Build a detect intent model

Detect intent models use natural language processing (NLP) to understand what your customers are talking about and provide them with the information that they need, you’ll first need to set up intent recognition.

Step one: Choose a topic for your intent model

Begin by choosing a single topic, such as “Exchange” or “Return” for your intent model.

By choosing a single topic, you can keep your model focused and organized. You can create more models for your other intents later.

Select at least two of your most frequently asked questions within the topic that you’ve chosen.

For example, if your topic is “Returns”, your two most commonly received phrases might be:

  • “I want to return my product”

  • “How do I return my product?”

More examples:


Plant care

Questions about common pests and diseases that might affect your customer’s houseplants, and strange plant symptoms.


Inquiries from customers who want to purchase plants or plant accessories.


Complaints about service and orders.

Step three: Create match phrase variations

Next, it’s time to generate variations of these intent-matching phases. Matching phrases are example questions or inquiries that you might expect to receive from your customers. By providing our machine learning model with these example training phrases, you will train it to analyze and categorize incoming messages as accurately as possible. Think about the different ways in which your customers might phrase their intent.

For example, the phrase “I want to return my product” might also be phrased like these questions:

  • “How do I return my product?”

  • “I need to return this product”

  • “Return product”

Come up with as many variations of each intent as you can think of. Ten is a good number to aim for!

More examples:

Training phaseIntent

I’d like to buy a monstera


What small houseplants can I buy?


What plants are suitable for a bathroom with no natural light?


I have tiny black flies in my soil

Plant care

Why is my cactus mushy?

Plant care

My ficus keeps dropping leaves

Plant care

My order has missing items


I never received my order


When my plants arrived they were already dead


Step four: Define irrelevant content

Irrelevant content is any phrase that your customer might write that doesn’t correspond to the intents in your model.

By default, the model will automatically categorize a question as “irrelevant” if it can’t find a matching intent.

You can choose to add irrelevant content when you’re training your model, to provide it with examples of phrases that you don’t want it to detect.

Step five: Test your model

Once you’ve built your model, you can test your model to see how well it works. If you don’t get the results you expect, you can go back to your intents and refine them to improve the accuracy of the intents the model provides.

Step six: Use your detect intent model

Now that you’ve created your detect intent model, you can deploy this model using the detect intent action within Flows.

Last updated