Why workplace AI adoption is an organizational design problem

By Fynn Feldpausch, VP of Engineering, Haiilo

There’s a story I keep hearing.

A company invests in AI. They run pilots. The demos are impressive. The board is excited. The vendor deck promises transformation. For a few weeks, maybe even a few months, there’s momentum.

And then… nothing really changes.

Usage plateaus. Teams revert to old workflows. The “AI strategy” becomes a slide in a quarterly update. Eventually someone says, “The technology just wasn’t mature enough.”

I don’t buy that.

After years building and scaling software platforms – and now leading engineering at Haiilo – I’ve come to a different conclusion:

AI adoption is rarely a capability problem. It’s almost always an organizational design problem.

And until we treat it that way, we’ll keep mistaking weak adoption for weak technology.

Read: The 2026 guide to AI in internal comms

‍

No one owns AI end to end

In most organizations, AI sits in a gray zone.

• IT evaluates vendors and manages security risk
• Innovation teams run pilots
• Business units are “encouraged” to experiment
• Legal sets guardrails
• Executives talk about strategy

But who owns AI adoption end to end?

In most companies, the honest answer is “no one”.

And that matters, because AI without ownership is just technical debt.

When no one owns the outcome, all you get is drift. Tools deployed without clear success criteria. Pilots run without accountability. Teams carrying risk without authority. And eventually, AI is something that “exists”, but doesn’t fundamentally change how work happens.

See: How AI works at Haiilo

‍

Adoption follows accountability, not capability

It’s assumed that if AI is powerful enough, people will use it. That’s not how organizations work. We all know that.

I’ve seen incredibly capable AI systems sit untouched because no leader was accountable for changing workflows around them. And I’ve seen simpler automation tools transform teams because someone was measured on making them stick.

This is where organizational design can undermine ambition. Companies say, “We want to be AI-first.” But they don’t change incentives. They don’t redefine decision rights. They don’t reallocate ownership. And so AI remains a side project.

The platform team paradox

Let’s talk about platform teams for a minute.

In many companies, platform or IT teams are tasked with “enabling AI.” They evaluate vendors, ensure compliance, integrate systems, manage infrastructure.

They carry the risk. But they don’t control business priorities. They don’t define workflows. They don’t own performance metrics in sales, HR, operations, or customer service.

They’re responsible for safe deployment, but not empowered to drive adoption.

This creates a structural paradox:

• If AI fails, it’s seen as a technology failure
• If adoption stalls, it’s framed as a change management issue
• But no one sits at the intersection with clear authority and accountability

Over time, platform teams become conservative, not because they lack vision, but because they absorb downside risk without upside control.

And we wonder why AI momentum can be so slow…

Read: How to boost adoption of your tools

‍

Bad incentives block adoption

Organizational design shows up most clearly in incentives.

Imagine a customer support team being told to “leverage AI to improve productivity.” But their performance metrics still reward ticket volume, not resolution quality. Or speed, not experimentation.

Or a marketing team is asked to use AI-generated content, but they’re judged harshly for any mistake or tone inconsistency.

Or a compliance team is evaluated solely on risk minimization.

Each team behaves rationally according to its incentives. But collectively, those incentives block adoption.

AI often requires:

• Short-term productivity dips
• Workflow redesign
• Experimentation
• Cross-functional coordination

If incentives reward stability over iteration, AI will stall.

‍

System boundaries matter more than features

One of the biggest adoption killers is unclear system boundaries.

Where does AI sit in your operating model?

Is it:

• A productivity assistant individuals can use at will?
• A governed enterprise capability?
• A feature embedded inside core platforms?
• A centralized service team runs for others?

If you don’t define this clearly, it’s a problem.

Employees don’t know what’s approved. Managers don’t know what’s expected. IT doesn’t know what it’s accountable for. Legal doesn’t know where the risks are contained.

Clear system boundaries create that psychological safety. They define where experimentation is encouraged, where governance applies, and who decides what.

AI thrives when there is clarity. It withers when there is ambiguity.

‍

Why this matters for tech leaders

If you’re a CTO, CIO, or Head of Engineering, this probably feels familiar.

You’ve been asked to “lead AI transformation.” But transformation doesn’t happen in the architecture diagram. It happens in org charts, incentive models, and performance reviews.

You can modernize the stack. You can integrate APIs. You can secure the data layer. But if no executive peer owns adoption outcomes, you’re fighting a losing battle.

This is why so many AI initiatives quietly shift from “transformational” to “experimental.” It’s safer organizationally. It creates less friction. It avoids forcing structural change.

But it also massively limits impact.

‍

How to design AI for adoption

So, what does “better” look like?

From what I’ve seen, successful AI adoption shares a few structural traits:

1. A named executive owner

Not a committee. Not a task force.

One executive accountable for measurable AI-driven outcomes across functions. That person must have cross-functional authority, not just influence.

Ownership changes conversations and forces clarity.

2. Incentives that reward workflow change

If leaders are only rewarded for protecting existing KPIs, they’ll protect existing workflows.

Adoption requires redefining what good looks like. That may mean:

• Measuring augmentation, not just output
• Rewarding experimentation
• Accepting short-term disruption for long-term gains

3. Embedded AI, not bolted-on AI

The more AI feels like “another tool,” the less likely it is to stick.

Adoption improves when AI is embedded inside platforms people already use. Where workflows already live.

This reduces cognitive load. It reduces switching costs. It reduces fear. And it makes AI feel less like a mandate and more like a natural evolution.

4. Clear risk containment

Platform teams need to define boundaries of responsibility. Legal needs clarity on guardrails. Employees need to know what’s safe to try.

When risk is contained, you can expect a bump in experimentation.

When risk is ambiguous, everyone hesitates.

Download: How to make AI really boring (& why that's a good thing)

‍

AI failure without tool-blaming

It’s tempting to blame tools. It’s cleaner. It avoids uncomfortable internal conversations.

But most stalled AI initiatives I’ve seen weren’t caused by weak models. They were caused by weak ownership structures.

The tool did what it was designed to do. The organization didn’t. And that’s actually good news. Because organizational design is within our control.

We can:

• Clarify ownership
• Align incentives
• Redefine system boundaries
• Give platform teams authority alongside responsibility

These are leadership choices, not vendor dependencies.

‍

The real transformation question

When companies say, “We want to adopt AI,” what they’re really saying is:

“We want to change how work happens.”

That’s not a procurement decision. It’s an organizational design decision.

Technology can accelerate transformation. But it cannot compensate for structural ambiguity.

If you want AI to scale, don’t start with features. Start with org charts. Start with incentives. Start with decision rights.

Design for adoption. Because in the end, AI doesn’t fail because it isn’t powerful enough.

It fails because it was no one’s responsibility to make it matter.

‍