BANKS AND FINANCIAL INSTITUTIONS

A voice AI employee for your bank — in Uzbek, inside your perimeter

At night and at peak hours there are not enough agents, and customers expect an answer in their own language. The AI employee takes the call, answers from bank data and logs the outcome in the CRM.

Get a demo callContact us
Incoming call23:14UZ RU
  1. Caller00:02Dollar kursi qancha bugunWhat is the dollar rate today?
  2. AI00:05Kursni hozir bank tizimidan aytamanI will give you the rate from the bank system
  3. Caller00:11Ha, а когда дойдёт перевод?Yes, and when will the transfer arrive?
  4. AI00:14I can tell you. Which country are you sending to?
  • Language: Uzbekfrom the first phrase
  • Request: dollar rateunderstood without a menu
  • Rate from bank systemsnot a pre-recorded phrase
  • Switch to Russiananswer in their language
  • Outcome logged in CRMtopic, language, next step
0:20

Inside the bank perimeteraudio and transcripts never leave it

24/7first line, nights and weekends3 languagesUzbek, Russian, English4 departmentsvoice and text on one platform
BANK DEPARTMENTS

One AI agent platform for every department of the bank

Support, collections, sales and PR run on one platform and one knowledge base. AI agents answer by voice and text and write the outcome of every contact to the bank’s CRM.

Calls and chats in one queue, answered in the customer’s language

Channels

  • Inbound calls
  • Telegram
  • Instagram
  • In-app chat
  • Web chat

What the AI agent does

  • Balance, rates, application status, card blocking — around the clock
  • Answers inside the app, where the customer is already signed in
  • Hands complex cases to an agent with a short call summary

Outcome: Request, topic and outcome land in the bank’s ticketing system

In-app chatAI agent onlineuz
  • CustomerKartam bloklandimi?Is my card blocked?
  • AI agentHa, karta vaqtincha bloklangan: PIN-kod uch marta notoʻgʻri kiritilgan. Blokdan chiqarish uchun murojaat ochdik.Yes, the card is temporarily blocked: the PIN was entered wrong three times. We’ve opened an unblock request.
  • Ticket createdTopic: card block · language: uz
Sample scenario
THE UZBEK VOICE

Global models do not deliver the Uzbek voice

That is why a customer who speaks Uzbek still waits for an agent. We train our own speech recognition and synthesis models and run them inside the bank perimeter.

  • Understands Uzbek and mixed Uzbek-Russian speech
  • Answers in a natural voice, male or female
  • Takes rates, terms and statuses from the bank’s systems
0:20

Rate and transfer — mixed speech

UZ / RU
  1. Customer speech«Dollar kursi qancha bugun»
  2. RecognitionUzbek, then mixed speech
  3. Bank knowledge baseTransfer terms — by the bank’s rules
  4. Action in the systemRate — from the bank system
  5. Speech synthesisA spoken reply in the caller’s language

The whole path stays inside the perimeter: audio and transcripts never leave it

Customer speech, recognition, the bank knowledge base, an action in the system, speech synthesis: the whole path runs inside the bank perimeter.

A DAY ON THE LINE

Nights and peaks stop being a hole in the schedule

At night and at peak hours there are more calls than agents on shift. Whatever the shift cannot pick up goes to the AI employee, and headcount stays the same.

Line coverage · 00–23 h
  • left unanswered
  • handled by people
  • handled by the AI employee
An illustration, not a measurement: we will plug in your contact-centre data

Today calls go unanswered at night and during two daytime peaks. With an AI employee the AI picks them up, and shift headcount stays the same.

One night call — two routes

Today

A call at 23:14

Customer
Hears the ringtone and hangs up
Outcome
By morning nobody knows who called
With an AI employee

The same call at 23:14

AI
Answers in the customer’s language
Outcome
The question is resolved or a case is opened in the CRM

Today a night call ends on the ringtone. With an AI employee it ends with an answer or a case in the CRM.

  • The AI employee’s lines absorb the peak, with no overtime
  • At night the same scenarios and knowledge base are at work
  • Complex conversations stay with the bank’s agents
What happens to a call nobody is free to answer
  1. The call joins the queueNight or peak hour, no agent is free
  2. Today: dropped without a traceNo topic, no reason, no task
  3. With AI: an answer in the caller’s languageA free line exists at any hour
  4. Resolved or loggedAn answer from the knowledge base or a case with context

The call joins the queue. Today it drops with no record; with an AI employee the question is resolved in the conversation or reaches the CRM as a case with context.

QUALITY CONTROL

You hear a sample — you will see everything

Today you judge line quality by a couple of recordings someone listened to. The system reviews every conversation and shows where the first line goes wrong.

One hundred conversations of a shift

2 of 100 listened to100 of 100 reviewed

A supervisor manages to hear only a handful of every hundred conversations. The system reviews all hundred, and the supervisor can still spot-check as before.

  • not reviewed
  • listened to by a supervisor
  • reviewed by the system

Today a small share of conversations is reviewed; with AI review, all of them are. Spot-checking recordings is still possible.

  • Enquiry topics

    Cards, loans, deposits, transfers, blocks

  • Reasons for hand-off to an agent

    Where and why a conversation goes to a person

  • Gaps in the knowledge base

    Frequent questions the bank’s rules do not answer

Contact-centre daily wrap-up in Zukko.AI: number of operators, total calls, inbound, outbound and the average conversation quality score for the selected period
Shift wrap-up: operators, line volume, inbound and outbound, average call score

We publish no accuracy figure: we measure it on your calls by an agreed methodology

THE DATA PERIMETER

You are in control, we deliver

Servers sit in the bank’s data centre or in one agreed with the regulator. The bank runs IT and information security.

Bank perimeter
Inbound
  • Telephony and SIP trunk
  • Messengers and the web channel
  • Mobile app
AI core
  • Recognition and synthesis
  • Language model
  • Bank knowledge base
  • Audit log

Audio, transcripts and personal data never leave the bank network

Bank systems
  • Core banking (ABS)
  • CRM and ticketing
  • Agent workstation

Bank channels pass the enquiry to the AI core deployed inside the bank perimeter; the core works with core banking, CRM and the agent workstation; the diagram has no external endpoints.

Public cloud · external servicesNot a single arrow leaves the perimeter

  • Where it is deployed

    Your security team chooses.

    • In the bank data centre
    • A private perimeter in an agreed data centre
    • A hybrid scheme

  • Storage and computation are resident

    Recognition, the answer and synthesis stay inside the perimeter.

  • The bank governs, we deliver

    A public cloud does not fit this kind of delivery.

  • Models trained on your calls are yours

    If the contract ends, we hand them over to the bank and delete our copies.

You walk into the security committee with the answers ready

How platform security is built
CONTROL AND RESILIENCE

Who governs what, and what happens on failure

Before the contract is signed you can see what the bank owns and what Zukko owns. Whatever fails, the customer still has a route to an agent.

  1. Level 0

    Normal

    The AI leads the conversation, an agent joins by the rules Customer: on the line; AI employee: leads the conversation; Bank systems: responding; Agent: on escalation.

  2. Level 1

    The bank system is not responding

    The AI answers reference questions, the rest goes to an agent Customer: on the line; AI employee: reference topics; Bank systems: not responding; Agent: everything else.

  3. Level 2

    The AI is unavailable

    Calls follow the usual contact-centre route Customer: on the line; AI employee: unavailable; Bank systems: responding; Agent: standard route.

  4. Level 3

    A stop decided by the bank

    The customer goes straight to an agent Customer: on the line; AI employee: stopped by the bank; Bank systems: responding; Agent: directly.

At every level the customer has a route to an agent

Where responsibility splits

Where responsibility splits
AreaBankZukkoHow it is verified
Infrastructure and security toolinggovernsdeliversDeployment record
Data access and permissionsgovernshas noneRole model, access log
Stopping the automationgovernstakes no partThe switch is on the bank side
Recordings and retentionownsdoes not exportRetention policy

You get a dashboard, not a console: the Zukko team runs the rollout

HOW IT WORKS

From the call to a record in the bank systems — in one pass

The AI employee carries the enquiry through to a record in the bank system. If a person is needed, the conversation goes to an agent together with its context.

  1. 1

    Enquiry

    A call, Telegram or web chat; language from the first phrase

  2. 2

    Understanding

    Gets the question with no “press 1” menu

  3. 3

    Identification

    Verifies identity by the bank’s rules

    Required on 76 % of calls in a financial contact centrePCI Pal, global data, 2025

  4. 4

    Action

    Balance, card block, status — via core banking or CRM

  5. 5

    Writing the result

    Topic, outcome and task — in the CRM

  6. 6

    Hand-off

    By topic, customer request or signs of irritation

The agent’s shift

Agents only get the conversations that need a person

The AI closes routine requests. A complex conversation arrives with its context, so the customer does not have to repeat anything.

Decisions that matter to the customer are made by a bank employee

Hand-off with context23:16
Topic
International transfer: when the money arrives
Language
Uzbek, then Russian
What the AI did
Gave the rate, asked for the destination country
Why an agent
The customer asked for a person
On the card
Transcript and case history
A sample card, not bank customer data

First-line scenarios

  • Balance and statement
  • Card block and reissue
  • Loan application status
  • Payment reminders
  • Rates and transfers
  • Account opening
  • Disputed transaction review
  • Product terms
  • Branches and opening hours
  • Routing to an agent
  • Outbound campaigns
  • Satisfaction survey

The bank chooses which scenarios run and in what order

MONEY AND ROLLOUT

What a minute costs and when it starts working

This is what a live-line minute costs today. We will name the price of an AI minute after measuring it on your traffic.

The cost of a live-line minute

  • An agent minute with overheads~3,200 UZSZukko calculation, July 2026
  • An outsourced contact-centre minute1,550–2,900 UZSZukko calculation from CIS quotes, 2026
  • An agent minute by payroll alone~1,100 UZSZukko calculation from the GorodRabot.uz median, July 2026
  • A voice AI minute inside your perimetercalculated after measurementdepends on the load profile and the deployment scheme
The calculation excludes infrastructure, integration and training of the bank team.

How the rollout goes

  1. 1

    Discovery and measurement

    The starting scenario is chosen

  2. 2

    Security review

    Your security team checks the scheme and access

  3. 3

    Test perimeter

    AI on part of the traffic, measured on your calls

  4. 4

    Scaling

    New scenarios and departments on the same platform

Discovery and measurement, security review, a test perimeter on part of the traffic, scaling.

NEXT STEP

Test the AI on your own bank products

One call: the AI holds a voice conversation about your bank’s products, and we walk through the deployment scheme.

Get a demo callContact us

A conversation on your own rules · we review the scheme with your security team

QUESTIONS AND ANSWERS

What gets asked before the first meeting

Where is the data physically stored and processed

Inside the bank perimeter: recognition, the answer, synthesis and analytics all run within it. Not only storage but computation is resident.

Who governs the infrastructure and what is needed from bank IT

The bank governs it — these functions are not outsourced. We need SIP access to telephony, exchange methods with core banking and CRM, and resources inside the perimeter; the effort is named at the discovery and measurement stage.

How is banking secrecy observed and what happens to recordings

Access is split by role, every action is logged, personal data is masked. The bank sets the retention period and deletion on schedule is automatic.

How confidently does the AI handle Uzbek speech

We train recognition and synthesis ourselves, including mixed Uzbek-Russian speech. We publish no accuracy figure: we measure it on your calls by an agreed methodology.

Do you have rollouts in the banking sector

No. Today Zukko.AI works for a national airline on text enquiries and for a national medical centre on voice and conversation review. We say this before the tender, not after it.

What SLA do you give and how does escalation work

Availability and response-time commitments are fixed in the contract after measurement on your traffic — we name no figure upfront. Escalation is triggered by topic, by customer request and by signs of irritation, and carries the transcript and the enquiry history.

Which bank departments does Zukko.AI work in

Zukko.AI works in four bank departments on one platform. Support: inbound calls, Telegram, Instagram, web chat and the chat in the mobile app. Collections: outbound calls on overdue payments; the promise to pay is written to the CRM. Sales: calls to customers who have given consent; replies to comments and messages; call-backs on unfinished applications. PR: replies under posts, with complaints handed to an employee. All AI agents run inside the bank perimeter on one knowledge base and write the outcome of every contact to the CRM or ticketing system.

BEGIN

Let us start with one scenario on your own traffic

We pick one voice scenario, deploy it inside the bank perimeter and measure the result on your calls.

Get a demo callContact us

Deployed inside the bank perimeter · your security team checks the scheme