← Back to Blog
What Is a Knowledge Graph? Definition, Examples and Uses in Customer Experience
Customer Experience10 min readNovember 16, 2026

What Is a Knowledge Graph? Definition, Examples and Uses in Customer Experience

What a knowledge graph is, its three building blocks, real examples from search engines to enterprises, how it differs from a graph database, an ontology and a knowledge base, how it helps AI give grounded answers, and how to build one for customer experience.

Listen to this article

0:00 / 10:48

Quick answer

A knowledge graph is a network of real-world things (entities) and the named relationships between them, usually organised by an ontology that defines which kinds of things and relationships exist. For example: the complaint "late refund" happens at the "refund" step, which is part of the "returns" journey. Because the connections are stored explicitly, people and AI can follow them to answer questions such as "which rule is behind this complaint?" instead of guessing.

Most company data sits in tables and documents: customers in the CRM, products in the catalogue, rules in contracts and policies, complaints in reviews and tickets. The most useful knowledge is often in the connections between them. Which complaint happens at which step of the customer journey, and which rule applies there? A knowledge graph stores those connections directly.

This guide explains what a knowledge graph is, its building blocks, real examples, how it differs from a graph database, an ontology and a knowledge base, why it matters for AI, and how to build one for customer experience.

What is a knowledge graph?

A knowledge graph is a model of knowledge that stores entities and the meaningful relationships between them. It has three building blocks:

  • Entities (nodes): the things themselves: a customer, a product, a complaint theme, a journey step, a rule.
  • Relationships (edges): named links between entities, such as "happens at", "is a kind of" or "applies to".
  • Ontology (schema): the shared vocabulary and its rules: which kinds of entities exist, how they form a hierarchy, which relationships are allowed, plus definitions and synonyms. Its rules let software infer facts nobody wrote down: if a late refund is a kind of payment issue, every late refund also counts as a payment issue.

Facts in a knowledge graph are often written as triples: subject, predicate (the relationship), object. A small, fictional example from online retail:

SubjectRelationshipObject
Late refund (complaint theme)happens at stepRefund
Refundis a step ofReturns journey
Returns journeyis the journey ofOnline order
Refund time policyapplies toRefund
Late refundis a kind ofPayment issue

Read together, the chain says: a late refund is a payment issue at the refund step of the returns journey for online orders, and the refund time policy applies at that step. No single table holds that sentence; the knowledge graph does.

Knowledge graph examples

  • Search engines: Google introduced its Knowledge Graph in 2012. It powers the knowledge panels that show facts about people, places and organisations next to search results.
  • Open knowledge: Wikidata is an open, community-edited knowledge graph that anyone can query and reuse.
  • E-commerce: product graphs link products, categories, attributes, brands and "often bought together" relationships to improve search and recommendations.
  • Healthcare and life sciences: graphs connect drugs, diseases, genes and studies, so researchers can find indirect links.
  • Finance: graphs of companies, owners and transactions support fraud detection and know-your-customer checks.
  • Customer experience: complaint themes linked to products, journey steps, rules and the reviews that mention them.

What is an enterprise knowledge graph?

An enterprise knowledge graph connects a company's own data: customers, products, processes, documents, rules and the feedback customers leave. These usually live in systems that do not talk to each other. Once they are linked, a question that spans several systems ("which policy is behind the complaints about our returns process?") becomes a query instead of a week of meetings.

Knowledge graph vs graph database vs ontology vs knowledge base

What it isBest for
Knowledge graphEntities and named relationships, with meaning usually defined by an ontologyFollowing connections and answering "how is this related to that?"
Graph databaseSoftware that stores and queries nodes and edgesRunning a knowledge graph, or any network data, efficiently
OntologyThe vocabulary: kinds of things, allowed relationships, definitions, synonymsMaking data mean the same thing everywhere
Knowledge baseA searchable collection of answers, articles, rules and documentsAnswering questions in plain language

Put simply, the graph database is the engine, the ontology is the vocabulary, and the knowledge graph is the connected content. A knowledge base is the wider idea: any store of knowledge people or AI answer from, such as articles, rules and documents. A knowledge graph is one way to structure that knowledge, and many AI systems combine the two.

Why knowledge graphs matter for AI

Large language models write fluent answers, but they do not know your company's facts, and when they do not know, they can guess. A knowledge graph grounds the model: the system first retrieves the relevant entities and relationships from the graph, and the model then answers from them. This approach is often called GraphRAG (retrieval-augmented generation over a graph); Microsoft Research published a well-known version in 2024 that builds a graph from documents and uses it to answer questions about a whole collection.

What a knowledge graph adds to AI:

  • Connections, not just similar text: it can answer questions that need several hops, such as "which complaints are linked to a rule in our terms and conditions?"
  • Explainable answers: you can show the path the answer followed and the evidence behind each link.
  • Consistent vocabulary: synonyms such as "refund", "money back" and "reimbursement" map to one concept, so counts and answers do not split.
  • Your rules, not generic guesses: the answer follows how your industry and your company actually work.

Knowledge graphs in customer experience

Customer feedback is spread across reviews, surveys, support tickets, calls and social posts. A knowledge graph for customer experience connects:

  • the themes customers talk about, such as late delivery or billing errors;
  • the products and services they are about;
  • the customer journeys and the steps where each theme happens;
  • the rules that apply: policies, terms and conditions, regulation;
  • the evidence: the actual reviews and comments behind each link.

With those links in place, a CX team can ask better questions. Where in the journey does this complaint happen? Which other issues come up together with it? Does it come from customers not knowing a rule, or from a process that does not follow it? Is it growing or fading?

Pivony Knowledge Base (PKB)

Pivony brings together your company knowledge base, the dynamics and rules of your industry, and your workflows, with inference on top. Pivony Knowledge Base (PKB) applies the ideas in this guide to customer experience. It is built on an ontology for your sector and brand: CX themes, products, customer journeys and their steps, rules and regulations, and the links between them. In the Pivony console you can:

  • Explore the ontology as a table or as a network view: start from any concept, open its neighbours and read its definition and synonyms.
  • Open the concept map of any concept: where it sits in the customer journey, which themes are complained about together with it more often than chance (with how many reviews mention both), and which rules and conditions it is linked to, with what each rule says. The reviews behind the concept are one click away.
  • Follow themes over time: each theme's status compares the last 30 days with the 90 days before, and the evidence reviews are one click away.

The same knowledge serves the rest of Pivony. Sonic Advisor answers data questions from your knowledge base, by voice or text. Pivony MCP (an add-on) lets AI assistants such as Claude ask your organisation's knowledge base why-questions: root causes, whether an issue is still happening, competitor comparisons. Sonic Engage then reaches at-risk customers on WhatsApp, starting from the root causes, through your rules and approval flow. Each organisation's PKB is set up with its Pivony account manager. Trusted by Samsung and Allianz.

How to build a knowledge graph for customer experience

  1. Start from questions. List the questions you want to answer, for example "which steps of our claims journey cause the most complaints, and which rules are involved?" The questions decide what goes in.
  2. Define the ontology. Choose the kinds of entities (themes, products, journey steps, rules) and the relationships between them. Write a definition and synonyms for each concept. Reuse your process maps and industry standards.
  3. Extract entities and relationships. Take structured data from systems such as the CRM and the product catalogue, and use AI to pull themes and links out of unstructured text: reviews, tickets, call transcripts, policies.
  4. Link and deduplicate. Map synonyms and duplicates to one entity, so "money back" and "refund" count together.
  5. Keep the evidence. Store where each fact came from, so every link can be checked against real reviews or documents.
  6. Make it usable. Store it in a graph database or similar store, and give people a way to explore it: search, a network view, a concept map.
  7. Keep it current. Give each part an owner, have experts review new rules, and update it when products, journeys or regulations change.

Common mistakes

  • Starting with the technology instead of the questions the business needs answered.
  • Modelling everything: an ontology so large that nobody maintains it.
  • No evidence: links without sources are hard to trust and harder to fix.
  • A one-off project: a knowledge graph that is not updated drifts away from how the business works.

Related reading

Next step

See how a knowledge base built on your industry's rules explains why customers complain: book a demo, or request a free root cause analysis on an export of your reviews or tickets.

Frequently asked questions

What is a knowledge graph in simple terms?

A knowledge graph is a map of facts: things such as customers, products, issues and rules, connected by named relationships such as "happens at" or "applies to". Because the connections are stored explicitly, people and AI can follow them to answer questions instead of guessing.

What is an example of a knowledge graph?

Google's Knowledge Graph, which powers the knowledge panels next to search results, and Wikidata, an open knowledge graph anyone can query, are well-known examples. In a company, a knowledge graph can link complaint themes to the journey steps where they happen and the rules that apply there.

What is the difference between a knowledge graph and a graph database?

A graph database is software that stores and queries nodes and edges. A knowledge graph is the content: entities and relationships whose meaning is defined by an ontology. A knowledge graph is often stored in a graph database, but the two are not the same thing.

What is the difference between a knowledge graph and an ontology?

An ontology is the vocabulary: which kinds of things exist, which relationships are allowed, plus definitions and synonyms. A knowledge graph uses that vocabulary and fills it with real facts, such as which complaint happens at which step of the customer journey.

How are knowledge graphs used with AI and LLMs?

A knowledge graph grounds an AI model in your facts. The model first retrieves the relevant entities and relationships and then answers from them, an approach often called GraphRAG. Answers can follow connections across several steps and show the evidence they come from.

What is an enterprise knowledge graph?

An enterprise knowledge graph connects a company's own data, such as customers, products, processes, documents, rules and customer feedback, across systems that otherwise do not talk to each other, so teams and AI assistants can answer questions that span them.

#knowledge graph#what is a knowledge graph#enterprise knowledge graph#knowledge graph examples#ontology#graphrag#customer experience

See root causes on your own feedback

Book a 15-minute demo, or get a free RCA in 48 hours with no sales call.