What the Post-call tab of the AI4CALL assistant does
The call ends, the work does not. The Post-call tab handles everything that happens after the conversation: from here you can extract structured data and send it to a webhook, or use tools to have the assistant carry out follow-up activities. These are two distinct modes: the first delivers information to an external service, the second lets the assistant perform actions once the call is over.
In this video tutorial we start with the webhook and move on to Post-call with tools, following step by step the examples shown in the configuration screen.
Post-call with webhook: extracting structured data from the call
A webhook is the address of the service that should receive the result: enter it in the dedicated field. With Add property you define each piece of information to extract from the call. In our example we ask for the reason for the call, using the String type: the description tells the assistant to summarise it in one sentence and not to invent details. If the data does not come up in the conversation, the result must flag it as not available.
Data types: String, Number, Boolean, Enum and Object
The type defines the shape of the data the assistant returns to the receiving service, and therefore how that service will be able to read it.
- String: free text.
- Number: a numeric value.
- Boolean: true or false.
- Enum: limits the choice to predefined values.
- Object: groups several properties into a single structure.
The Required checkbox means the field must be present in the result. It does not authorise the assistant to invent its content when the conversation does not provide it.
Enum example: classifying the request
The category property can contain information, appointment, support or unclassifiable: that last value covers calls where no clear category emerges. Write in the description when each value should be chosen, because a controlled list makes routing requests inside the receiving system far easier.
Object example: grouping the request data
With Object you group the data of a request into a single structure: in the example we place the reason for the call inside it. Each inner property also has a name, a type and a description. Use this structure when the receiving service requires an object with several fields, and agree names and types with whoever manages that integration.
The identifiers sent in the headers: caller, called and callid
The screen also shows the identifiers transmitted in the headers. Caller identifies who called, called the recipient, callid the individual call: these are the values that link the result to the correct conversation. Remember to replace the address shown in the examples with the one of your own service.
Post-call with tools: actions after the call
Post-call with tools lets the assistant carry out activities once the conversation ends. Turn on the feature and select from the catalogue the actions you want to make available, using the filters to find the tools you need. The instructions guide the model in choosing which actions to run after the call.
Example: an email summary with the send email tool
In the example we select the send email tool of the SMTP instance: the goal is to send the company recipient a summary that helps handle the request. Before using it, check that the chosen instance is configured and that the recipient is the right one.
In the instructions we ask for a single email, covering the reason for the call, the confirmed data and the next action. We also add that it must not include details that did not come up: this precision is what prevents repetitive messages or summaries that attribute requests the customer never made.
Testing and saving your settings
You can use the main model or a configuration dedicated to Post-call. The test is there to check the result, but it can actually run the selected tools: with an email tool, that means genuinely sending the message. Check recipients and instructions before running it. When you are done, save the assistant to make the settings effective.
Post-call at a glance
Webhook: the address of the receiving service and the properties to extract (String, Number, Boolean, Enum, Object), with descriptions that forbid inventing details. Headers: caller, called and callid link the result to the correct conversation. Tools: actions chosen from the catalogue and guided by your instructions, such as sending a summary email. Before testing, check recipients and instructions; when finished, save the assistant.
Frequently asked questions about Post-call
What is the difference between webhook and tools in Post-call?
The webhook sends the structured data extracted from the call to the address of an external service. Tools, instead, let the assistant carry out activities after the conversation, such as sending a summary email with the send email tool of an SMTP instance.
What does the Required checkbox mean?
It means the field must be present in the result. It does not authorise the assistant to invent its content when the conversation does not provide it: if the data does not come up, the result must state that it is not available.
Does the test actually run the selected tools?
Yes. The test is there to check the result, but it can actually run the selected tools. With an email tool that means genuinely sending the message, so check recipients and instructions before starting it.
Let's configure your Post-call together
Want to work out which data to extract from your calls and which actions to automate afterwards? Let's talk, with no commitment: we'll show you the AI4CALL assistant on your real use cases.