What to Do if the Knowledge Needed by People and AI Agents Isn't Documented Anywhere

In a previous post, I pointed out that Agentic AI will fail badly if AI Agents aren’t fed with the specific enterprise knowledge they need to perform complex, multi-step tasks. But how should this knowledge be obtained and codified so that an AI Agent can be truly effective?

Before we can answer that question, let’s understand a bit more deeply what an AI Agent needs to know in order to be effective.

To produce the best results, any agent - whether human or machine, needs to have knowledge from multiple domains - what we call multi-domain knowledge.  From this perspective, the agent needs to have knowledge from 3 distinct, related domains:

• The source of the tasks it needs to perform - this is knowledge about the situations that trigger the agent to perform actions - let’s call this the source domain.

• The target of its tasks - this is knowledge about the people or systems that use the results of the agent’s actions  - we’ll call this the target domain.

• Specific steps the agent needs to take under various circumstances as it performs its tasks - we can call this knowledge the action domain.

As an example, let’s say that we want to design an AI Agent that is intended to improve telecom customer experience - something we hear about a lot when it comes to AI Agents.

Let’s say that we want our AI Agent to notice when a customer isn’t getting the level of service that they paid for (maybe too many dropped connections), and to take some action when it happens - like informing them that it’s not their device that’s misbehaving - it’s the network that’s having a problem.

We also want the AI Agent to let the customer know when the situation has been fixed, and maybe even give them a credit on their bill, if the situation was bad enough.

With this seemingly simple example in mind, what are the 3 domains that our AI Agent needs to have knowledge from?

• In this case, the source domain is network service performance - the AI Agent needs the right knowledge to properly identify when the customer’s device isn’t receiving the right level of service from the network.

• The target domain is the customer - among other things, the AI Agent needs to know what information it can provide to the customer, under what circumstances, what items can be changed on their bill, etc.

• The action domain is Customer Experience (CX) - the AI Agent needs to know what steps it can take in order to measurably improve CX  - like not telling them that a service-impacting situation is fixed when it actually isn’t.

Here’s a business process diagram that illustrates what the AI Agent needs to know. It’s a simplified example of the type of artifact that DomainStream’s Knowledge Management Method™ produces, but it’s essential to capture properly - so it can be agreed and aligned across all the enterprise stakeholders involved, before feeding it to our AI Agent.

Business Process Diagram for a Customer Experience AI Agent

So while this example appears simple, the domain-specific knowledge that our AI Agent needs to operate effectively - and avoid wreaking havoc with customers, is anything but simple. Each of these domains has its own set of rules, metrics and dependencies that need to be captured and codified before they can be ingested by an AI Agent.

But even more importantly, it is the inter-domain knowledge that is essential for our AI Agent to do its job properly.

For example, what interactions should happen with customers when the network causes a full service interruption vs. a partial service degradation, like slow speeds? How long should the AI Agent wait before informing customers - to avoid needlessly annoying them with transient network conditions? How many service-impacting issues should result in a credit to a customer? This kind of inter-domain knowledge involves deep expertise from both the service performance and the Customer Experience domains.

Over the myriad use cases that we’ve worked with our clients, this kind of inter-domain knowledge typically isn’t documented anywhere. This knowledge needs to be created - by working with multiple domain-specific subject matter experts and capturing it properly, so it can be ingested and used effectively by our AI Agent. 

Another topic that I’ll cover in an upcoming blog is how to capture the knowledge needed by AI Agents to accurately measure and learn whether or not they are actually achieving their desired outcomes - like improved Customer Experience.

By this point, I think it’s becoming clear why accurate, multi-domain enterprise knowledge is essential for AI Agents to truly take off.  If this knowledge isn’t documented anywhere, no amount of Web scraping or other automation provided by enterprise knowledge management tools alone can solve the problem.

DomainStream’s Knowledge Management Method™ (KMM), combined with our multi-domain experts are purpose-driven to work closely with all your stakeholders and experts, using the latest tools to assemble the multi-domain knowledge needed for your specific use cases - whether you’re building an AI Agent or building an enterprise knowledge base that helps people do their jobs more easily and effectively.

Contact us to find out more.