Get Info services to complete all security checks and typical data gathering, saving your agents the time.
Configuring a Get Info Assistant
After creating a Natterbox AI component from the Routing Policies, simply select the Get Info App.
Set the initial message to the customer
Treat the Get Info Assistant like a receptionist
Give them a name
Provide the welcome
Clearly state you need obtain some information from the customer
Example “Good morning, thank you for calling Natterbox. My name is Alice and I'm an AI, before I get you assistance I will need a few details from you”
Define all the information that is needed within the goal prompt - see more on the goal prompt here
Set the message to the customer once all information has been captured
Set a friendly message for when the customer is off handed
Define a variable for each detail being captured
This is how you can dynamically use the information as part of the routing policy (i.e. to perform lookups in Salesforce and/or present information to the agent ahead of answering a call)
Click ‘Add Variable’
The Variable Name must not contain any spaces
Choose the most appropriate data type to match the information being captured - see here for examples and use of the data types
Date (i.e. a date of birth)
String (i.e. a name)
Number (i.e. a case number)
Boolean (i.e. a yes or no answer)
Selection (i.e. a single selection from a list pre-defined options)
Multi Selection (i.e. a one or more selections from a list pre-defined options)
Set the format
Depending on the data type, you need to define how the data will be formatted once it’s captured
Date - YYYY-MM-DD
String - Regex
Number - 0-1000
Boolean - true or false
Selection (i.e. Starter Tier, Standard Tier, Enterprise Tier)
Multi Selection (i.e. Product 1, Product 2, Product 3)
Use the Variables within the Routing Policy - see the Quick Start Guide for a full example of this
Defining the Goal Prompt
Make the goal prompt clear and specific about what the assistant should achieve. The default goal prompt is:
Your goal is to efficiently gather information from a Customer. You must:
- Ask for each item individually
- After receiving each item, repeat it back and ask the Customer to confirm it.
- **Do not ask for the next item until the current one is confirmed.**
- Wait for confirmation before continuing.
- If the Customer declines to confirm, politely ask them to repeat or clarify.
- Only proceed with validation once all required fields have been collected and confirmed
This prompt reinforces a structured and step-by-step interaction model where each piece of information is validated before moving on. The assistant should clearly explain what it is collecting and why.
You can customize this goal for your use case. For example:
Your goal is to efficiently gather information from a Customer so that you can verify their identity using their Full Name, Date of Birth and Postcode.
Variable Data Types
When setting a variable data type follow these best practices :
Provide a clear, meaningful variable name (e.g.,
CustomerFullName) that helps guide the assistant in referencing the item.Include a 2–3 sentence description that defines the variable’s purpose, expected structure, and any formatting rules.
Optionally, set the variable type (e.g., Text, Date, Number) to further guide the assistant.
If using the Text type, you can specify a pattern using regular expressions. The assistant will validate user input against this pattern and ask again if the input doesn’t match.
Example Variable Definition
Name: CustomerFullName
Description: The customer’s full name, typically including first name, optional middle names, and last name. It should be a single string in natural name order (e.g., "Jane Alexandra Smith"), without abbreviations, initials, or titles. This field is used for identification, personalization, and verification purposes.
Why this is a good description:
It clearly defines the expected structure, includes an example, and explains why the data is needed. This helps the assistant gather the information accurately and ensures consistent handling across customer interactions.
Handling References and Codes
Information like part codes or booking references can cause problems if not handled correctly. When a customer speaks a code, the AI engine may receive it either as a single string or as fragmented segments. For example, the code AB1234R might be received in any of the following ways:
ab1234ra b 12 3 4 ra b 1 2 3 4 r
This inconsistency can cause issues unless it's properly managed, including:
The AI engine may fail to recognize the code accurately.
If you're passing the code to a backend tool or macro, it may be formatted incorrectly, causing failures downstream.
The Problem with Strict Format Expectations
If we take the following Goal prompt as an example:-
You are in the role of a specific Assistant. Continue to use the name and personality traits defined in ##PERSONA##.
You are an expert agent for an events business. Your job is to get booking references from the caller so that they can pass security. Booking references should be 2 letters, followed by 4 numbers, followed by 1 letter. If the reference you hear does not follow this format, then assume it is incorrect and ask the caller to repeat their booking reference.
It is super important that you get their booking reference right so once you have a reference you believe is valid, repeat it back to the caller and wait for them to confirm before completing your task.
This defines a booking reference as Booking references should be 2 letters, followed by 4 numbers, followed by 1 letter. This prompt is likely to fail in practice because the AI may expect the code to be received exactly as "AB1234R" and reject alternative, but valid, spoken variations.
Recommended Prompt Strategy
To make your AI more robust, improve your Goal prompt to account for realistic input. For example:
You are an expert agent for an events business. Your job is to collect a valid booking reference from the caller. A valid booking reference always has this format: 2 letters, followed by 4 numbers, followed by 1 letter — for example: AB1234R.
The customer might say the reference with pauses (like “A B 1 2 3 4 R”) or in groups (like “AB12 34R”). That’s expected. You should combine the characters into a single 7-character code, with no spaces, and convert it to uppercase.
-You may silently clean and normalize the input to combine it into the correct format.
-NEVER say or imply that you are reformatting, converting, or cleaning the reference.
-NEVER mention “spaces”, “formatting”, or “normalization” to the customer.
If the customer’s input cannot clearly be interpreted as a 7-character reference in the correct pattern (2 letters, 4 digits, 1 letter), ask them to repeat it.
Once you have a reference that matches the correct format, repeat it back as a single string and wait for the customer to confirm before proceeding.
The improved prompt is more effective because it is designed with real-world speech variability in mind. It includes a clear example (AB1234R) to set the expected format, and explicitly acknowledges that customers may speak references with pauses or in grouped segments (e.g., “A B 1 2 3 4 R” or “AB12 34R”). This helps the AI handle different delivery styles without misinterpreting valid inputs.
It also instructs the AI to silently normalize the input—combining characters and converting to uppercase—without ever mentioning formatting or processing to the customer. This avoids breaking the conversational flow or causing confusion. Finally, the instruction to repeat the reference back and wait for confirmation adds an essential validation step to ensure accuracy before continuing.
Reinforcing in Tool Descriptions
You should also reinforce these expectations in your Variable descriptions. For example, in the Defining Variables section, update the variable descriptions as follows:
Variable Name Description:
A booking reference to be validated. Booking references are 7-character codes consisting of 2 uppercase letters, followed by 4 digits, and ending with 1 uppercase letter.Validation Regex:
Use this pattern to enforce strict validation in tools or macros:^[A-Z]{2}[0-9]{4}[A-Z]$
Data Type Examples
Date - (Date of Birth)
The Assistants are able to detect and capture dates in the many different ways your customers may communicate them but to ensure you’re able to perform a lookup match you will need to pick the same format of how you store you date data.
String - (Post Code)
For most strings, such as First name and Last name, you can keep the pattern matching empty to ensure you capture everything as-is, but in certain situations such a Post Code you need to ensure data is transformed and matches the format in your CRM.
The Pattern field expects a Regex.
Here are some useful Regex examples :
UK Post Code
^([A-Z]{1,2}[0-9][0-9A-Z]?\s?[0-9][A-Z]{2})$US Zip Code
^\d{5}(?:-\d{4})?$AU Post Code
^\d{4}$US Social Security Numbers
^(?!666|000|9\\d{2})\\d{3}-(?!00)\\d{2}-(?!0000)\\d{4}$AU Tax File Number
^\d{9}$Number - (Support Case)
Just like String, if you’re looking for a general number you can leave the minimum and maximum values blank, but if you know the number has a specific range then you can set that range to reduce the likelihood of false positive hits you may encounter.
Boolean - Existing customer
Using a boolean variable is a create way to capture a yes or no type answer to a question, such as whether your customer already has an account with your business.
In this example you use this information to change the destination or priority of the call based upon your business needs.
Selection - (Starter Tier, Standard Tier, Enterprise Tier)
Selection is a great way to provide your customers with some set options to select one from, for example determining what product tier they are currently on or looking to upgrade to.
Note remember to ask your AI to describe all the available options before getting an answer.
To add an entry simply type it into the selection field and click ‘Add Tag’.
Multi Selection - (Product 1, Product 2, Product 3)
Just like the selection variable, the multi selection acts the same but serves the ability to choose multiple options rather than just one. For example you the Assistant could gather all the various products they are either currently using or are interested in hearing more about
