Many organizations and knowledge management (KM) teams have already taken the first step with AI. They’ve launched a pilot with maybe a chatbot, a generative AI search tool, custom agents, or a focused use case tied to a specific business problem. And often, that pilot works.
It answers questions. It saves time. It gets people’s attention. So why does progress so often stall after that?
The reality is that launching the technology is typically the easy part. The harder challenge is what comes next. Scaling, sustaining, and embedding AI into how work actually happens every day. That’s where many organizations find themselves stuck in what we might call the “messy middle”.
The Messy Middle of AI and Knowledge Management
The messy middle sits between early success and enterprise-wide adoption. It’s the phase where the excitement of a successful pilot meets the complexity of real-world implementation. This isn’t considered failure. It’s not even about resistance in the traditional sense. It’s what happens when a promising technology runs head-on into the realities of content quality, governance, and human behavior. And it’s completely normal.
From APQC’s research on change management and many years of experience working with organizations, we know that a sustainable transformation doesn’t come from deploying new tools alone. It requires intentional effort to help people work differently, change how decisions are made and improve how knowledge flows across the organization. And this is exactly where the messy middle gets…messy.
Where AI and KM Initiatives Start to Break Down
AI doesn’t create knowledge from nothing; it draws from what already exists. That means it quickly exposes the reality of your content ecosystem. If your knowledge repositories are outdated, duplicative, inconsistent, or irrelevant, AI will surface those issues at scale. In other words, AI doesn’t fix content problems, it just makes them more visible.
This is why APQC has long emphasized the importance of strong content management practices. And in the context of AI, those practices aren’t just helpful, they’re essential.
Governance and Ownership Get Fuzzy Fast
In a traditional organization, content ownership is already a challenge. Now with AI, it becomes even less clear. Who owns:
- the source content?
- the AI-generated response?
- the ongoing improvements and evolution of the system?
Without clear answers, accountability will break down fast. KM, Digital, and business teams can end up operating in parallel instead of in partnership. APQC’s continued guidance on content governance is especially relevant here. Clearly defined roles and responsibilities are critical for sustaining any KM initiative, including content management and AI.
Trust Is the Make-or-Break Factor for AI Adoption
AI introduces a new dynamic where users must decide not just whether to use a system, but how much to trust it. If trust is too low, it has a negative impact on adoption and people begin reverting to their old habits. Which means the intended investment will never be realized.
If trust is too high, there is also risk. People begin to rely on incomplete or incorrect answers without verification. Remember the adage garbage in equals garbage out? Recently, an APQC member shared something they learned at a conference which resonated with me. With generative AI, garbage in may now equal “gospel” out.
Most organizations today are landing somewhere in between where usage is inconsistent, and confidence levels vary. Building trust with your AI solutions requires more than just accuracy. It requires transparency, feedback mechanisms and loops, and clear expectations about how AI should (and shouldn’t) be used.
Behavior Change Often Lags Behind AI Capability
Even when technology works, people don’t automatically change how they work. They might:
- continue searching the way they always have,
- hesitate to rely on AI outputs, or
- fail to contribute and maintain content in ways that support the system.
This is where many AI and KM efforts can often stall. The capability is there, but the behavior hasn’t caught up, and this is fundamentally a change management challenge. As APQC’s research consistently shows, tools don’t change behavior—people do. Intentional and ongoing communication, upskilling, leadership reinforcement, and alignment with process and daily workflows all play a key role.
Why AI in KM Is More Than a Technology Implementation
At its core, the issue is simple: many organizations approach AI and KM as a technology implementation. But it’s really a cultural transformation that requires:
- clear ownership,
- strong governance, and
- sustained attention to how people adopt new ways of working.
Too often, we prepare the technology for our people; but not our people for the technology.
How KM Teams Can Scale AI Beyond the Pilot
The good news is that the messy middle is manageable if you approach it intentionally.
- Fix content before scaling AI. Start with the knowledge that matters most. Focus on high-value, high-use content, clear ownership, and defined lifecycle processes. You don’t need to fix everything, but you do need a solid foundation.
- Define roles early (and revisit them often). Be explicit about who is responsible for content quality, AI outputs, and governance decisions. And recognize that these roles may evolve as your AI capabilities mature.
- Invest as much in change management as in technology. This is where many organizations underinvest. Effective change management includes:
- communicating not just the what, but the why,
- upskilling people in the context of their daily work, not only providing a training course, and
- reinforcing new behaviors through leadership and incentives.
If you want different outcomes, you have to encourage and support different behaviors.
- Measure what actually matters. Usage metrics only tell part of the story. To understand impact, look at: time saved, rework avoided, time to competency, and improvements in decision speed or quality. These are the measures that resonate with the business—and sustain momentum.
The Real Work Starts After the AI Pilot
The messy middle isn’t a sign that AI + KM isn’t working. It’s a sign that the real work has begun.
Organizations that push through this phase don’t just deploy AI tools. They build the structures, habits, and trust needed to make knowledge flow more effectively. And in doing so, they move closer to what KM has always strived to achieve: helping people get the knowledge they need, when and where they need it, and in a way that resonates best with them, in order to do their work better.
AI can accelerate that vision but only if we’re ready for everything that comes AFTER the pilot.
To learn more, see APQC’s From Business Knowledge to Collective Intelligence: The New Edge in Performance and Navigating AI’s Impact on Knowledge Management.