A credit union or bank connects a chatbot to their SharePoint and tests it with a small group of employees. They quickly run into hallucinations, outdated information, and knowledge gaps. They then invest 12 months and expand headcount to get procedures to a place where the AI can use it.
That story isn't rare. It's close to the median outcome.
We've had a version of this conversation across regulated industries, financial services, and healthcare: leadership approved an AI initiative expecting a fast rollout. Instead, they got a spotlight that highlighted the true state of their knowledge management practices.
AI didn't fail these organizations. It exposed something that's been true for years, but nobody had a reason to look at it closely enough to see it.
Why This Keeps Happening
There's a line some of these organizations have heard, and it's a reasonable-sounding one:
"Don't worry, this is just a matter of updating some procedures..."
That line holds up in a world where policies, procedures, and regulations don't shift under your feet week to week. That's a fair description of most software. It is not a fair description of operations in a regulated industry.
Procedures in regulated industries are a moving target by design, not by neglect. Tools change. Teams change. Regulations change.
Layer on top of that the reality that most complex procedures aren't a straight line. They branch, with decision points that depend on a customer's specific situation, an account type, a regulatory exception. And a meaningful share of the real answers were never written down at all. They live in the head of the person who's been doing the job for twelve years.
"Garbage in, garbage out" isn't a new idea.
What's new is that most leaders didn't know they had garbage until they tried to feed it to something that demanded clarity.
Why the Old Tools Can't Fix This
SharePoint, shared drives, Word documents, and legacy knowledge bases were never built for this level of scrutiny, complexity, or velocity. They're storage tools. They hold knowledge. They were never designed to operationalize it.
That distinction matters more than it sounds like it should.
Having something written down somewhere and having something an employee can actually find, trust, and act on in the moment are two different problems. The old tools only solved the first problem. And for a long time, that was good enough, because nobody was asking the content to do more than sit there and be technically true.
AI is asking it to do more.
The Pipe and the Water
Here's the way to think about what's actually happening:
For years, the assumption was that the problem was access. There's water underground. There are people who need it. The trouble was getting it to them.
The chatbot is the pipe. It made the water easy to reach. What almost nobody checked was whether the water was any good.
What organizations are finding, over and over, is that the water was never clean. It's polluted with outdated steps, contradictions between two versions of the same procedure, and gaps where the real answer only ever existed in someone's head.
People try to drink it. It tastes awful. It makes them sick, in the form of a wrong answer given to a customer or a compliance exception nobody caught. And then they stop drinking. They stop trusting the chatbot, and they go back to the escalation desk or to guessing.
A System, Not a Project
If you were trying to provide drinking water in this scenario, you would never treat "cleaning the water" as a one-time project. To have any faith in the water, you would need to have a system in place that ensured the water was free from any and all contaminants.
Organizations in regulated industries don't just need a better pipe. They need a system that keeps the water clean on an ongoing basis: capturing it, updating it, maintaining it, and building in the feedback loop that catches contamination.
Organizations that are successfully adopting AI are finding that "updating procedures" is no longer a project you run once every few years. "Ensuring procedures are accurate and followable" is a skill that organizations have to systematize. Haphazard processes with no clear accountability aren't sufficient anymore.
The "Jobs To Be Done" of a Modern Platform
A platform built for this has to solve problems that the old tools were never asked to solve. These include:
- Capturing knowledge on a larger scale and across a larger number of Subject Matter Experts (SMEs)
- Formatting knowledge for performance
- Ensuring knowledge is accurate and that knowledge gaps are filled
- Speeding up feedback loops between employees and SMEs
- Ensuring knowledge can be used in the flow of work
All of these points to an increase in agility, velocity, and accountability.
A modern platform has to get knowledge out of people's heads far more efficiently than writing procedures the old way ever did. And it can't solve that by handing the job to AI outright. Using AI to auto-generate procedures produces something that reads fine and that nobody will ever actually use. Same root problem, just faster.
Capturing knowledge at that scale means distributing ownership out to the business units that actually hold it, not routing everything through one person. The moment a single documentation owner becomes the bottleneck, the content stops keeping pace with the business, and it never catches back up.
It has to surface duplicates and contradictions as a built-in function, not a periodic audit somebody remembers to run twice a year. Content health has to be structural, not an event.
Sustaining that accuracy means separating content by how often it actually changes. Stable reference material and fast-moving, high-frequency procedures need different maintenance models. Treating them the same is exactly why maintenance breaks down in the first place.
Traditional systems were built for a few people to author content that, if we're honest, very few people ever used. A modern platform has to run constantly, taking in feedback, pushing out changes, adapting to what the business actually looks like this week.
What Is This New Category?
Many organizations are treating this as just an evolution of their intranet, LMS, or document storage system. But when you look at the criteria above, you can quickly see that these systems weren't designed for these jobs. An LMS is for learning through exposure and memorization, an intranet is for communication, a document storage system is designed for containing PowerPoint decks, forms, and spreadsheets. None of these were specifically designed to handle complexity and change for procedures in regulated environments.
We call this new category of software a Knowledge Operations Platform. A Knowledge Operations Platform isn't just designed to store knowledge. It helps you build a system that operationalizes your knowledge: it ensures that knowledge can be trusted, updated, and applied in the moment of performance.
The Point for Leadership
The real decision isn't whether to invest in AI. The decision is whether or not a generic knowledge storage system is robust enough to support a systematized approach to operationalizing the organization's knowledge.
A software company can run its documentation on a wiki and a few disciplined engineers, because its underlying logic doesn't move the way we just described. Regulated industries don't get that luxury. That's not a maintenance gap you close with better discipline. That's a different category of problem, and it needs a different category of tool.
If your AI project stalled, or got pulled back after launch, the honest question isn't whether the AI failed. It's whether anyone ever checked the water it was drawing from.
