How it works

Everything in Caller revolves around one idea: a recording comes in, it is turned into text, and that text is compared with what you have said matters. This page explains that path and the rules that repeat on every screen.

1. The recording comes in. You upload it from the app, or your telephony system sends it through the API. At that moment the minutes of audio are reserved and the call is filed with its agent, its date and its campaign, if they can be recognized.

2. It is transcribed. The audio becomes text, with the voices separated and each sentence marked with its time.

3. It is analyzed. The AI reads the transcript with the template that applies to the call. It works out which voice is the agent and which is the customer, and returns a title, a summary, the main topic, how the call ended, how the customer felt and the points for each criterion.

4. It is ready. The call appears with its score. If something in it needs a person to listen, the call also goes into the To review queue.

Steps 2 and 3 happen on their own, usually within a few minutes. You do not need to click anything or keep the page open.

State	What it means	What you can do
Transcribing	It is queued or being turned into text.	Wait.
Analyzing	It has a transcript and the AI is scoring it.	Wait.
Ready	It has a transcript and a score.	Open it.
Needs a template	It was transcribed when the account did not have a template yet.	Create a template. Then, in Calls, select those calls and click Analyze again.
No analyses left this month	It was transcribed, but the period’s ceiling of analyses has been reached, which only happens when many calls are analyzed again.	It is analyzed on its own when the period renews.
Analysis failed	The AI tried three times without success.	Click Retry in the list.
Failed	The recording could not be transcribed.	Click Retry in the list.

In Calls, the In progress view gathers the calls that are still moving forward, and Failed gathers the ones that have stopped.

1. If the call is in a campaign that has a template, that template is used.
2. In any other case, the account’s Default template is used.

The default is the first template you created, until you choose another one with Use as default in a template’s menu.

Each criterion in a template is worth a maximum number of points. The score of a call is the points it earned out of the points possible, expressed on a scale of 0 to 100.

A template with four 10-point criteria adds up to 40. A call that earns 31 has a score of 78.

The same scale is used on every screen:

Score	Rating
80 or more	Good
60 to 79	Could improve
Below 60	Critical

A person can change the points of any criterion with Correct the score. The AI’s original figure is kept next to the corrected one.

For each criterion, the AI explains why it gave those points and cites the sentence in the call that it relies on. From there, Listen jumps to that moment in the recording.

A quote is shown only if that sentence exists in the transcript. If the AI returns a quote that does not appear in the text, the quote is discarded: the criterion keeps its explanation, but with no link to a specific moment.

A call goes into To review when it meets at least one of these conditions. They are listed in order of importance, and the first one the call meets is the one shown in the list.

Reason	When it applies
Mandatory criterion missed	A criterion marked as mandatory earns less than half of its points, even if the rest goes well.
Critical score	The score is below 60.
Upset customer	The customer ends the call with negative sentiment.
Unresolved	The call ends with no solution and no agreed next step.
Long silence	A quarter of the call or more is silence.

The call leaves the queue when someone opens it and clicks Mark as reviewed.

The analysis classifies the end of each call as one of four outcomes:

Outcome	Meaning
Resolved	What the customer asked for was solved during the call.
Follow-up agreed	Something is still pending and the next step was agreed.
Escalated	The call was passed to another person or department.
Unresolved	It ended with no solution and no next step.

Caller estimates how the customer feels at the start and at the end of the call: positive, neutral or negative. It does not estimate the agent’s mood, and that is deliberate: the EU AI Act (Article 5(1)(f)) prohibits inferring a person’s emotions in the workplace, so the analysis is limited to the customer.

Term	What it is
Call	A recording with its transcript and its analysis.
Template	The list of what the AI checks in a call.
Criterion	A check in the template that gives points.
Data field	Something the AI notes down from the call without scoring it, such as the reason for a cancellation.
Campaign	A group of calls (a team, a queue, a product, a client) with its template and its language.
Client	An optional label for grouping campaigns, meant for anyone who handles calls for several companies.
Agent	The person who took the call. They can have an access of their own, which shows them only their calls.
Document	A script, protocol or policy that the criteria can consult when scoring.