Every large company has a graveyard of documents nobody properly buried. A benefits policy revised two years ago, but the old version still circulates as an email attachment. A code of conduct with an entire chapter about a corporate structure that no longer exists. A collective bargaining agreement replaced at the latest base date, while the three-year-old PDF stays open in half a dozen managers' browser tabs. These documents weren't deleted, didn't disappear, didn't go wrong on their own. They just died without anyone announcing it, and they keep being cited with the same authority they had when they were alive.
The problem isn't having an outdated document — that's inevitable in any company that changes policy with any regularity. The problem is that nothing, absolutely nothing, in the normal workflow signals that a document has died. It stays in the same place, with the same file name, with the same authoritative look it had the day it was published. Whoever goes to use it has no way of knowing, just by looking, whether they're reading a rule in force or a fossil.
How big is this hole
An MIT survey, Project NANDA, found that around 95% of generative AI pilots at companies showed no measurable impact on results by mid-2025 — and the cause wasn't the model, it was the knowledge base behind it: unstandardized and ungoverned. Gartner reaches a similar conclusion from the opposite angle: by the end of 2026, 60% of enterprise AI projects will be abandoned because the knowledge and data feeding them were never ready.
The zombie document is the concrete root of that abstract number. There's no way to have a reliable knowledge base when part of it is dead and nobody knows which part that is. And the problem isn't exclusive to AI — it predates it: an HR analyst who answers a benefits question based on an expired document makes the same mistake an AI would, except nobody audits the analyst by the same standard used to audit the model.
Why nobody kills the dead document
The answer is simple and uncomfortable: killing a document takes work and has no owner. Reviewing an entire policy, deciding whether it still holds, formally archiving the old one and publishing the new one are tasks that require time from someone already overloaded answering new questions, not checking old ones. So the old document just stays, stacked behind the new one, unlabeled, unannounced, with no one responsible for saying it no longer applies.
The result is that a document's authority, within corporate culture, never had a built-in expiration date. A contract has a clear term. An internal policy, in most companies, has none of that. It's read as permanent truth until someone, by accident or by luck, discovers it no longer is.
The cost of the ghost
The real damage from a zombie document isn't the document itself, it's the decision made on top of it. Labor guidance based on the wrong collective bargaining agreement can create liability. A benefit explained under the old rule means two employees treated differently for the same question, a month apart. None of these errors shows up as a system failure — it shows up as bad human judgment, when it's actually a signaling failure: nobody announced that the document had died.
That's why treating document validity as metadata, not assumption, has stopped being a document-management luxury and become an operational prerequisite, whether or not AI is involved. A document marked as expired keeps existing, but stops circulating as if it were current rule — which already solves most of the problem without needing to delete anything.
AskLisa
Every department has a zombie document circulating right now, and most don't know which one it is. The question isn't whether a dead policy is being cited as alive inside the company. It's how many decisions have already been made on top of it without anyone noticing.
At AskLisa, every document has an age, a term, and an anniversary. And if it dies, everyone knows. Including your AI.
