(This post was originally shared on my LinkedIn Profile.)
We talk obsessively about AI readiness. Plug in the right tools, train the model, scale the adoption. But I keep hearing the same quiet concern in rooms full of knowledge and learning leaders: What happens when we feed the beast garbage?
The garbage has a name: ROT. Redundant, obsolete, trivial content. And if you’re serious about AI readiness—or frankly, about knowledge management at all—you need to stop treating ROT disposal as a nice-to-have technical task. It’s a governance problem. It’s a people problem. And it’s non-negotiable.
The ROT Problem Is Your AI Problem
Here’s the uncomfortable truth: the organisations struggling most with AI implementation are the ones drowning in ROT.
Redundant documents that say the same thing three different ways. Obsolete processes locked in PDFs from 2015. Trivial information cluttering search results so badly that finding what you actually need takes longer than just asking a colleague. Add multiple systems, no single source of truth, and inconsistent version control, and you’ve created a perfect petri dish for training data that’s incomplete, contradictory, or simply wrong.
Now feed that into an AI system. The model doesn’t know intent. It doesn’t know that the 2015 process was replaced by something newer. It can’t distinguish between the authoritative procedure and the three near-duplicates authored by well-meaning teams in different divisions. It learns from what you give it. Garbage in, garbage out—but faster, at scale, and with the veneer of algorithmic confidence.
That’s not an AI problem. That’s a governance problem. Which is a people problem.
Enter ART: Accurate, Reliable, Trustworthy
The antidote to ROT is ART: content that is accurate, reliable, and trustworthy.
Accurate means it reflects reality as it currently stands. The process described is the process actually followed. The policy hasn’t shifted. The template is current. Accuracy requires human judgment—someone who understands the business has to say, “This is true today.”
Reliable means it’s consistent and findable. Identical information doesn’t exist in five formats. Metadata is correct. Version history is transparent. There’s a clear path to the current, authoritative version. Reliability requires governance: standards, ownership, maintenance schedules.
Trustworthy means people believe it enough to act on it. That only happens when they’ve seen it stay accurate and reliable over time. Trustworthiness is built through culture—through demonstrating that the knowledge function takes stewardship seriously.
None of these are technical properties. They’re human commitments, enforced through governance structures.
The Governance Gap
Here’s where most organisations trip. They treat quality as a technical project: cleansing tools, metadata standards, taxonomy implementation. These are important. But without governance, they’re cosmetic.
Governance asks the hard questions:
- Who owns this content? (Not who wrote it. Who’s responsible for keeping it true?)
- Who decides what’s redundant, and by what criteria? (Delete the 2015 process, but who makes that call, and what if someone was relying on it for context?)
- When does content become obsolete, and who’s checking? (That vendor partnership doc from 2012—is it history or a hidden liability?)
- What happens when we find contradictions? (Two teams think they own the same process. Now what?)
- Who’s accountable for trustworthiness? (Not a system. A person or team.)
Governance without people is theatre. But people without governance quickly become overwhelmed, inconsistent, or burnt out. You need both.
Keeping Humans in the Loop
This is where AI readiness becomes interesting. The temptation is to assume AI can automate the governance problem away—classifiers that identify ROT automatically, algorithms that spot inconsistencies, systems that flag obsolete content.
Some of that is useful. But it’s not sufficient. And if you try to make it sufficient, you’re back where you started: garbage in, garbage out.
Humans stay in the loop at the critical junctures:
Defining what matters. What counts as redundant rather than useful context? What’s a necessary variant rather than an inconsistency? These are judgement calls. Tools can flag; humans decide.
Ownership and accountability. An algorithm can tell you content needs updating. A human has to take responsibility for updating it, and for saying “I’ve reviewed this and it’s true.”
Interpretive consistency. Policies that sound the same might apply differently in different contexts. Humans who understand the business catch that. Algorithms amplify the confusion.
Escalation and exceptions. ROT-clearing sometimes reveals gaps or conflicts in your business logic, not just your documentation. Those need human attention, not automated resolution.
The goal isn’t to remove humans from governance. It’s to remove humans from the repetitive, low-judgment parts so they can focus on the parts that require discretion, accountability, and real understanding.
The Real ROI
Organisations that have tackled ROT properly—through committed governance, clear ownership, and sustained human attention—report something interesting: it’s not just cleaner data. They report faster decision-making, fewer duplicated efforts, lower training time for new staff, and better cross-team collaboration.
And when they do deploy AI, it works. Not because the tool is magic, but because the foundation is solid.
The quiet concern I keep hearing? It’s legitimate. You can’t automate your way out of a governance problem. But you can solve it. It just requires treating knowledge stewardship as seriously as you treat technology adoption.
That’s the difference between readiness and recklessness.