Vocodia Executive Interview
Brian Podolak, CEO and Co-Founder, Vocodia
Click image to download the interview ebook.
Vocodia is an Artificial Intelligence Company specializing
in sales and customer service for call centers.
CrmXchange Managing Partner Sheri Greenhaus conducted an
in-depth conversation with Brian Podolak, CEO & Co-Founder, Vocodia
Holdings Corp after the CCW conference in Las Vegas.
Please give us an overview of your company.
In a nutshell, our service is a call center with virtual or
AI agents. We are one of the only turn-key solutions out there. We basically
build digital contact center agents using trained AI for a client, providing
technical resources that make it as simple and easy as possible.
How do they know that the customer is happy?
The number one thing they look at is the AI DISA (Digital
Intelligence Service Agents). We look at how effective DISA is at handling the
conversations and the goal is at least 80% of the time. What a lot of people
don’t realize is that AI still has to be manually fed data to work correctly.
So we take questions and answers from conversations and feed it into DISA to
train it faster. If the other 20% of conversations has to be transferred to a
supervisor, generally they are questions that haven’t been asked before, or
maybe it was asked in a certain way that the AI doesn’t understand. When that
happens, we flag it, learn from it, and implement new information as a
Are you doing the QA on these calls? Who is listening?
Not only are the calls listened to via the recordings but
there is also a transcription. We have our own internal QA process. Clients can
work with them in the beginning if they’d like and we work in conjunction in
that team. They soon realize how quickly the DISA is trained and it’s not
Is it in the cloud?
Everything is cloud based.
How often does the client need to listen and can they
interrupt if something is wrong?
Traditionally in real time we have a whisper or a barge
function. As of right now, we listen to calls afterwards. We are not listening
to calls in real time. If we have about 500 recordings or so beforehand to
train the system, by the second week the system is ready to go.
Your website says the DISAs never go off script? If a
customer is aggravated, is that a good thing that they can’t go off script?
It is correct that, they never go off script. With the
technology today, we believe it’s a good thing that they stay on script for
compliance. We recommend at any given time that the customer can say “operator”
to transfer to a supervisor if needed. We are usually right around the KPIs of
a human agent in terms of calls that get escalated to a supervisor.
Does the person does not know they are talking to a machine?
Customers almost never realize that it is an AI. You notice
on inbound, its ok to be robotic looking. We have actually seen in certain
verticals it is a better customer experience when the person is not speaking to
a human. For example, with debt collectors, that collection agency calls - if it’s
a person it’s a little embarrassing. If it’s a machine, that person might talk
more freely. So right now, we are working with a debt collection agency and we
are seeing that most people prefer to talk with the digital agent.
I can see that with debt collections. What about financial
or health - issues that are more sensitive?
In healthcare, because there needs to be a licensed agent to
talk about specific topics, the digital agent does not address those issues. However,
if it is insurance doing wellness checks, that is where people prefer the digital
Talk about the scalability of this technology.
COVID taught us we need to be flexible. Let’s say it’s
Mother’s Day and you are a flower shop. Typically you need 50 agents, but since
it’s Mother’s Day you require 500. In that case, you can just switch it up easily
on the back end. From the time you make the change it takes 8-15 minutes to
ramp up and have 500 agents on the phone. Ramping back down to 50 takes 8-15
minutes as well. We can be seasonal and flexible.
Is this more business to consumer as opposed to business to
We do both! With B2B it's a lot of appointment reminders and
appointment setting, things like that, and with B2C it's more
Walk me through - a company says yes, were going to try this
- what are the steps?
There’s a lot of questions we ask at the beginning
(1) Logistics: hours of operation, number of agents,
(2) What is the script that you are following today? What
are the objections and rebuttals your agents are dealing with?
(3) Can we get 500 or so recordings of your agents? We want
to test recordings. With AI, you don’t want to test with just one voice. You
want to test male voices, female voices, individual voices. You can actually
A/B test with real data - which question did we lose a call? Where do people
get frustrated? We can actually see down to the second where this happens and
(4) Testing and adjusting
Usually it takes about 30-45 days until we are up and
Do you use a dashboard that shows the number of calls and
We have a client dashboard that is widget based. It can be
customized the way you want to see data versus the way I want to see data. If
you’re a line chart person, you can see it that way. You can pick and choose
elements like number of calls, etc. What makes us unique is a dash that talks
about how efficient the DISA is - how good it is at understanding reply, how
often it gets stuck. This could mean someone has a heavy accent, a question we
have never heard before, any anomaly. We flag profanity and inappropriate
behavior. To really get it perfect could take 90 days to six months.
Did you develop this technology?
My partner James and I go back 20 years. I had the call
center and sales experience and he has the technology experience so we
collaborated on this.
Do clients give you KPIs that they want to meet?
100%. Of course. Contact rate, transfer rate. Sometimes they
can be different and if we don’t have it on our dashboard already, we will add
You are in a crowded field. Did you find people saying ‘I
already know about virtual agents’?
Not at all! We had some questions about compliance. Our
biggest piece of feedback was “we tried virtual agents and it didn’t work”
because the solutions they used were NOT turnkey like we are.
When was your first installation?
We have been developing this for two and a half years. We
started in April of this year (2021) with beta clients that we were able to
test and tweak with. We went into revenue 60 to 90 days ago. We have a ton of
clients ready to turn on in January.
What other industries are you working with?
We have had a couple BPOs come to us to offer to augment
what we are doing. Health insurance, home security, doctors’ appointments. A
lot of this is new and you have to learn it, which is a challenge. We talked
about collections agencies, which is a big one for us. Fintech is huge and we
are talking to a few hedge funds. Banking inbound customer service - lost
credits cards, etc. - that is where we can come in too. We are dealing with a
large casino dealing with reaching out to players to get them to come back. We
are talking to a client that doesn’t want to use it for their sales process,
but to reach out to clients who haven’t been back in a while for a win-back
campaign. There are so many pieces that require people on the phone, so we are
helping out in all of these areas. There’s a lot of interesting models we are
testing on. There are a few we are not ready for, like a consultative-type
sale, and like medicine in terms of making a diagnosis, but there is so much we
CAN do on the customer service and sales side.
Does the name Vocodia have mean?
We wanted to CODE something for VOICE so VO for voice. COD
for code. And “IA” at the end. It’s also a musical instrument. You might
recognize what a vocode is from 70s music.
What was the overwhelming feedback from your trade show?
The number one thing was compliance over and over again. The
second was availability.
In the last minute or so that we have, what else would you
want our audience to know?
The only thing we haven’t covered – If there is any AI you
want to implement, if you don’t have enough data to train the models, it’s
going to fail. We took a couple clients that for some reason didn’t have access
to recordings - we tried to get around that - and it didn’t work. We didn’t
take any money and we were just testing. AI is going to have to be trained by
your smartest people.