Skip to content
EN
Demo
Platform
CallsQueues and routingIVR (phone menu)Smart transferAuto dialerBrowser-based softphoneClick-to-call
Messaging and customer dataUnified inboxForm builderCustomer portalCustomer success platform
ManagementHistorical reportingReal-time monitoringShift scheduling
Deployment and securityIn our cloudOn your serversHybrid deploymentData security
Platform →
Solutions
IndustriesRetailReal estateBanking and insuranceTravelHealthcareEducation
Customer serviceFAQ self-serviceOrder status without an agentAfter-hours coveragePeak days without overloadConversation quality assurance
SalesLead qualificationAppointment bookingRepeat salesLoyalty program
Outbound communicationPayment and overdue remindersNotifications and confirmationsPost-interaction survey
Who it's forBy company sizeBy role
Solutions →
AI agents
For customersVoice AI consultantAI consultant in chat
For employeesAI assistant for calls and salesAssistant for work questionsNew hire onboarding
TechnologyKnowledge baseSpeech analyticsAI connectorModelsData security
ServicesAI agent trainingMachine learning
AI agents →
Integrations
CRM systemsSalesforceHubSpotDynamics CRMSAP CXZoho CRMPipedrive
ERP systemsSAPNetSuiteBusiness CentralSageOdoo
Help desk and team chatZendeskFreshdeskServiceNowSlackMattermost
Messaging and socialWhatsAppViberInstagramFacebookSMSLive chat
Telephony, API, and AIAsteriskAMI, AGI, and ARIKamailioFreeSWITCHAPI and protocolsMCP connectors
Integrations →
Pricing
Company
ResourcesBlogCalculatorsComparisonsLibraryGlossary
ServicesOpen SourceImplementationDevelopmentSupportTraining
ContactsContact usAboutFacts and figures
Company →
Language: English
Demo

AI agents

Which AI model is right for your contact center

The model is the part of an AI agent that understands the customer's question and writes the answer. It determines how many inquiries AI resolves without an agent, whether a customer hangs up over a pause, and what each conversation costs. We connect cloud models and open models that we deploy on your servers, some of them free, and we choose based on how a model answers your customers' real questions: accuracy, speed, and cost. If a better model comes along, we swap it in without rewriting the agent.

Get a quotePricing
Cloudor your own serverYour choice
Paidand free open modelsMatched to the task and budget
2-3models comparedTested before choosing
$0per minute when the model runs on your serversPay for the server, not for conversations

Experience

What we test before choosing a model

Choosing the wrong model costs money: wrong answers, abandoned calls, and unnecessary spending.

Your terminologyHow the model handles the names of your plans and products and your abbreviations, not just everyday language.
Response latencyThe time from the customer's question to the first word of the answer – on a call, delay is most noticeable.
Long conversationsWhether the model loses the thread when there are many turns and knowledge base content is added to them.
An honest "I don't know"Whether the model admits there's no answer or makes up a plausible but wrong explanation.
Cost at volumeWhat a month of work costs at your conversation and message volume, not a single test dialog.
Provider updatesWhen a provider updates a model, its answers change – we re-check them against reference answers.

Tasks

Different tasks, different models

Contact center tasks vary in complexity, and one powerful model for everything usually means overpaying.

Calls and chats: accuracy and speed matter

A live conversation requires a model that understands your customers' language and your rules well and answers without delay. On a call, three models work in sequence: the first turns the customer's speech into text, the second – the language model – works out the question and writes the answer, and the third reads it out loud. We choose each one separately: for example, speech recognition runs on-premises and speech synthesis in the cloud. The higher the accuracy, the more conversations AI handles to the end without transferring to an employee.

  • Accurate answers
  • No pauses on calls
  • Recognition and synthesis chosen separately
Task 1 of 3

Different AI agents on the platform can run on different models at the same time.

Hosting

Where the model runs

Cloud

The model runs in the provider's cloud – no hardware needed

The model's answers reach the AI agent over the internet, so you don't need to buy a server. Only what's needed for the answer leaves your system: the text of the conversation and the selected information, and personal data can be masked before sending. When data must be processed in a specific country, we choose providers that process it there.

  • No server of your own
  • Providers in the country you need
  • Pay by volume

On-premises

An open model in your server room

We install an open model on a GPU server; sometimes spare capacity on the analytics server is enough. Conversations are processed inside your company, and there's no per-minute charge – the only cost is hardware.

  • Data stays in-house
  • No per-minute bill
  • Requires a GPU

Switching models

The model changes, the AI agent stays

When a better or cheaper model comes out, we connect it in place of the old one and test it on the same reference answers. The knowledge base doesn't change, and if the new model answers differently, we adjust the instructions. This way you cut costs without rewriting the whole agent, and you can start in the cloud and move on-premises later.

  • Checked against reference answers
  • Same knowledge base
  • Start in the cloud, move in-house later
More questions resolved without peopleThe more accurately a model answers your questions, the more inquiries AI resolves and the fewer employees you need.
Fewer abandoned callsA fast model answers without a noticeable pause, so customers don't hang up and go to a competitor.
A lower cost per conversationSimple tasks run on an inexpensive model, complex ones on a powerful model, so you don't overpay for suggested replies and call notes.
Data stays where your rules requirePersonal data is processed in the provider's cloud or on your servers – as the law and your security policy require.

Launch

How we choose a model

  1. 01RequirementsWe find out which data may leave your systems and what AI needs to do.
  2. 02SamplesWe collect customer questions with answers checked by your expert.
  3. 03CandidatesWe shortlist two or three models that fit the language, task, and budget.
  4. 04ComparisonWe compare accuracy, speed, and cost for all candidates on the same list.
  5. 05ConnectionWe launch the chosen model and review conversations with your team for the first few days.
  6. 06ReviewWhen new models or updates come out, we run the comparison again.

Requirements

What the model needs

Cloud model
Internet access, billed by volume
Your own model
A GPU server on-premises
Voice
Speech-to-text and text-to-speech – cloud or yours
Samples
A list of questions with correct answers from an expert
Cost of your own model
Only the server; minutes are free
Selection and comparison
From your support hours

Preparation

Decisions to make before choosing

Can data leave your systems?If not, you need a self-hosted model.
What will AI do?Conversations, suggested replies, call review.
Do you have a GPU server?Sometimes the analytics server is enough.
Do you need voice?Then we add speech models.
Who will check the samples?An expert who knows the correct answers.
Must data stay in one country?A provider in that country or your own server.

Q&A

Frequently asked questions

Which model is best?

There's no universal answer: one is stronger at reasoning, another is faster and cheaper, a third writes better in your customers' language. That's why we compare candidates on your questions and choose by the numbers.

Can we avoid third-party cloud models altogether?

You can run open models on your own hardware. We test their quality the same way we test any other model.

How well do models understand our customers' language?

Modern models are fluent, but they differ in how they handle your terminology and your customers' conversational speech. That's exactly what we test on the samples.

Are free models worse than paid ones?

Not always. For suggested replies, summaries, and call review, free open models often do just as well, while a complex customer conversation may need a paid one.

What exactly does a cloud model receive?

The text of the turns needed for the answer and the selected knowledge base articles, plus the audio if speech recognition runs in the cloud. Phone numbers and ID details are masked, and for full isolation, the model runs on your servers.

What if the cloud model provider goes down?

The AI agent switches to a backup model that's been tested on your question set in advance, and the customer keeps talking without noticing anything.

Is the cloud cheaper than our own server, or the other way around?

It depends on your tasks and data requirements. We calculate both options for your case: the cloud by volume, your own server by hardware cost.

How do we check that the model pays off?

Look at how three metrics change: the share of inquiries AI resolved without an employee, the share of abandoned calls, and the average cost of an AI conversation.

What hardware do we need to run a model ourselves?

You need a GPU. The platform's analytics server often works too, if it isn't fully loaded – we check this during selection.

Can we change the model later?

Any time, with no need to rewrite the agent. We first check the new model against reference answers and adjust the instructions if needed.

Do we need to fine-tune the model for our company?

Usually not: the AI agent does fine with instructions and your knowledge base. We save fine-tuning for a narrow subject area or a distinctive speaking style.

Machine learning →

Who chooses the model?

We make a recommendation based on the comparison and show you the numbers; the decision is yours.

How much does model selection cost?

Comparison and connection are billed in support hours, with no licenses. Cloud models are billed by volume.

Support plans →

Glossary

Terms on this page

Cloud deployment (cloud)
The platform runs on the vendor's servers, and agents connect over the internet. Learn more →
On-premises deployment
The system runs on the company's own servers, not in the vendor's cloud. Learn more →
Knowledge base
All answers, policies, and prices in one place – for agents and AI. Learn more →
Abandonment rate
How many callers hang up before an agent answers. Learn more →
Speech-to-text
Turns spoken conversation into text. Learn more →
Personal data
Anything that points to a specific person: a name, a number, a voice. Learn more →
Text-to-speech
Reads text aloud in a human-like voice. Learn more →
Model fine-tuning
An existing model is trained further on company materials so it speaks the company's language. Learn more →
Large language model (LLM)
A neural network that understands questions asked in everyday words and answers the same way. Learn more →
Hosting
You rent a server from a provider instead of buying one. Learn more →
Backup
Data is regularly copied to a separate place so it can be restored if something goes wrong. Learn more →

More

More on AI in the contact center

Knowledge base Agents, chat, and voice AI give customers the same answer. Learn more Speech analytics Your entire call volume gets scored, not a handful of calls someone had time to listen to. Learn more Data security The AI agent knows exactly as much about a customer as it should. Learn more AI consultant in chat Customers write whenever it suits them and get an answer right away – not the next morning. Learn more

Blog

Fresh from the blog

All articles →
Services · 7 min read Custom development of your contact center platform for your processes When contact center settings are not enough: integrations, modules, AI workflows. How to tell development from configuration and estimate the payback. Read the article → Management · 9 min read How many agents does your contact center need? Erlang C in plain English Calculate contact center staffing with Erlang C: workload, service level, shrinkage, and headcount by shift – all in one worked example. Read the article → Integrations · 7 min read CRM and phone system integration: what it does and how it works How to connect your phone system to your CRM: calls and recordings in the record, screen pop, click-to-call. What to decide first and how to check it works. Read the article →

Updated: