Harvard Identified Seven Frictions That Kill AI Rollouts. You Probably Have All Seven.

A new piece in HBR from Karim Lakhani, Jen Stave, and Microsoft’s Jared Spataro lays out seven “last mile” frictions that prevent AI from moving beyond isolated pilots. Their core finding won’t surprise anyone who’s watched enterprise AI deployments up close: “The primary obstacle to progress is rarely model quality or data availability, but rather the ‘last mile’ of transformation where technical capability must meet organizational design.”

That’s the academic version of what we’ve been saying here since day one. The tools work. The organizations don’t.

Here are the seven frictions, made concrete:

• Pilot proliferation. Dozens of teams running independent AI experiments with no coordination. Each pilot proves value in isolation but none of them connect, so the organization can’t compound what it’s learning.
• Productivity gaps. Individual time savings that don’t translate to team or business-level outcomes because workflows weren’t redesigned around the new capability.
• Process debt. Legacy processes that were already broken before AI arrived. Automating a bad process just produces bad outputs faster.
• Tribal knowledge hoarding. Critical context lives in people’s heads, not in systems. AI tools can’t access what isn’t documented, and the people who hold that knowledge aren’t incentivized to share it.
• Governance gaps. No clear framework for who approves AI use cases, who’s accountable for outputs, or how to handle failures. So everything moves slowly or not at all.
• The efficiency trap. Organizations that use AI purely to cut costs, missing the opportunity to redesign work and create new value. You save 20% of someone’s time but don’t give them anything better to do with it.
• Training deficits. Generic “here’s how to prompt” training instead of role-specific workflow integration. People learn what the tool can do but not how it fits their actual Tuesday morning.

If this list feels familiar, it should. The Copilot 3.3% adoption data we covered is what these frictions look like in aggregate — proven technology that doesn’t scale because the organization around it wasn’t redesigned to absorb the gains.

The pattern keeps repeating. Companies buy tools, run pilots, see promising results in controlled conditions, and then can’t figure out why those results don’t spread. These seven frictions are why. Every one of them is an organizational design problem, not a technology problem.

Before your next AI tool purchase, run through this list honestly. How many of these seven frictions are active in your organization right now — and what are you doing about the ones that don’t involve buying more software?