These jobs will migrate to other
clients issues to help our other needs from clients, not only
customer support, but maybe people today that they are in customer
support, they will be on sales. They will be on how we can
achieve customer success. So for example, how can help our
clients to use our product better. And not only handling issues. Hello and welcome to the
conversations with Zendesk podcast. I'm your host, Nicole Fonders. Today, we're talking with Guilherme
Kohlberg, head of CX and customer care at XP Incorporated, a leading
financial services company in Brazil that has the mission to help Brazilian
people to invest their savings better. We wanted to speak with XP because
they're an early AI success story. And Guilherme shared with us how they
first approached implementation, what they've discovered along the way,
the impacts they've seen already, including things like reduced handle
time and greater agent satisfaction. If you're interested in taking
the next steps with your own team, we actually recently released
a new guide for CX leaders covering the best Best Practices
and Tips for AI Implementation. It's free and it's
available at zendesk. com slash intelligent dash cx. And if you're looking to dive deeper
into that topic, I encourage you to sign up for our upcoming webinar,
Unlocking the Power of AI for CX, which is coming up on June 4th. We'll be teaming up with experts from
AWS and IDC to discuss everything you need to know about AI for CX, from
trends to challenges and much more. The link to register for that
is available in our show notes. Thanks. Also, please be sure to follow Zendesk
on LinkedIn to hear about this event and others that are coming up soon. And as always, we
love hearing from you. So if you have a question or
some feedback on the podcast, you can drop us a line anytime
at CWZpodcast at Zendesk. com. All right, on to the show. My conversation with Guillerme was
recorded in person at Zendesk's RELATE conference in Las Vegas in April. Before we get to that, we're
going to hear from some Zendesk customers that we spoke to on
the expo floor about how they're thinking about and using AI. Brandon, can you tell us a
little bit about how you're currently using AI, either for
work or in your personal life? To us, AI is just another tool
in the toolkit to say, this is how we can help customers faster. This is how we can help agents become
more efficient at what they do. At Barnes Noble, we're just kind
of getting our feet wet and doing a little testing to see where we
can put it into some workflows. I think it would really be helpful
for our team and especially customers. Our knowledge base is kind of complex. So. Having some AI chatbots will
be really helpful for us. It's hard getting knowledge
built, um, and created. So being able to make that easier for
our agents to be able to share the knowledge that they see every day. That's what we're kind
of really looking at. It's really useful in
like writing content. It helps me like change the tone. How do I make something shorter? So that's actually really useful. We're partnered with Ultima already. We love it. It was plug and play, out of the
box integration, really reduced a lot of traffic for our agents. It's driven all the automation, right? And it's just been a
really dreamy partnership. I absolutely love ChatGPT to tell
me what the best car is going to be for me for my next car. Ooh, I love that. What a great use case. Saving you time. I actually just took a three week
trip to South Korea and Japan. Oh, wow. I went eight years ago and it was
a blast, but a headache to plan. We used ChatGPT. It took us 30 minutes to plan
a trip for an entire city. We followed that list. And it was, it was excellent. A huge thank you to everyone who
spoke with us for that segment. It sounds like there are a
lot of different things that people are using AI for. So let's take that deeper now
and hear all about how XP has had some great successes with using AI
to power their customer service. Here's my interview
with Guilherme Colbert. Guilherme, welcome to
Conversations with Zendesk. How are you doing today? I'm fine, and you? I'm doing quite well. Now, of course, we're
recording this episode live from RELATE24 in Las Vegas. How has the conference
been for you so far? It's amazing. It's amazing. We came here last year for RELATE,
and this year we saw how all the technology evolved, and we are much
more mature in our Zendesk program, so it's been really insightful for us. Awesome. So, tell us a little bit
about your organization. What is XPI do and what does
your support structure look like? Yeah. So, XP is a disrupted
company from Brazil. We are called the Brazilian
Charles Schwab because we are a financial services company. We were born in 2001 and we
started helping individuals in how to invest their money better. And at the beginning, for example,
our founder, he believed that XP would become a university
because we did education programs and then people became clients
and then they started to invest. Nowadays we have a network of more
than 15, 000 financial advisors. They help their clients and
videos to help their money. And in your CX department, we
organize in a matricial way, in the way that we have all the
technology team, products teams. And all the capabilities that
we need in order to help our clients to be better served. Got it. A lot of organizations that we're
talking to here are excited about AI, just like you were talking about, are
just starting to get going with it. But I understand that you've actually
already started to implement some AI into your support use case. Tell us a little bit about how
you got into it and what those initial conversations were like. Like every company, we
wanted to do everything. So we start implementing
and start with the big bads. The thing that we learned last
year that the basics that you have to crawl before you walk. So we started trying to implement
this bot that will connect directly to end users and then we will start
an automation that is really complex and then Zendesk relate 2023. It was, uh, game changer first
because I really liked the approach that Zendesk took
and we took that internally. 'cause okay, you have
all this technology. We all know this is a trend. But let's start the way
that the AI can help agents. So we start using native
features from Zendesk. And then we start going
deeper into client's journey. So right now we use for agents
productivity, but we also use directly with end users and also to start
some more robust alternations behind. Got it. So what were some of the biggest
challenges as you got started and what would you advise other companies
as they're getting started with it? I would say the biggest challenge
that we still face in this challenge is how You, you approach AI inside
a big company because AI is the gold rush and every department
wants to implement it and every people in the organization
wants to try to experiment. Will you have a more
centralized approach or a more decentralized approach? I don't have the answer yet. And I hope I can have next year,
but we start more decentralized at the middle of last year, we
moved to a more hybrid approach where we have this central team
that is directly connected with the CTO and the architecture
team that they are enabling. Okay. How we will consume
different AI features. And then we have these
different business units there are evolving on it. So we do have two ways that we do. One is using Zendesk. So using our partners,
their native features. So that's much more easier to
implement because it's just plug and play and then you
connect, it was really easy. And the second was, okay, how we can
build internally our capability or internal knowledge base, not only
for customer support, but also for sales, also for other domains and
how we can assure that we provide. Safety for our clients
with their data. What have been your biggest
wins with that AI so far? What kinds of results have you seen? The first move that we made, it
was, okay, let's turn it on Zendesk AI and let's call a group of 10
agents and let them then use it and see how check the results. And then we saw the group
where monitor CSAT, we monitor average headline time. And then in one week we saw like
this, okay, but it's not only the 10 agents that are using, but. Almost a hundred percent of the
agents they are using is in the SKI. What is happening? So the agents, they discover
all by themselves the features and they start to use it. And when we saw they were
sharing experience inside our Microsoft Teams group. Okay. You can do that. You can do that. And then it was nice for us because,
okay, that's the, really the goal that agents first, they have to
see AI as a tool to help them. And regarding the results, I mean,
we achieved really good results. So people that were using Zendesk
AI, especially the features like change tone, expand, we saw an
increase in success around five points, and we saw a decrease in
average handling time and around 5%. So you've really seen some substantial
business impacts it sounds like, and obviously more to come. Yes. Yes. A lot. A lot. That's good to hear. And then we start using other
features such as summarize. Because one problem that we
have as a big company, I believe every company has its problems. So you have different teams handling
different types of requests. It's normal that clients,
sometimes they get transferred. It's not what we want. We are working hard on
smart routing right now. But when this, this happened, like
the streamerized features, it helps a lot because the agent doesn't
have to read all the conversation. And one thing that is nice to
mention that 70 percent of our clients, they contact us by chat. Okay. So it makes sense that there would
be a lot of AI engagement around that, given that you're using that. Yeah. It's easier for us to implement it. Normally clients prefer voice, and
voice, I believe it's much harder to do this first to implement
it than it is for messaging. Very good. You mentioned that your agents
had gone through, started finding some of the features, you saw
them adopting things maybe that you hadn't even trained them on. What has their reaction been? It sounds like they're excited. Do they overall really like it? Has it been a mixed bag? What would you say? They like it. They like it a lot. We didn't have Problems like,
okay, but the AI would take my job because, uh, in XP we have this
really entrepreneurial sentiment. So people want, they, they
want to be in customer service, but they want to be agents for
like three, four, five years. And then they, they will
move on in their careers. So they know that that's
something transitory. What we want is that they can
handle much more complex issues. Okay, I don't want a human being
creating things that a robot can do. So it sounds like it's really
evolving their role into something that's going to be a little bit more
fulfilling, something that's going to help them build their career. Are there other ways that you
see your agent's roles evolving with the implementation of AI? One of our main goals inside
customer service is that, okay, how can we help XP to not have a
customer service department anymore? Because our product is so digital,
it's so frictionless and they, they really engage in this perception. So for example, when we do like this
continuous improvement programs. All the agents, they
like to participate. So they come with ideas, they
come with, okay, we have to fix this bug inside our app, or we
can write this, this information better in our knowledge base. So clients can help themselves. And with AI, there are some agents
that they are really into the AI. So they tell us. Ideas, oh, but I know, and
I can start this automation. And some of them,
they know how to code. So they start to build their
own little processes so they can automate their workflows. So we really like the sentiment. Okay. We have goals, so we have to help her. We have to improve our
customer experience. We have to reduce costs since we have
to be much more scalable than we are. And they like this idea
and they help us a lot. That's great. So we talked a little bit about
how the agents have responded. How have your customers responded? Or do they even know the difference? Before Gen AI, most of the
people hated bots, right? Especially in Brazil. I want to talk with a
human, not with a robot. And what we did is we invest
a lot of time creating these 100 percent automated flows
for our bot experience. So, for example, if you want to
get your tax documents from XP in order to file your taxes, you
can do it, everything on WhatsApp or web chat, whatever you want. We expend a lot of energy trying to
build this a hundred percent automated flows for clients and they love it. We saw an increase
in deflection rates. We saw an increase in satisfaction
and also a decrease in costs. That's fantastic to hear. That's fantastic. That's the dream, right? The goal. The goal. But the challenge here is how
you work with other departments. So we are not like this incumbent,
this really old company. But we are not a company that was
born like three, five years ago. So we do have some legacy systems. We do have multiple
process, multiple flows. We have to have this
continuous improvement program with the entire company. I mean, it's always going to
be an ongoing process, right? Cause of course, once you get
one thing optimized, some new technology will come up or there'll
be something else that you want. We like to say it's an infinite job. Yes. We have job for, because when
we solve today's problems, the clients will want more. The client, he or she doesn't compare. XP with other financial companies,
it compares with Netflix, it compares with Google, it compares with Amazon. So the expectation are increasing. You talked about how you've been
designing these flows and that your customers have responded positively. How are you making sure to balance
human side of engagement with the new technology and the AI that's coming? That's hard. That's hard. I would say that it's
right in your pocket. The best example is how you create
chatbots, because you can create a chatbot that is deflect a hundred
percent of your tickets, but then some clients will go crazy. You don't want that. You don't want that. Or you can create a chatbot that
the first thing that went wrong, you already goes to a human
and then you increase costs. So what's the balance? So two things were really clear for
us, when the clients start to swear. It's not a good sign, but it's
a good sign that you need to do something a little different. Yeah. So we train our bot to identify
if the client is angry. Right now we are testing the
sentimental analysis to, to do that. But when the client is swearing,
he or she goes directly to an agent and an escalated channel. And secondly, the mix between how
you make a more conversational bot, ask ChatGPT, or a decision tree. Okay. Now we have mixed models where
the client writes what he wants. And then when we know, okay,
he's talking about an issue with his credit card, and then we
direct with a list of options. And then we put what we have as
a hundred percent automated flow. And he always have at the end an
option to speak with an agent. Right now we are, we reaching our
about 65 percent of deflection rate. And we start last year with
50, so that's a huge advance. And we still, we saw NPS increases
and our CSAT score increases. So that's why, how we balance. Great. It's a long journey. Even if AI is disrupting everything,
it's not from one day to another. You've started on the AI journey. What do you think is
coming up next for you? So I believe for the next two,
three years, many companies, they are facing this challenge as how
we can connect all the companies, how we can connect the systems,
how we can build a knowledge base. So if everything is integrated,
if with one vendor or multiple vendors, but if it's integrated,
it's much more easier. Then you can start to
plug AI inside that. So say this next few years will
be all about how we can integrate and connect our infrastructure. Or a medium or big company,
I would say it's the same. And secondly is how you can gain a lot
of traction by decentralizing again. So you have the capabilities and
then you decentralize that by different teams that they can
work on their own use cases. For example, at XP, you're talking
about customer service, but we do have a sales team that using AI in order
to sell credit cards, for example. So they are engaging prospects. So we just talked a little bit
about your near term vision for AI, what's coming in the
next two or three years and how you'd like to see it integrated. What's your long term vision or your
big dreams for AI in the long term? I don't like to talk about long
term view, but if I could bet, I would say that many of the
work today will be done by AI. Like all the tasks that can,
that are silly, that can be automated, that will be done by AI. We are using a lot AI
with agent supervisor. One day AI will train. So you don't need this
agent supervisor anymore. These jobs will migrate to
other clients issues to help our other needs from clients. Not only customer support, but maybe
people today that they are in customer support, they will be on sales. They will be on how we can
achieve customer success. So for example, how can help our
clients to use our product better and not only handling issues. So I do believe AI is going to disrupt
a lot in, especially in your work environment, but also other fields. For example, in Brazil, we have
a really poor population and this poor population receives
ads from the government, but they do have a cell phone. They do have access for the internet. So how can we make all this,
this process much better for the population and much more easier
for government by using AI? So I believe that's a huge
amount of things that we can achieve as a society. It'll be really exciting to see
what direction all of that goes in. What has been the biggest
surprise to you or the biggest result that was unexpected
out of your AI implementation? We discovered that
there was this agent. That he was using AI in order to write
his boss an email to get a raise. So he was going to Zendesk, write,
okay, how can I get a raise? And then he was making this really
nice and text telling why he should get a raise at this time. Well, he's showing some initiative and
a lot of cleverness with doing that. The tools are there, so
you just have to use it. And I think people that are
not using they are missing it. To build a building. You have all the tools, but we
still, like, only use your hands. Yeah. For example, we do
make a lot of analysis. So analysis and
PowerPoint presentations. And right now we see many employees
starting using AI and different kinds of AI in order to build presentations,
in order to make analysis. So for example, five years
ago, I was reading client's reviews on a Excel spreadsheet. Right now, I don't need
to do that anymore. So I just have a prompt on GPT
that does make that for me. That's great. It's a good time savings. What is the biggest thing that you
have learned that you would recommend to other businesses that are just
starting to think about using API? You have to learn to crawl
before you learn to walk. Start small, start using like
different, like easy features. Start by using AI with human help. You can get more confidence, you
can test, and then you can learn. If you start only with the big
bads, that will be much more difficult to achieve any results. So that was our journey. And right now we are seeing
this paying off a lot. I've learned so much from you
today, and I'm really excited to hear more businesses starting to
adopt AI the way that you have and see some of those same benefits. I have one last question for you,
and it's our favorite question to ask on the Conversations with
Zendesk podcast, which is, do you have an example of a time
that you were the customer and received some really great support
or had a really great customer experience with another company? I had one like recently. My second daughter, she
was born a month ago. Oh, congratulations. So my older one, she's three years
old now, and she's kind of jealous. Yeah. So I spend a lot of time with her. And then I started going with her
to a park called Parque da Mônica. That is a, it's a comic
in Brazil, really famous. And then I said, okay, if I'm going to
stay a lot in Sao Paulo, maybe I, let me check if they have this year pass. And then I checked online, they
have the year pass, I bought it. And then the day when I arrived
at the park with the year pass, Guilherme and Sophie, Sophie
is the name of my daughter. Is that you? Yeah, it's me. We want to thank you because
we didn't have a network pass. We're just testing something
that if, if it's going to be appeal for our clients. And you are the first one to buy it. You and Sofia. So they gave us a lot of presents
and comics and really nice things. Oh my gosh. That's an amazing story. How wonderful. Yeah. Thank you so much for taking
the time with me today. I hope that you enjoy
the rest of Relate. Yeah, I will. Wonderful. Thank you so much. Thank you. If you missed Relate, you can
still catch the recordings. Just visit our show
notes for the link. And if you enjoyed today's
episode, please leave us a review on iTunes or Spotify. Be sure to subscribe or get
notifications for future episodes and join us next time
when I'll be speaking with Brent Pliskow, Vice President
of Customer Support at Upwork. Until next time, I'm Nicole Saunders
for Zendesk, the intelligent heart of customer service.
