2020 deepVB® Project

Conversational AI with Real-world Results

scroll down

Alfa Bank Company Profile

Alfa Bank Ukraine is part of Alfa Group Consortium:

One of the largest privately owned investment groups; commercial and investment banking, asset management, insurance, retail trade, and water utilities.


The group includes 5 banks: Alfa-Bank Ukraine, Alfa-Bank Russia, banks in the Netherlands, in Belarus and in Kazakhstan with representative offices in Cyprus, the United States and the  United Kingdom.

Voice Biometrics Project Goals

Measure Omilia VB technology in Ukraine

with mixed language speakers

Faster Authentication

Improve Customer Experience


Reduce Agent Costs

Minimize Fraud

Reduce fraud by adding biometric factor of authentication

Voice Biometrics initiative targets

Minimise long authentication time:

Identification takes up to 70 seconds for certain operations, so customer experience and agent costs skyrocket.

Improve Authentication success rate:

Often, clients don’t readily remember the answers to Auth questions, and end up disqualified and frustrated

Minimize Service Denial due to ID/AUTH failure:

21% of customers do not manage to successfully authenticate, and are forbidden to retry for next 24 hours.

Timeline: Project completion in less than 5 months!

Service Flow: How It Works

Passive Voice Biometrics through the IVR

Stage 1

Enrollment – Create Voiceprint

1. User First call to the system becomes the users VB Enrolment call

2. On Successful Identification Agent will press button on agent desktop, and Voice Data is passed to the VB system for enrolment.

3. VB system creates customer Voiceprint and enrols the user

Stage 2

Normal Operation: VB authentication

1. Enrolled User calls IVR and starts talking to agent

2. System compares VB and displays color coded

real time results to the agent desktop

Confidence Levels


Positive ID

0,1 – 0,98

Permissible ID


Negative ID

Technology Challenges

1. Train the VB system without properly annotated speech data

a. No speaker labels, some recordings from same speaker

b. Sanity checks needed, data augmentation with different noises

2. Language and Dialect independence

a. A single VB model for Ukrainian, Russian and Mixed.

b. No need for Language Identification modules

3. Keep at the same time

a. LOW False Acceptance rate (FAR) and

b. HIGH % of users successfully authenticated by VB in the first dialogue turn.

4. Effective Agent – Console Integration

5. Optimization of workflows for security and legal compliance

a. Non-enrolled users should be asked if they want to use VB

b. If yes, they should be sent to an agent to enrol them

c. Security questions can be skipped in case of VB verification

Beyond State-of-the-Art Authentication Results

1. Fast Enrolment with minimum 25 sec speech sample!

2. Success in creating Voiceprint – 95% – on the first try

3. Successfully authenticates more than 85% of users from the first dialog turn! 

4. State-of-the-art authentication rates with only between 2-4 seconds of speech – enough time for correct matching with sample

Confidence Levels:


Confidential SR – True Client

0,1 – 0,98

Permissible SR – With extraneous noise, poor communication, a “sick” voice, or outdated voiceprint


Negative SR – Definitely Not the client

Beyond State-of-the-Art Security and Customer Experience Results

• Greatly improved customer experience

– No denial of service due to forgetfulness

– Decrease complaints

– Increase Client Satisfaction on 13%

• World Class LOW FAR (False Acceptance Rate) of 0.2%!!!

• 90% Reduction of Authentication Time 

• 100% Reduction in Fraud rates in the contact center

• 20% agent savings on service line

• Cost optimization $193K

Lessons Learned

• Voice biometrics technology is successful in terms of combating fraud. 

• Voice biometrics are the most effective additional authentication factor, both in terms of cost reduction and customer experience.

• Pivotal point is that VB technology must enrol users with

– not more than 25 sec of speech and 

– successfully authenticate with 2-4 seconds of speech

• Because in any other case both enrolment and authentication can become very difficult to achieve in a single call.

• Customers love the IVR experience the shorter it is, and Voice Biometrics is the key to shortening authentication by up to 90%


Our proven Omni-Channel technology is aimed at:

Large Corporations (200+ agents / 4+ million calls per year), Integrators & Contact Center Service Providers.

If you represent a relevant business and would like to arrange a demonstration of our technology and learn how it can transform your customer care, fill out our form and we will get in contact with you to get the ball rolling.

    First Name
    Last Name
    Business Email
    Job Title
    Number of Employees