually happening is that three specific, very ordinary problems are quietly absorbing all the effort that should be turning into results.
- You're solving the same problem more than once. Someone on your team builds something that works, a smarter intake process, a faster first-pass on resumes, and it stays with that team. Nobody hid it on purpose. There was just never a moment when it got handed off. So six months later, a different team starts again from a blank page, and the org pays for the same learning curve twice.
- AI gets bolted onto a process that was already broken. This one is sneaky because it feels productive. You take a workflow that everyone privately agrees is clunky, and you speed up the slowest part of it with a tool. It feels like progress for about a quarter. Then you notice the same complaints are still coming in, just faster. If the underlying steps were built on outdated assumptions, AI just executes them quickly.
- You and your technical partners are speaking two different languages. This is the one HR leaders bring up most, usually with some embarrassment, as if it reflects badly on them. It doesn't. You know the operational reality: what breaks down, what the actual users need, but turning that into a scoped technical brief isn't a skill HR was ever trained for. Meanwhile, your technical team can build almost anything you ask for, but without the operational context, they often build the wrong thing well. So you end up bringing in outside help to bridge the gap, and the thing that gets built belongs to the vendor, not to you.
These are structural gaps that won't close because someone tries harder, but because something changes about how the work happens.
What redesigning a workflow actually looks like
This is the part that gets skipped most often, because it's slower than it sounds: AI works best when it's applied to a workflow you've rethought.
Take hiring. The legacy version usually runs:
- job request → manual posting → resume flood → a two-week screening slog → human interview.
The instinct, when AI shows up, is to bolt a tool onto the slowest step, which is usually the screening, and call it transformation. But it isn't. It's the same workflow, with one stage running faster, and all the same assumptions still baked in underneath it.
An AI-native version of the same workflow looks structurally different from the start:
- a skills-gap map instead of a generic job post, automated sourcing built around that map, a deep profile match instead of a keyword filter, and then, still, a human interview.
The human judgment doesn't disappear. And should never do. It just moves to where it actually adds value, instead of being spent on the parts a machine can do as well or better. That redesign is the actual work. Adding AI on top is the easy part everyone does first, and it's also the part that almost never produces the result leadership is asking about.
Who should actually own this
Ask four executives who should own AI-driven workflow redesign and you'll get four confident, different answers: IT, HR, Ops, Product. All four answers are reasonable. All four are also, on their own, wrong.
Park it in IT, and it becomes procurement and licensing. Park it in HR, and it becomes training modules and compliance language. Park it in Operations, and it optimizes today's process instead of questioning whether it should exist in its current shape. Park it in Product, and only the product org changes while everyone else watches.
The more useful answer: the team that owns the workflow owns its redesign, not a central department sitting above it. What actually needs to exist centrally is a small cross-functional group with real executive sponsorship, because redesigning how people work is a change-management effort, and those don't survive without air cover from the top. HR, IT, Ops, and Product each bring something real to that group: skills, tooling, process knowledge, the actual surface where the work happens, but none of them runs it alone.
This is also, not coincidentally, the strongest case for bringing in outside facilitation. Cross-functional work is exactly what one internal department can't run neutrally. Whoever's "hosting" the conversation tends to unconsciously steer it toward their own priorities. A structured, facilitated process is built specifically to hold that space evenly, which is harder to do from inside any one team's chain of command.
What this looks like in practice
A few things worth retiring immediately, because they're actively working against you:
- Stop routing all the thinking through one central AI team. It creates a bottleneck and strips ownership from the people who'll actually have to live with the result.
- Stop asking "how can AI do this task?" and start asking "what should this workflow look like?" The first question optimizes a task. The second one questions whether the task should exist in its current form at all.
- Stop letting pilots live forever in pilot mode. AI that never reaches production is useless, and everyone in the room usually knows it.
- Stop overpromising fast transformation. Unrealistic timelines create disappointment, and disappointment creates resistance that's much harder to undo than slow, credible progress.
- Stop treating the human factor as a footnote. Fear, trust, manager capability, and reskilling carry as much weight as the technology itself. Often more.
What replaces all of that is a structured process that takes one validated problem, maps the workflow as it actually exists today, redesigns it before introducing AI, and brings the right cross-functional group together to do it in the same room, at the same time. It ends with one of three honest outcomes: scale it, iterate on it, or stop.
If there's a single, concrete signal that you're preparing for how work is actually changing, rather than just talking about it, it's this: pick one workflow. Hiring, onboarding, performance review, whichever one is causing the most friction right now. Redesign it end-to-end, not just the slowest step. Prove that AI can improve speed, quality, and human judgment at the same time, in one real process, with real people.
Let's talk about which workflow is worth redesigning first. Get in touch and we'll help you pick the right workflow, bring the right people into the same room, and turn experimentation into a decision worth acting on.
