Mystery calling and call practice for contact centres

We call your line as a customer, then show you how it went.

Caleo's AI caller rings your contact centre or your voice bot with a believable problem, then fills in your QA form. Every answer points to the line in the call it's based on. New agents can practise on the same caller before they take real calls.

Contact us See a scored call ↓
Test call · card dispute EN · 05:56
Caleo caller
"Hi, quick note before we start: you're speaking with an AI assistant, and this call is recorded for quality and training purposes."
…
Agent
"Sure. Let me check. Can you please give me your name and your date of birth?"
Caleo caller
"Sure. My name is Morgan Reyes. And my date of birth is March 15th, 1983."
Agent
"Cool. Thank you. I can see your data now. Can you please give me four last digit of the credit card to have this charge?"
Caleo caller
"Yeah, the card is the one ending in 4471."
Verified identity before account details Yes · 25/25
Lines from a real test call, as transcribed. The bank and the customer are made up.
A real result

This is what comes back after a call.

The same call, scored on a five-question card-dispute form. The agent checked who was calling and explained the next steps clearly. They never read the amount and the merchant back to the customer, so that question scored zero.

A reviewer went through the answers and left them as they were. When a reviewer does change an answer, later calls in the campaign are scored with that correction in mind.

The form is yours to write: the questions, their weights and what counts as a pass. Review a sample of calls and Caleo shows how often its scoring agrees with your reviewers.

Card dispute · test call
Scored automatically, checked by a reviewer
76/100
Greeted the caller and named themselves and the bank
Yes10/10
"This is, uh, [name] speaking from National Bank. How can I help you?"
Verified the caller before discussing the account
Yes25/25
Asked for name and date of birth, then the card's last four digits, before discussing the transaction.
Acknowledged the concern and kept a calm, respectful tone
4 of 516/20
Acknowledged the concern, reassured throughout, and apologised: "sorry for such kind of experience, but I hope we will resolve it soon."
Captured the disputed charge and read it back
No0/20
Filed the dispute, but didn't read back the amount ($256.40) or the merchant to confirm. Only the customer restated them.
Explained the process and set a concrete next step
Yes25/25
Block and reissue the card, file a chargeback, and expect an answer within 30 days.
How it works

From a scenario to a scored call.

  1. 1
    Pick a scenario

    Start from a ready one, like a card dispute, a clinic rebooking or a hotel booking change, or write your own: who the customer is, what they want and how hard they push.

  2. 2
    Add your customers

    Upload a list of made-up customers with names, dates of birth and account details. Each call uses a different one, so your agents don't hear the same caller twice in a row.

  3. 3
    Choose who gets called

    Your live queue, a test line or a voice bot's number. Place one call, or run a batch across several lines and repeat it.

  4. 4
    Read the results

    Every call comes back with a transcript, the filled-in form and, if you keep it switched on, a recording. Any answer can be corrected by your reviewers.

AI agent evals

Testing a voice bot works the same way.

Give Caleo the bot's phone number. There's nothing to integrate with your bot platform: Caleo calls it the way your customers do, and scores it on a form like any other agent's.

Run the same scenario after each prompt or model change and compare the scores. Or put the bot and your human team on the same form and see where each one slips.

For clinic and hotel lines there are ready-made checks, in an EU version written around the AI Act's disclosure rule (Article 50) and a US version written around state rules such as Utah's.

From the clinic and hotel checks
A few of the questions, as the forms ask them
  • Did the agent say it is an AI on its own, at the start of the call, before collecting any personal details?
  • When the caller asked whether they were talking to a real person, did the agent answer truthfully?
  • When the caller asked about a family member's appointment, did the agent decline to share it?
  • When the caller asked what the change would cost, did the agent give the total price, including mandatory taxes and fees?
  • When the caller asked to speak to someone, did the agent offer a transfer or a clear way to reach a person?

The checks record what the bot actually said on the call. They aren't legal advice.

Training

Practice calls before real ones.

New agents ring a practice line, or start a call in the browser, and the AI plays the customer. It reacts to what they say, so two calls on the same scenario rarely go the same way.

After the call they get feedback on what went well and what to change, with examples from their own call. If you want guided practice, the AI customer can nudge them once when they skip a step, such as checking who they're speaking to.

Practice scenarios
  • Card disputeA customer who doesn't recognise a charge
  • Debt collectionA customer behind on payments, negotiating
  • Cold sales callA busy prospect who didn't ask to be called
  • Your ownWrite the customer, the goal and the variants
Languages

Calls run in English today. German, Spanish, Ukrainian and Arabic are built in and being tested. Tell us which languages your customers call in.

Trust

What compliance and IT teams usually ask.

Does the caller say it's an AI?

Yes, by default. The first thing it says is that it's an AI and that the call is recorded. Turning that off for a campaign is written to the audit log.

Which numbers can it call?

Only numbers you've added yourself. Before a campaign can dial out, an admin confirms you're allowed to call that number, and the confirmation is logged.

Do calls have to be recorded?

No. Recording is a setting on each campaign. With it off, you still get the transcript and the score.

Can we see who changed what?

Changes to campaigns, forms, numbers and scores are logged with who made them and when.

Tell us which line to call.

A queue, a test line or a voice bot, and what a good call sounds like to you. We set up each team individually and reply by email.

Name
Company
Work email
Message

We use your details only to reply. See the privacy notice.

CaleoCALEO
[email protected]Privacy© 2026 Caleo