How to define an intent?

A guide for creating intents so that the AI-Chatbot can sufficiently answer the queries.

When creating intents or considering which should be answered by the chatbot, the following must be taken into account:

A typical intent consists of an overarching intention of the user. This intention can be recognized based on typical user queries. The goal when creating intents is to ensure that the Chatbit is able to recognize the user's intention and respond appropriately.

Do the five typical user requests represent the same intention?

In the following example the typical user requests do not represent the same intention: 
Intent to be created: Change delivery
Five typical requests:

  1. I want to change the delivery date.
  2. I want to change the address.
  3. I would like to receive my delivery earlier. 
  4. How long does it take for my package to be delivered?
  5. I would like to cancel my delivery.

User requests 1 and 3 have the same intention: 
The users want to change the delivery date.

Requests 2, 4 and 5 differ in the intention of these requests:
Request 2: The users want to change the delivery address.
Request 4: Users want to know how long a delivery will take.
Request 5: Users want to cancel their order.

Even if the answers for these intents could be bundled thematically from the company's point of view, as they are similar or even the same, this must be viewed from the perspective of the AI. 
The AI reacts based on the incoming user requests and not the response content. Therefore, several intents should be created to ensure that the bot is able to correctly recognize the user's intent.

A positive example of what one of these intents could look like:
Intent to be created: Change delivery date
Five typical user requests:

  1. I would like to change the delivery date of my order.
  2. Could I have my order delivered earlier?
  3. Is it possible to specify a different delivery date?
  4. The delivery date is too late for me.
  5. How can I postpone the delivery date?

Intents that contain multiple intentions, as in the example above, i.e. where the 5 typical user enquiries are thematically different from each other, are a kind of catch-all. They prevent the queries from being recognized properly and the correct answer from being displayed. There may even be intents that have already been created that would cover one of the enquiries.


If several intentions are summarized in one intent, the performance of the chatbot deteriorates.


In addition, the AI would recognize over time from the enquiries received that separate intents are needed here. The AI suggestions then result from this.

All intentions can be covered with more specific subdivided intents.
It is therefore advisable to create individual intents that imply the corresponding variants. It is then of course possible to create the same answer for all of them.


Product names are only interchangeable variables for the AI, which do not influence the intention behind the question.


Accordingly, intents that differ only by individual words can be merged and divided into the individual subcategories within the answer using the multiple choice selection.
However, creating a new intent with very similar questions would have a negative impact on AI comprehension.