ARTICLE SUMMARY
  • RAG is the technology that enables AI to query external data and serve end-users company-specific AI summaries.
  • Every company has a “junk drawer” of knowledge that houses unreliable information.
  • Typical AI implementations fail because companies treat it as a one-and-done project, but AI requires continuous maintenance. 
  • A Knowledge Ops System — including the right methodology, software, and team — helps businesses prepare their operational knowledge for AI and keep the information up to date, allowing AI to deliver accurate, up-to-date information.
Greg DeVore

By: Greg DeVore on February 18th, 2025

Print/Save as PDF

Why Your AI Needs Better Operational Knowledge

Artificial intelligence (AI) is only as good as the information you feed it.

Yet, many organizations still fall into the trap of thinking AI is a plug-and-play solution. They assume that if they collect a bunch of documents and import them into an AI system, it will magically provide accurate, useful answers. 

Unfortunately, that’s not how it works. 

If your knowledge management system is messy, your AI will be unreliable, producing inconsistent and even incorrect information. 

This is why operational knowledge matters. 

In this article, you’ll learn how AI operates, where businesses go wrong with AI implementation initiatives, and how you can better leverage AI with a Knowledge Ops Platform.

RAG: The technology behind AI and why AI needs clean knowledge

So, why does AI need you to adopt a better system for managing your company knowledge?

First, let’s look at the technology behind AI to understand the need to prepare operational knowledge before implementing an AI system.

What is Retrieval-Augmented Generation (RAG)?

Retrieval-Augmented Generation (RAG) is a method that combines information retrieval with AI-generated responses. It allows you to mix your company-specific knowledge with highly trained AI models. The combination of these two sources of knowledge is used to generate AI answers and responses.

How does RAG work?

RAG starts the same as other AI systems, like ChatGPT. It pulls from a Large Language Model (LLM). These cast a wide net to gather general information, which means they won’t know your company’s policies, workflows, or specific industry nuances.

What makes RAG different is it allows you to specify where you want the AI system to pull your company-specific knowledge from. RAG employs retrieval algorithms to fetch relevant information from external sources (i.e. knowledge bases, databases, or document collections).

Pulling from external sources is what allows you to deliver your AI-generated responses on operational knowledge specific to YOUR business.

However, RAG alone doesn’t fix the problem. To understand why, let’s first explore the major pitfalls of AI delivery.

Operational Knowledge Definition: Operational knowledge is the actionable knowledge employees need to perform tasks, answer questions, and troubleshoot problems. It is complex and constantly changing. It includes step-by-step procedures, troubleshooting guides, policies, and expert insights that keep business processes running smoothly.

The challenge: You have a junk drawer of knowledge

Junk drawer of knowledge strategy — where the fuel to AI doesn't exist.

Every company has a “junk drawer” of knowledge. 

What does that mean? You have information scattered across different tools, stuck in employees’ heads, buried in emails, or hidden in outdated documents. This makes it nearly impossible for employees (and AI) to find the right information when they need it.

This “junk drawer” is often a knowledge base or knowledge repository, like SharePoint. It is most likely a location where people store documented policies and procedures and then promptly forget about them.

The result is your “junk drawer” houses unreliable information. The information is inconsistent, outdated, and incomplete. When we work with organizations that claim to have good documentation, we find that at least 50–75% of their operational knowledge is not documented

That’s why AI and employees are pulling from resources that serve them incorrect information and cause them to make mistakes.

When knowledge is buried in cluttered documentation, AI struggles to extract accurate and useful insights. It makes it difficult for end-users to get the right answer.

Also, operational knowledge gets messy in your junk drawer. Without a structured system in place, knowledge becomes fragmented, making AI-powered solutions ineffective.

Common mistake — treating AI as a one-and-done project

Organizations make one major mistake in how they approach implementing AI into their knowledge management strategy. They look at AI as a magic fix — but AI isn’t magic. Feeding AI with accurate operational knowledge takes continued effort. 

If you want AI to be effective in supporting your employees, AI needs continuously updated knowledge to stay accurate. It’s a constant project that requires the right team, the right strategy, and the right software.

What happens during a typical AI implementation initiative

Typical 4-step process organizations use to implement AI using RAG.

Let’s look at this from the perspective of how organizations typically handle an AI initiative. When organizations decide to add AI tools to their operations, this is typically what happens.

1. Gather documentation

Businesses pull together all their existing knowledge. This information is often found in:

  • Emails
  • PDFs
  • Spreadsheets
  • Internal wikis
  • Slack, Teams, or other chat message channels
  • Etc.

2. Import it into a data lake

Once you’ve gathered the documentation, then they store all of that knowledge in a centralized location.

(Side Note​​: Not all AI implementations use a data lake. Some use vector databases or other knowledge retrieval systems. For this scenario, we’re talking about those AI systems that use a data lake.)

3. Train the model or use RAG

Next, AI is either trained on the data or uses RAG. The goal is to get AI to retrieve relevant information from the data lake when answering questions.

4. Test and refine

Then they test their work … AND the results aren’t what they wanted.

Why the results aren't what the organization wants

What happens when businesses test RAG without accurate operational knowledge?

Many companies set up their AI system and quickly realize it’s not working the way they expected. Here’s what happens:

  1. They test and refine the system, but the AI gives inaccurate or incomplete responses.
  2. They go back to step one — gathering more documentation — thinking the issue is with the quantity of data rather than its quality.

The real problem is that they don’t have the right data and it’s not structured in a way that AI can use it effectively. In short, they haven’t established a data lake with good data hygiene for their operational knowledge.

That’s where a Knowledge Operations System comes in.

The Solution: A Knowledge Operations System

4-step Knowledge Operations System for AI

Instead of treating AI as a standalone tool, businesses need a structured Knowledge Operations System to support it.

The system is a strategic approach that involves:

A Knowledge Ops Platform allows you to update and refresh content as the operational knowledge changes.

You have the process, you have the software, you have the team to update the data lake that RAG is going to use. It is a way to fill the data lake with the most accurate information. Plus, it allows you to constantly update and refresh the information as operational knowledge changes.

Watch this video to see how it works, and keep reading for more details.

1. Identify knowledge gaps

Before implementing AI, businesses must identify knowledge gaps that could lead to inaccurate or incomplete AI responses. These gaps often include:

  • Inconsistent procedures across teams – Employees follow different processes, leading to confusion.
  • Employees getting stuck mid-process – Unclear or incomplete steps cause delays and errors.
  • Tribal knowledge locked in experts’ heads – Key insights aren’t documented, making them inaccessible.
  • Unaccounted-for variables in how-to guides – Instructions only present the ideal step-by-step instructions. They are missing less common scenarios.

When you identify knowledge gaps, you are collecting a list of everything you need your employees to be able to handle on their own. This way, you can ensure you can build out that knowledge to include in your data lake.

Identify knowledge gaps with a Find & Follow Workshop

To identify knowledge gaps, we recommend running a Find & Follow Workshop. This workshop helps you work through everything you need an employee to do so that you can create learning resources to support them. 

Learn more about a Find & Follow Workshop and how to identify knowledge gaps with this 3-minute video.

2. Capture knowledge from SMEs (subject matter experts)

AI can’t generate useful answers from scattered, unstructured information. That’s why businesses need to extract knowledge from their experts and centralize it.

There are a few ways you can capture the knowledge in your SMEs’ heads. You can either have:

  1. SMEs document the step-by-step procedures on their own OR
  2. Content authors can interview SMEs and document the process

A Knowledge Ops Platform is a system built specifically to help organizations handle operational knowledge. It helps you quickly capture SME expertise and make it easily accessible to other employees.

The Knowledge Ops Platform has AI-powered tools to assist you in documenting and sharing your operational knowledge.

🚨Feature Alert: With the ScreenSteps Knowledge Ops Platform, you can use the Audio AI Article Generator to record a SME talking through a procedure. The Audio AI then turns the recording into a digital guide.

3. Refine and approve

AI delivers useful insights only when it pulls from clear and structured knowledge resources. That’s why refining and approving knowledge is essential.

When creating your digital guides, design them in a way that is easy for a human to understand. AI, too, needs a clear path for the step-by-step instructions.

Format the guides so that they are findable, followable, and scannable. That way, when they are added to your data lake in your Knowledge Ops Platform, it is easy to summarize.

Consistency in design improves both employee performance and AI efficiency.

4. Review and certify

If knowledge is outdated, AI responses will be inaccurate. 

Create a plan to continuously update the digital guides in your Knowledge Ops Platform (aka your data lake). 

Establish regular reviews to ensure accuracy. Assign an article owner. Schedule reminders to review and certify the content in that article.

Additionally, provide employees with a way to provide feedback on articles. Those who are handling the procedures will be the first to notice if something isn’t working.

The results: AI that actually works

Without a holistic Knowledge Operations System, AI is worthless in the knowledge management space. 

BUT, if you have the right combination of methodology, software, and team, AI can be a game-changer for how your business manages and transfers operational knowledge. 

The ScreenSteps Knowledge Ops Platform — combined with the Find & Follow Framework — enables businesses to leverage the power of AI. This means:

  • More relevant, accurate, and contextually appropriate AI responses
  • Improved employee efficiency and reduced errors
  • AI that truly enhances business operations rather than causing confusion and mistakes

Ready to empower employees with AI resources? Talk to a ScreenSteps expert to learn how our Knowledge Ops Solution can help your organization make AI work for you.

 

About Greg DeVore

CEO of ScreenSteps