Conversation
Management System
The next step after the Content Management System.
The name says what it does. Where a Content Management System lets you manage content, a Conversation Management System lets you manage the conversations: the knowledge behind them, the instructions, the actions and the rules the AI has to follow.
How the web evolved
Every twenty years the web makes the same move. Something that used to be locked in the code comes loose and becomes manageable. Anyone who lived through the arrival of the CMS will recognize what is happening now.
-
1
1991 to 2003
Static websites
Code = Content
Every page was handwork. Changing a sentence meant changing the source code.
-
2
2003 to 2024
Content Management System
Code ≠ Content
Content becomes manageable. The organization manages pages, the developer manages the system.
-
2024 to now
Generation 3
Conversation Management System
Code ≠ Knowledge
The website becomes generative. Intent takes center stage. Not pages, but conversations.
Why a CMS on its own is no longer enough
A CMS is designed around the structure of a document: pages, menus, navigation and hyperlinks. That structure assumes the visitor will find their own way. Most of the time they do, until a question does not fit on one page.
AI looks at something else: what someone wants, in what situation, which knowledge fits and what needs to happen. None of that is described in a CMS, because nobody built a CMS for it.
A CMS manages
- Pages
- Menus
- Navigation
- Hyperlinks
AI works with
- Intent
- Context
- Knowledge
- Actions
The visitor searches
Clicking on, hoping the answer is on the next page
- Home
- Products and services
- Applications and forms
- Search results (34)
Four steps, still no answer.
The visitor says what they want to achieve
One intent, one answer, with a citation
I want to report a change of address, but I am subletting.
Answer by vragen.ai
You can do that online. You will need DigiD and the date you are moving. If you sublet, you report it with a statement from the main tenant. I can open the form for you.
What is a Conversation Management System?
A Conversation Management System is the software layer where an organization manages its knowledge, conversations, instructions, actions and AI governance. The CMS decides what is on a page. The Conversation Management System decides what happens in a conversation.
That layer sits between the website and the systems behind it. It decides which knowledge the AI may use, which model processes that knowledge, which actions the conversation may carry out, and it records what happened.
Website
Conversation Management System
Intent, context, rules, answer
- Knowledge
- AI models
- Business rules
- Skills
- Integrations
- Monitoring
- Analytics
- Human in the loop
CMS versus Conversation Management System
The same website, but you manage something else.
| Topic | Content Management System | Conversation Management System |
|---|---|---|
| Managing | Content | Conversations and knowledge |
| Structure | Pages | Intents |
| Interaction | Clicking | Asking questions |
| Search | Navigating | Understanding |
| Success | Pageviews | Goal reached |
| Personalization | Segments | Context |
The building blocks of a Conversation Management System
Ten components that you set up separately, and that only produce a usable answer together.
-
Conversation engine
Manages how a conversation unfolds: the session, the context and what was said earlier.
-
Intent recognition
Works out what someone wants to achieve, instead of just the words they typed.
-
Knowledge layer
Makes content, documents and data sources usable as knowledge an answer is allowed to rest on.
-
Skills
Bounded tasks the system can carry out, such as starting an application or running a calculation.
-
Business rules
Set out what the AI may answer, where it may refer people on, and what it may carry out.
-
Integrations
Connect the conversation to the CMS, the CRM, the case management system and existing APIs.
-
Analytics
Show which goals visitors have and whether they actually reach them.
-
Monitoring
Tracks quality, cost and model behavior over time, with an auditable trail.
-
Human in the loop
Keeps editors and staff in control: reading along, correcting and taking over.
-
Widgets
The places where the conversation becomes visible: website, app, portal or intranet.
How the components work together
Every question follows the same route. You set up each step separately.
- Question
- Intent
- Knowledge
- Rules
- Action
- Answer
- Insight
Why this matters now
AI is not new. What is changing is what visitors have got used to. They ask a question somewhere else and get an answer, and they expect the same on your site. Four developments are making the page a less obvious place to start.
-
AI search
Search engines return answers instead of links. The click through to a website is no longer the starting point.
-
AI agents
More and more often an agent reads the site instead of a person, to arrange something on someone's behalf.
-
Generative interfaces
The AI assembles the screen per question. A fixed page is then one of several possible outcomes.
-
Systems that act
Systems increasingly settle things between themselves. An organization needs to decide what they may do.
Navigating was never what your visitors wanted. It was the only way to reach their goal.
Where it is used
It works in any sector, because the shape of a question does not change. What does differ is the knowledge underneath, and how far an organization wants to go in what the AI may do.
-
Government
Residents describe their situation and end up at the right scheme or application.
-
Healthcare
Patients and their families get explanations from approved protocols and patient information.
-
Publishers
Readers search the archive with a question and get the original article alongside it.
-
Business
Visitors compare products and services in their own words instead of through filters.
-
Knowledge platforms
You open up a large documentation set without changing its structure.
-
Customer support
The conversation handles recurring questions. The rest goes to a colleague with the context attached.
Frequently asked questions
Is this the same as a chatbot?
No. A chatbot is an interface. A Conversation Management System is the layer underneath it, where you manage the knowledge, the rules and the actions. A chatbot can be one view of it, the way a page is one view of content in a CMS.
Do I still need my CMS?
Yes. The CMS remains the place where content is created and managed. A Conversation Management System uses that content as a knowledge source and adds a layer for conversations. The two do not replace each other.
Does this work with Drupal?
Yes. The connection runs through the existing APIs and the sitemap. Drupal's content structure stays as it is; the conversation layer reads that structure and keeps the knowledge current.
Does this work with WordPress?
Yes, on the same principle. A Conversation Management System is independent of the CMS: it connects to the content that is already there, whichever system publishes it.
Can you use several AI models?
Yes. The choice of model is a setting. You use a different model per task, and you swap out a model without setting up the knowledge, the rules or the skills again.
Is RAG the same thing?
No. RAG is a technique for retrieving knowledge when a question comes in. A Conversation Management System is the place where you set that technique up, bound it and measure it. RAG is one component of it.
How is this different from an AI assistant?
An assistant helps one user with their own work. A Conversation Management System manages the conversations an organization has with its audience: the same knowledge, the same rules, and traceable afterwards.
In practice
vragen.ai is a Conversation Management System
We built vragen.ai along these lines: a management layer for knowledge, conversations, instructions, actions and governance.
Answers come only from the organization's own content, with a citation. Editors stay in control of what may be said.
- Answers only from your own content, with citations
- Knowledge and instructions you manage yourself, without development work
- A choice of models, and a setup in which the whole chain runs in the Netherlands
- Insight into intents, goals and where visitors get stuck
Where we are today
vragen.ai follows these principles. Knowledge management, citations, instructions and insight all work today. We are building on what comes after that. Our roadmap says when we expect it. If you want to know what is possible today, just ask us. We keep developing, so this page always trails a little behind.