What is a knowledge base?
A knowledge base is a curated, searchable collection of information, such as help articles, policies, product details and procedures, that customers, employees and AI systems draw on to answer questions consistently, with each entry kept current through ownership, review dates and a record of past versions. It records what is true and how things work; it does not by itself decide what should happen next in a conversation.
Organizations often run more than one. An external knowledge base is the public help center. An internal one guides employees through procedures and exceptions. A growing third kind is written mainly for machines, structured so that an AI system can retrieve the right passage. Keeping all of them accurate is the discipline called knowledge management.
How a knowledge base works
The same content increasingly serves three audiences, and each needs something slightly different from it.
| Audience | How they use it | What they need from it |
|---|---|---|
| Customers | Search a help center or follow suggested articles | Plain language, answers that are easy to find, current details |
| Employees | Look up procedures during calls, chats and case work | Exact steps, exceptions spelled out, fast lookup |
| AI systems | Retrieve passages to base their answers on | Short self-contained sections, clear scope and dates, no contradictions |
Behind every article sits a lifecycle: drafting, review, publication, periodic re-review and retirement. Each article needs an owner who answers for its accuracy and a review date that triggers a check. The most common structural failure is duplication, where the same question is answered in three places and only one of them was updated.
Why knowledge bases matter
In a Gartner survey of 5,728 customers, only 14% of customer service issues were fully resolved in self-service. The most common reason for failure, in 43% of cases, was that customers could not find content relevant to their issue.
Content is where self-service most often breaks: an article is missing, out of date or contradicted by another. When an AI system answers from that content, the same failures reach customers faster and with more confidence, because a fluent answer built on a stale article reads as authoritative. A knowledge base has become infrastructure for AI, and it needs the upkeep that implies.
Where knowledge bases are used
- Help centers. Public answers on products, fees, account management and policies.
- Employee guidance. Procedures, scripts, exception handling and escalation paths for service and sales teams.
- New-hire training. A reference that shortens the time to work independently.
- AI assistants. The approved source that retrieval-based systems answer from.
- Compliance reference. The current, approved wording for statements that must be made a certain way.
A knowledge base is the wrong home for customer-specific data, such as balances or application status, which should come live from systems of record. It is also not a process engine: an article can explain how to apply for a loan, but completing the application is a different kind of system.
What a good knowledge base looks like
- An owner for every article. Accountability for accuracy is assigned, not assumed.
- Review dates. Articles are checked on a schedule and whenever the product or policy changes.
- One answer per question. Duplicates are merged, and related articles link to the canonical one.
- Written for retrieval. Sections make sense on their own, with the product, region and date they apply to stated plainly.
- Version history. Earlier versions are kept, so it is possible to show what an article said on a given date.
- Gap tracking. Searches and questions that find nothing are logged and turned into new content.
How to measure a knowledge base
- Search success rate. The share of searches that end on an article rather than a new search or a contact.
- Accuracy on audit. The share of sampled articles that are fully correct and current.
- Freshness. The share of articles past their review date.
- Gap rate. Questions asked by customers, employees or AI systems that no article answers.
- Contribution to resolution. How often an article was involved in a question being resolved without a follow-up contact.
Rules for help content in regulated industries
Help content about financial products is customer communication, and some of it counts as advertising. The main US points follow; insurers, lenders and other regulated sectors have further rules of their own.
- Credit terms in help content. Under Regulation Z, stating certain terms when advertising closed-end credit, such as a down payment, the number of payments or a payment amount, triggers further disclosures, including the annual percentage rate. A help article promoting a loan can count as advertising.
- Savings rates. Under Regulation DD, an advertisement that states a rate of return on a deposit account must state it as an annual percentage yield.
- Delivering required disclosures. Posting information in a help center does not deliver a disclosure the law requires. Under the E-SIGN Act, required consumer disclosures may be provided electronically only after the consumer has consented to electronic delivery.
Frequently asked questions
What is the difference between a knowledge base and a knowledge graph?
A knowledge base is mainly a collection of documents and articles. A knowledge graph stores facts as connected entities and relationships, such as a product linked to its fees and eligibility rules, so software can reason over them.
Can AI write knowledge base articles?
It can draft them from tickets, transcripts and product documentation, which speeds up coverage. A subject-matter owner still needs to review each one, since an error in the knowledge base spreads to every answer built on it.
How often should articles be reviewed?
On a fixed schedule set by risk, often quarterly for pricing and policy content, and immediately whenever the underlying product or rule changes.
Is a knowledge base enough to run an AI agent?
No. It gives an agent facts to answer with, but not the process, decisions and actions needed to complete a task.
What is a knowledge gap?
A question that people or systems ask often but that no current article answers. Tracking gaps is the most direct way to decide what to write next.
Related terms
Knowledge answers. Agents finish.
Encore AI's agents do not need a knowledge base built for them.
Scale your best to be the best.
Book a Demo