5 tips for beginners to perfect their Dialogflow agent

1. Create a simple agent and get straight to the point

A smart agent is one that is trained to process input data clearly, accurately, and as quickly as possible.

Think of yourself as the director and the agent as your new assistant. No matter how impressive the assistant is, if they don’t help save you time and instead make things more complicated, they won’t be effective.

Consider the following conversation example:
User: "Find me a good Korean restaurant nearby."
Agent: "Okay, where are you located?"
User: "I’m in Thu Duc District."

Now, let’s analyze this conversation.

Did you notice something wrong with the agent’s response? The agent is trying to extract additional information from the user, but this approach can lead to vague or inaccurate responses, making the conversation less efficient.

A smoother conversation would be:
User: "Find me a good Korean restaurant nearby."
Agent: "Okay, can you provide me with your address?"
Or even better:
Agent: "Okay, which street and district are you in?"

By rephrasing the question, the agent avoids misunderstandings, reduces unnecessary back-and-forth, and makes the interaction smoother and faster.

2. Use clear and precise parameter names

Parameters play a crucial role in training an agent in Dialogflow. However, many developers struggle with using them correctly, and even making mistakes in naming parameters can cause significant issues.

Naming parameters is similar to naming variables in code:

• Too long → Difficult to manage later.
• Too short → Hard to understand.

For example, if your agent collects user information, you might name the action getUserDetails. However, you don’t need to repeat "user" in every parameter, such as userName or userMail, because the action already provides the context.

Another common mistake is using generic parameter names, which can cause conflicts and errors.

For instance, suppose you need to collect two locations from a user: current location and destination.

If both parameters are named location, the system won’t distinguish between them, leading to data being overwritten and lost.

A better solution is to use currentLocation for the starting point and destLocation for the destination.

By naming parameters correctly, you minimize errors and improve agent performance. Mastering this skill requires continuous practice and refinement.

3. Be cautious with sys.any

Dialogflow offers system entities that help recognize and process common data types such as dates, numbers, emails, and addresses.

Among these, sys.any is a flexible entity that can extract any content from user input without distinguishing between data types. This makes it easy to use, but also prone to overuse, leading to unintended results.

Consider an online flower shop chatbot where users can order flowers by specifying the type they want.

If you use sys.any to capture the flower type, the following scenarios may occur:

• User: "I want to order a bouquet of roses."
• User: "I haven’t decided yet, just browsing."

In the first case, the bot correctly understands that the user wants roses. However, in the second case, sys.any might mistakenly interpret "just browsing" as a flower type, causing confusion.

Since sys.any captures any input, irrelevant responses from users may be misinterpreted, making the chatbot unreliable.

To avoid this, only use sys.any when you cannot predict the possible user inputs.

For example, when asking users for open-ended feedback:

1. Agent: "Please share your thoughts on our service."
2. User: "The service was great, but delivery was a bit slow."
   → Here, sys.any is useful because it captures the full response without requiring a specific format.

4. Set the correct lifespan

Lifespan determines how long an agent retains context in a conversation, measured in the number of exchanges.

Let’s analyze the following conversation:

1. User: "I want to send an email."
2. Agent: "Okay, who do you want to send it to?" → Intent 1: Asking for recipient information
3. User: "Send it to a@example.com."
4. Agent: "Is this the correct email: a@example.com?" → Intent 2: Confirming recipient email
5. User: "Yes, that’s correct."
6. Agent: "Okay, I will send the email now!" → Intent 3: Sending the email

In this example, the mailID parameter appears in steps 3 and 4, meaning it should persist for at least two exchanges.

• If the lifespan is too short (e.g., 1 exchange) → The mailID is deleted before step 4, preventing email confirmation.
• If the lifespan is too long (e.g., 10 exchanges) → The mailID could get overwritten if the user provides additional input later.

Additionally, some users might split their responses across multiple messages, like this:

1. User: "I want to send an email."
2. Agent: "Okay, who do you want to send it to?"
3. User: "Send it for me…"
4. User: "To the address…"
5. User: "a@example.com."
6. User: "Is that okay?"

To ensure all necessary parameters are received, a lifespan of 4-5 exchanges is recommended.

Managing lifespan effectively ensures smooth, efficient conversations and reduces errors in intent handling.

5. Don’t forget the help intent

A help intent is crucial for assisting users in understanding how to interact with your chatbot. Many users begin by asking general questions such as:

1. "What can this bot do?"
2. "I don’t know what to ask. Can you help me?"
3. "How do I use this bot?"

To handle these inquiries, configure a help intent with appropriate responses, such as:

• For a movie ticket bot:

  • "Hello! I can help you book movie tickets, check showtimes, or suggest good movies. Try saying: 'Book tickets for Avatar' or 'Showtimes for today'."

• For a banking chatbot:

  • "Hi! I can help you check balances, make transfers, and view account details. Try asking: 'Transfer money to account 987654' or 'What was my last transaction?'"

• For a food ordering bot:

  • "Hello! I’m here to help you order food. You can say: 'Order a burger and fries' or 'What’s the best dish today?'"

Including a help intent ensures that new users can quickly understand and interact with your chatbot effectively, making a great first impression!