AI-ready knowledge is becoming one of the biggest tests of whether customer service teams can scale AI safely. Many organizations are still most interested in comparing models. They ask which large language model is fastest, smartest, or most flexible. Yet for service teams, that may be the wrong starting point.
AI cannot resolve issues confidently if it does not understand the company it is serving. It needs product policies, processes, business rules, customer context, escalation logic, and the authority to act. For Hannah Deveney, Senior Director, Product Management at Zendesk, that makes knowledge a foundation for AI-led service:
“The model provides the reasoning, but the knowledge dictates the outcome.”
That distinction matters as customer service moves beyond AI assistants and toward more autonomous agents.
This is where the AI conversation becomes less glamorous but more useful. Service leaders can choose impressive models and launch polished pilots, but the outcome still depends on the knowledge those systems can reach. If that knowledge is fragmented, outdated, or difficult to access, AI may simply reproduce the gaps already inside the operation.
Why AI-Ready Knowledge Comes First
The pressure on service teams is easy to understand. Customers want quick answers, and they expect every interaction to carry context. Zendesk’s CX statistics show that 70% of customers expect anyone they interact with to have the full context of their situation.
That expectation applies whether the customer reaches a human agent, a self-service journey, or an AI agent. Yet many service organizations still manage knowledge as a static resource. Policies live in PDFs. Process updates sit in shared drives. Product details sit with veteran agents. Internal exceptions live in Slack or Teams.
That may work in a human-only environment, because experienced agents can fill in the gaps. It does not work as well when AI agents are expected to resolve issues on their own. Deveney compared it to hiring a strong employee and withholding the training:
“You wouldn’t hire a brilliant new employee, and then refuse to train them on your products. But that’s exactly what happens when companies deploy AI without a strong knowledge foundation.”
This is where many AI programs run into early friction. The model may be strong, but the business knowledge around it is not ready. That can lead to confident wrong answers, failed actions, and escalations that create more work for human agents.
Fragmented Knowledge Creates Confident Errors
One of the biggest risks is an AI agent delivering a confident but incorrect answer.
If a company’s return policies are split across multiple systems, with one version in a PDF, another in an internal wiki, and another inside a legacy tool, an AI agent may find the wrong answer and present it clearly.
That creates a new problem for the human team. Instead of simply handling the customer’s original issue, an agent now has to explain the mistake, rebuild trust, and correct the action. Deveney warned that poor knowledge can turn AI from a resolution engine into a source of extra effort:
“If you have your return policies split across three different systems with outdated PDFs, AI is going to pull the wrong answer. Instead of reducing ticket volume, you’ve increased your handle time and eroded trust.”
The lesson for CX leaders is direct. AI performance cannot be separated from the quality of the knowledge it can access. A weak knowledge base does not simply limit self-service. It limits agent assistance, automation, and the quality of AI-led resolutions.
Knowledge Has To Support Action
Modern service knowledge is no longer only about answering questions. In an agentic AI environment, knowledge also has to support action. An AI agent may need to process a refund, update an address, change a booking, check eligibility, or trigger a workflow.
That requires more than an article. It requires structured rules, clean permissions, reliable workflows, and system access that lets the AI complete work safely.
Zendesk’s own knowledge positioning reflects this shift. The company describes Zendesk Knowledge as a platform to build, connect, and surface knowledge across agents, AI, and self-service. It also highlights the ability to bring knowledge from multiple sources into a single platform, surface the right information in every interaction, and continuously identify content gaps.
That matters because human agents and AI agents use knowledge differently. A human agent may need a clear explanation of a policy. An AI agent may need the same policy in a structured format, with the relevant business rules and action paths attached. Deveney explained the difference:
“For a human agent, knowledge might be a really well-written paragraph explaining a policy, but for that AI agent, knowledge also includes business rules, API endpoints, and structured data that gives it permission to take action.”
That is a useful distinction for service leaders. AI readiness is not only about whether the right information exists somewhere. It is about whether that information is usable by both people and machines.
Human Agents Are The Knowledge Litmus Test
Service leaders do not need to start with a large technical audit to find early warning signs. The first clues often sit with human agents.
If agents regularly bypass the knowledge base and ask questions in Slack, Teams, or informal group chats, the knowledge environment is probably not trusted. If newer agents cannot find the right answer without tapping a veteran colleague on the shoulder, that is another warning sign. Deveney framed those behaviors as a simple test of knowledge health:
“If your people can’t find the right answer, AI won’t be able to either.”
That line should sit close to any AI planning process. Human workarounds are often treated as cultural habits, but they may point to deeper operational problems. They show where knowledge is missing, stale, hard to navigate, or written in a way that does not reflect real service work.
AI makes those gaps harder to ignore. When a human agent cannot find an answer, they may improvise. When an AI agent cannot find the right answer, it may escalate, stall, or act on the wrong information. None of those outcomes helps the customer.
From Static Articles To Dynamic Knowledge
The role of knowledge is also becoming more strategic. For years, many organizations treated the knowledge base as a publishing project. Teams created articles, reviewed them occasionally, and hoped agents or customers could find them when needed.
That approach is unlikely to support AI-led service at scale. Deveney argued that companies need to stop thinking of knowledge as static content and start treating it as a live operational asset:
“Knowledge has to become like a dynamic operational asset for them both.”
The “both” is important. Knowledge now has to serve human agents and AI agents in the same service system.
That means updates cannot sit in one part of the business while another team works from older information. If an AI agent learns from a new resolution pattern, that insight should be available to humans too. If a human agent identifies a process issue, that feedback should improve what AI can access.
That is also why new roles are beginning to emerge.
In an earlier video interview with CX Today, Deveney pointed to the rise of knowledge engineers. These roles move beyond documentation writing and focus on structuring data, processes, and tools so AI agents can act safely and human agents can work from the same source of truth.
That is a practical evolution. As service teams become systems of human and digital agents, someone has to design and maintain the knowledge layer that supports them.
What CX Leaders Should Check First
Before scaling AI, service leaders can ask a few simple questions.
Can frontline agents find the right answer quickly? Are policies current across every channel? Do AI agents and human agents use the same knowledge core? Are process rules structured enough for AI to act on them? Is knowledge updated continuously, or treated as a yearly cleanup project?
If the answers are uncomfortable, that does not mean AI rollout should stop. It means the knowledge work has to move up the priority list. The wider AI conversation often focuses on capability. The service conversation needs to focus on readiness too.
The most advanced model in the world cannot compensate for messy business knowledge. It may even expose the problem faster.
For service leaders, that is the real opportunity. Build the knowledge foundation well, and AI can help customers resolve issues faster, support agents with better context, and turn service operations into something more consistent and scalable. Ignore it, and AI may simply automate the confusion that already exists.
