Technology

AI agent or scenario: which to choose for a call campaign

← All articles

AutoCall.kz has two ways to talk to a customer without a live operator: an interactive scenario and a voice AI agent. Ads call both of them “smart calling”, and from the outside the agent looks like a slightly better scenario. It isn’t. They are built in fundamentally different ways, and the choice between them is not about “pricier or cheaper”. It is about whether you know in advance what people will answer.

A scenario runs on rails

🗺️
Interactive scenario
You draw the graph, the contact follows one of its branches

You build the flow in advance: play a recording → recognize the answer → if they said “yes”, go here, if “no”, go there, if they stayed silent, ask again. Every fork is drawn by your own hand. The call always follows one of the paths you planned for.

Besides playing audio and recognizing speech, a scenario can catch phone keypresses, connect the contact to a live operator, record an answer like a voice recorder, send an SMS to the contact or to you, and ping your system with an HTTP request right in the middle of the call.

What a scenario sounds like
Service“Hello! You have an appointment tomorrow at 2:30 PM. If you’re coming, say ‘yes’. If you want to reschedule, say ‘reschedule’.”
Contact“Yes, I’ll come.”
Service“Thank you, see you then. Goodbye.”
And if the contact says “Can I come, but a bit later, around five?”, the scenario will hear the word “come”, count it as “yes” and say goodbye.

An agent works from instructions

🎧
Voice AI agent
You describe the task in words, the agent runs the conversation on its own

There is no graph. You explain to the agent who it is, why it is calling and what it needs to achieve, the same way you would brief a new employee. From there it talks on its own: answers questions, handles objections, asks again if it didn’t understand, and pulls out of the conversation the fields you asked it to collect.

What an agent sounds like
Agent“Hello! This is Assel from Denta dental clinic. You have an appointment with us tomorrow. Is that still on?”
Contact“Can I come, but a bit later, around five?”
Agent“Of course, let me check. There’s a free slot at five p.m. Would that work for you?”
Contact“Yes, great.”
Agent“I’ve booked you for 5:00 PM. Have a nice day!”
The report will show: appointment confirmed, time moved to 5:00 PM.

That is the whole difference. A scenario works great as long as the contact answers the way you expected. An agent is needed where there are more possible answers than you can physically draw.

Point-by-point comparison

Feature Scenario AI agent
Predictability Full: it says exactly what you drew High, but the wording is different every time
Free-form answers from the contact Only preset words Understands them as they are
Questions from the contact Doesn’t answer Answers within its instructions
Keypresses (DTMF) Yes No, voice only
Transfer to a live operator Yes Not yet
Request to your system during the call Yes Yes
SMS during the call Yes Yes
What goes into the report Branches taken Your fields + full transcript
What setup takes Build a graph, record audio A chat with the assistant
Price per minute Lower Noticeably higher

What it costs

We calculate on the SILVER plan. The numbers are illustrative, just to show the scale:

Call type How it adds up Per call
Standard recorded message, 20 sec 20 sec × 0.56 ₸ 11.20 ₸
Scenario, 60 sec
with 2 recognitions
60 sec × 0.56 ₸ + 2 × 7 ₸ 47.60 ₸
AI agent, 120 sec 120 sec × (0.56 ₸ + 1.15 ₸) 205.20 ₸

Two things matter more than the numbers themselves.

A scenario doesn’t get cheaper automatically. Every speech recognition is a separate line on the bill. A scenario with seven questions in a row easily catches up in price with a short conversation with an agent, and it is far more boring for the contact to listen to. If there are many branches and even more recognitions, compare honestly before you rule out the agent because of price.

The agent is billed for the actual conversation. You set the maximum duration (from 2 to 10 minutes), and the maximum cost of the campaign is calculated from it. When you create the campaign, we hold the expected part: what such campaigns usually spend. If the campaign runs more expensive, we hold more from your balance. Conversations are almost always shorter than the limit, and we return the rest when the campaign ends.

The practical takeaway: give the agent a sensible time limit. The difference between 300 and 600 seconds means twice as much money held at the start of the campaign, even though the actual conversation will fit into a minute and a half either way.

When to choose a scenario

  • There are few answers and they are obvious. “Confirm or cancel”, “press 1 or 2”: there is nothing to draw here, everything is already clear.
  • The wording has to be word for word. Legally significant notices, debt reminders, anything you cleared with a lawyer. The agent will convey the meaning, but in its own words, and sometimes that is not acceptable.
  • You need the phone keypad. A noisy street, a bad connection, a contact who doesn’t want to speak out loud around other people: DTMF helps where voice doesn’t get through.
  • You need to connect the call to an operator. A scenario will transfer the call to your line, the agent can’t do that yet.
  • The campaign is big and the budget is tight. On tens of thousands of numbers, the difference in price per minute becomes decisive.

When to choose an agent

  • You don’t know in advance what people will say. This is the main sign. If, while drawing the graph, you catch yourself adding a branch for “what if they say something else”, that is a job for an agent.
  • The contact will ask questions. Sales, bookings, support, winning back lapsed customers: anywhere the conversation goes both ways.
  • You need to collect free-form data. Not “yes/no”, but a name, a convenient time, the reason for saying no, a comment. The agent will pull them out of natural speech and put them into the report columns.
  • Objections matter. “It’s too expensive”, “it’s not convenient for me”, “I’ll think about it”: a scenario stops there, an agent keeps going.
  • You need a transcript. Agent calls keep a full transcript of the conversation, and sometimes it is worth more than the result itself.

Can you combine them

Not in one campaign: the type is chosen at the start. But nothing stops you from running them in a cascade, and that is often the most sensible setup.

First, a cheap scenario across the whole list. It filters out those who don’t pick up, refuse right away or are no longer relevant. Then the agent, only for those who showed interest. The expensive tool works on the warm remainder, not on the cold list.

Technically, these are two campaigns: you export the results of the first one, put the interested numbers into a new list and run the agent on it. Through the API, it takes just a couple of requests.

How to choose in thirty seconds

Ask yourself one question: can you list right now all the answers you are going to hear?

If yes, pick a scenario: it is cheaper and more predictable. If you catch yourself thinking “well, it varies”, that’s an agent, and trying to draw such a conversation as a graph will end in a messy flow nobody wants to maintain.

Where to start

You can try both tools for free without launching a campaign. The scenario builder gives you a test call right in the browser: build a flow and hear it straight away. You create an agent in a chat with the assistant, then call yourself and talk to it live.

Fifteen minutes on both tests will tell you more than any comparison table, this one included.

Try both and choose

The test call is free, no need to launch a campaign.

Try it free →