> ## Documentation Index
> Fetch the complete documentation index at: https://docs.recepta.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Caller types

> Configuring what the agent asks and does for each kind of caller.

Not every caller wants the same thing. Recepta.ai sorts inbound callers into four types and follows the configuration you set for each — so a new lead gets qualified while a supplier gets handled in ten seconds.

## The four types

<AccordionGroup>
  <Accordion title="Potential customer" icon="user-plus">
    Someone who wants work done. This is the type worth the most configuration effort — it's where revenue comes from.

    The agent can collect contact details, service address, the service they need, custom qualifying questions, and anything else you specify, then book the job or hand off a fully-qualified lead. It can also push the new contact into your connected CRM.
  </Accordion>

  <Accordion title="Existing customer" icon="user-check">
    Someone you already work with — chasing an appointment, asking about an invoice, reporting a problem with recent work.

    The agent collects identifying details and the reason for the call, and can write back to your CRM. With a [field-service CRM](/integrations/field-service-crm) connected it can recognize them from their number and reference their job history rather than asking questions you already know the answer to.
  </Accordion>

  <Accordion title="Sales call" icon="bullhorn">
    Someone selling to you. Configure a short, polite close — take a name and callback number if you want them logged, or decline and end the call. Either way it's off your team's plate.
  </Accordion>

  <Accordion title="Other" icon="circle-question">
    Everything else: wrong numbers, job applicants, a neighbor about a parked van. The agent collects contact details and the reason, and routes or takes a message per your setup.
  </Accordion>
</AccordionGroup>

## What to collect

For each type you decide which details the agent gathers. The temptation is to ask for everything; resist it.

<Columns cols={2}>
  <Card title="Ask for what you'll act on" icon="circle-check">
    Name, phone, service address, and the job — enough to quote and dispatch. Add one or two qualifying questions that genuinely change how you'd handle the job.
  </Card>

  <Card title="Don't run an interrogation" icon="circle-xmark">
    Every extra question is another chance for the caller to give up. A field nobody reads later is a field that shouldn't be asked.
  </Card>
</Columns>

<Tip>
  Custom questions earn their place when the answer changes your dispatch — "is the unit accessible from outside", "single or two-story", "is there standing water right now". Preferences you'd ask on site do not.
</Tip>

## Emergencies

Emergency handling belongs in your [transfer rules](/receptionist/transfer-rules), not in caller-type configuration. A burst pipe should reach a human immediately, whatever branch of the flow it surfaces in.

## Getting the classification right

The agent infers the type from what the caller says. If you see calls landing in the wrong bucket in your [call logs](/receptionist/call-logs):

* Make your service list more specific in the [business profile](/receptionist/business-profile) — vague services make "potential customer" harder to recognize.
* Add the terms your customers actually use to the [knowledge base](/receptionist/knowledge-base).
* Connect your CRM so returning customers are identified by number rather than inference.