When a customer service team introduces AI, something happens to the work that few have planned for. The agents answer fewer tickets, but someone has to make sure that what the AI answers from is correct, and that it stays correct when prices, terms and products change. I have seen it in every AI rollout I have been part of: the decisive question is not which tool is chosen, but who owns the knowledge afterwards. This article describes that role.
Why does customer service need a knowledge manager now?
Because AI turns the knowledge base into customer service's most important means of production, and a means of production without an owner decays. When the help centre, the chat and the AI drafts in the inbox draw their answers from the same articles, the same error is repeated in every channel until someone corrects the article.
The industry has started to see it. In a Gartner survey of 321 customer service leaders in autumn 2025, 58 percent planned to upskill agents into knowledge management specialists, tasked with reviewing and curating AI-generated content. In the same survey, reported in a later press release, 84 percent said they plan to add new skills to the agent role, and almost 80 percent that they intend to move agents into new tasks.
What does a knowledge manager do?
The knowledge manager owns the knowledge base as a whole, while individual answers are still owned by whoever knows the subject. It is a steering role, not a writing role. These are the tasks:
- Owns the structure and the template. What an article looks like, which parts it has, what is shown to the customer and what is internal. Without a template ten people write ten kinds of articles; what the template should contain is described in the article on writing help articles.
- Prioritises what gets written. Reads the inflow, sees which questions lack answers and decides which are closed first. The method for that is in the article on knowledge gaps.
- Reviews what is proposed. AI drafts the agents have changed, new articles from colleagues, change suggestions from the chat logs. The knowledge manager decides what becomes an article and makes sure the subject owner approves the content.
- Holds subject owners to their answers. Returns are owned by logistics, invoices by finance. The knowledge manager makes sure they review their articles when something changes, and chases them when they do not.
- Removes and merges. Outdated articles, duplicates and campaign texts that have expired. It is the task that never gets done if nobody owns it.
- Follows up. Which articles are used, which get poor ratings, which questions still reach a person, and how often the AI drafts are sent without changes.
The Knowledge-Centered Service method, KCS, maintained by the Consortium for Service Innovation, is built on knowledge being captured in the workflow by everyone who resolves tickets, as a by-product of the work. But KCS also describes a specific role for whoever analyses the knowledge domain: according to the KCS guide on key roles that person should identify the articles that give the most value, see patterns in what is reused, and drive improvements in product, processes and policy that remove the cause of the most common tickets.
How does the role differ from agent, team lead and subject owner?
The knowledge manager owns the answers, the team lead owns staffing and the quality of the interaction, the agent resolves the ticket, and the subject owner is responsible for an individual answer being correct. This is how they differ:
| Role | Responsible for | Typical question | What the role does not do |
|---|---|---|---|
| Agent | Resolving the ticket, and flagging when the answer is missing or wrong | "The article on exchanges says nothing about sale items" | Does not rewrite the article alone but goes via the knowledge manager |
| Knowledge manager | The knowledge base as a whole: structure, prioritisation, review, follow-up | "Which ten questions lack answers this month?" | Does not decide the return policy, the subject owner does |
| Subject owner, for example logistics or finance | The content of their articles being correct | "The return period changes on 1 October, update the article" | Does not write the customer text, customer service does |
| Team lead | Staffing, queue, interaction quality, staff development | "How do we get through November?" | Does not own the knowledge base, even if it often ends up that way for lack of anyone else |
The most common confusion is between knowledge manager and team lead. That is precisely why the role should be its own, with its own time.
How much time does the role take?
Less than a full-time post in most teams, but more than zero, and the time must be protected. My benchmark after introducing the role in several organisations: in a team of five to ten people with a few hundred tickets a week, one person for about a day a week is enough, provided the agents flag gaps themselves and the subject owners respond to review requests. When AI is introduced, or when the knowledge base has to be built from scratch, more is needed for a few months. That is a benchmark from practice, not an industry figure.
The time usually already exists. According to Gartner's survey, 63 percent of customer service leaders are reducing agent headcount gradually through attrition and moving the capacity to higher-value tasks. Knowledge work is such a task, and it pays for itself in fewer tickets. A knowledge manager measured on tickets resolved stops being a knowledge manager within a month.
Who fits the role?
Someone who writes clearly, asks why and has the trust of their colleagues, and that is not always the fastest agent. The traits that matter:
- Writes clearly. The role is more about shortening and clarifying than about producing text.
- Is curious about causes. Sees ten tickets about the same thing as one problem to remove, not ten tickets to resolve.
- Dares to say no. To articles that are not needed, to subject owners who want marketing copy in, to managers who want "a quick FAQ".
- Has the team's trust. Agents should want to flag gaps to that person, not feel audited.
- Is comfortable with numbers. The role is driven by data: usage, ratings, recurring questions.
A warning from experience: the best agent is often the obvious choice, and that can be right. But the best agent is also the one who takes the most difficult tickets, and if you remove them from the queue without replacing the capacity you get a worse queue and a knowledge manager who is constantly pulled back into it.
How do you measure that the role has an effect?
By following whether the knowledge is actually used and whether it removes tickets, not by counting articles. Four metrics are enough:
- Self-service rate per question. The share of a given question resolved in the help centre or chat without a person. Should rise for the questions the knowledge manager prioritised.
- Share of AI drafts sent without changes. When the agents stop correcting the drafts for a ticket type, the article behind them is right. How that metric is used to decide what can be automated is covered in the article on AI drafts.
- Time from flagged gap to published article. Days, not weeks. If it rises, the role is understaffed or the subject owners are slow.
- Recurring questions removed. Questions that stopped arriving because the cause was fixed in product, website or process. That is the metric that shows the role is doing what KCS calls domain analysis, and it is the one that gives the most.
What to do
- Appoint one person and write down the remit in ten lines: what the role owns, what it does not own, how much time, and which four metrics it is followed on.
- Protect the time. Book it in the schedule as if it were a shift in the queue, and let nobody take it when the queue grows.
- Give the agents a way to flag gaps and errors in the tool they already work in, with one click, and make the knowledge manager the recipient.
- List the subject owners per topic area by name, and agree that they respond to a review request within a week.
- Start with the twenty most common questions. The knowledge manager makes sure each has a reviewed article used in the help centre, in the chat and as a draft in the inbox. In Supportifier the knowledge manager sees in one view which articles are used, which get poor ratings and which questions lack answers.
- Follow the metrics every month and report them to management as part of customer service's results, not as a side project.
Common questions
Does a small team of three need a knowledge manager?
Yes, but as part of someone's job, not as a post. The responsibility has to sit with a named person, otherwise it sits nowhere, and that applies regardless of team size. In a team of three a few hours a week is often enough, provided all three flag gaps and the person who owns the role gets those hours in the schedule.
Should the role sit in customer service or in marketing or product?
In customer service. That is where the questions arrive, where you see which answers work and where the knowledge is used every day. Marketing and product are subject owners for their parts and should review the content, but whoever prioritises and follows up has to sit close to the inflow. A knowledge base owned by a department that does not answer customers becomes documentation about the product, not answers to questions.
Does AI replace the need for a knowledge manager?
No, AI makes the need greater. AI can propose articles, find duplicates and point to questions without answers, but it cannot decide what is right when the terms change, and it cannot hold logistics to its return rules. Gartner's survey shows that 58 percent of customer service leaders plan to train agents to review and curate precisely AI-generated content. The role changes from writing to deciding, but it does not disappear.
What happens to the agents who become fewer?
Fewer than many think, and several of them change tasks rather than disappear. According to Gartner only 20 percent of customer service leaders had reduced headcount because of AI by the end of 2025, and Gartner predicts that half of the companies that cut because of AI will rehire by 2027. Knowledge manager is one of the roles agents are moved into.
Sources
- Customer Service and Support Leaders Must Prioritize Blending Human Strengths with AI Intelligence in 2026 — Gartner, 2025
- Gartner Survey Finds 91% of Customer Service Leaders Under Pressure to Implement AI in 2026 — Gartner, 2026
- KCS Principles and Core Concepts — Consortium for Service Innovation, read 2026
- KCS v6 Knowledge Domain Analysis Reference Guide: Key Players and Capabilities — Consortium for Service Innovation, read 2026
- Gartner Survey Finds Only 20% of Customer Service Leaders Report AI-Driven Headcount Reduction — Gartner, 2025
- Gartner Predicts Half of Companies That Cut Customer Service Staff Due to AI Will Rehire by 2027 — Gartner, 2026
Rickard Collander