Cristina Fonseca MYC'D Up with Tech Leaders promo image

AI-powered customer experience: Zendesk’s VP of product

In episode four of our newest podcast series, MYC'D UP with Tech Leaders, MyCustomer's Sabine Groven speaks to Cristina Fonseca of Zendesk about how to maximise your CX artificial intelligence offerings, and where she feels so many companies are missing the boat on AI. 

6th Jul 2023

In the latest instalment of MYC'D UP with Tech Leaders, we speak with Cristina Fonseca, VP of Product at Zendesk.

Before arriving at Zendesk, Cristina was already something of an AI expert – having previously founded and run two successful AI-focused companies.

With a strong background and passion for customer service alongside her expertise in the AI field, Cristina is perfectly placed to provide insights and wisdom on the hottest topic in the CX sector.

Drawing upon her considerable experience and success in both the AI and customer service arenas, Cristina provides detailed analysis of precisely how to get the best out of your AI programmes – including what you should and should not be automating, how to build trust in your AI offerings, and how AI could be the key to lowering agent turnover rates.

When we think about automating CX inquiries, the first task of AI for me, is understanding what should and should not be automated.


MYCUSTOMER: So let's dive in with the first question. You're a VP of product at Zendesk. And before we talk more about that, I'd love for you to talk about your background and how you ended up in that role.

CRISTINA FONSECA: It's a great question. So I'm currently a VP of product here at Zendesk, where I’m in charge of AI and machine learning. And my company was acquired by Zendesk, we were an AI company that did automation and AI for customer service. So it was clearly the perfect mix.

But before that, I studied engineering. So I have a technical background. And I built another very successful company called Talk Desk. So that got me into the CX industry. And then I left the daily operations of Talk Desk, and right after that, I was a little bit intrigued by AI, and especially trying to understand what was hype and what was real, and understanding where were the limits of the technology at that moment. So I studied AI. I did a nanodegree on AI in order to understand more. And then basically, I matched AI with the industry I knew better, which was customer service – and I created Cleverly, which was then acquired by CMS. So that has been my journey into CX and what led to me being currently the leader of AI.

MYCUSTOMER: Brilliant. That's an impressive CV indeed. So now that we have an understanding of your journey, let's move on to Zendesk AI. So for our listeners who might not be familiar with it, could you explain how it works and how it's used?

CRISTINA FONSECA: For sure. So Zendesk AI is our AI solution, which is basically about having AI that works off-the-shelf and provides value from day one to customers.

So the way it works is you click a button, you turn it on, it immediately starts labelling the requests that come in from your customer, so you understand what they are about. So for example, if customers are asking for a refund request, or asking for more information about specific billing queries – we label quite granularly, what's coming in. So we do this through intent, sentiment and language. And then when you have that as a base, you can build automation on top using our bots and deflection capabilities. You can programme intelligent workflows, routing requests to the best agents/ teams, you can prioritise certain queries – you are in control in regards to which workflows and automations you want to set up.

And then we have two other very important aspects. One is assisting agents with being more productive by suggesting to them replies, and other types of content that can help them speed up resolutions. And we have a very important foundation layer in terms of knowledge insights, it's not very common that companies help you manage your knowledge base. But AI really depends on data. So in order for you to have good AI algorithms, I need to have articles and replies to recommend to my customers via bots and cell service. I also need to have proper replies in order to recommend to agents so that they can speed up resolutions. And managing this knowledge base with knowledge insights is super important. So that's another layer and another area that's part of our Zendesk API offering. So it's basically like a suite of functionalities that plugs very well with your Zendesk ecosystem and helps you deliver automation capabilities with AI.

MYCUSTOMER: Okay, thank you for taking us through that. So my next question is about how AI has been making strides in customer service. And I'm curious to know a bit more about how it's transforming both the customer and the agent experience. So is that something that you can shed a bit of light on?

CRISTINA FONSECA: So let's start maybe with the customer side. The potential for AI and CX is just very, very obvious. That's why it's one of the industries that always comes up when we talk about oh, like, what's going to be the impact of this wave of AI, in businesses and so on. And the reason being, there's just a lot of repetitive tasks, and a lot of manual work. And one of those things is replying to customers with the same thing over and over. So there's certain types of queries for which the answer is always the same. Most of the time it's done by a human, but it could be done by a machine.

So for example, and I think, honestly, the first task when we think about automating customer service queries, the first task of AI to me is understanding what should be automated and what should not be automated. Because at the end of the day, we all believe in the potential, but at the same time, most of us don't really like bots. And why is that? Because the implementation of bots today is not very well thought. And to change that we need to understand bots are not the solution for everything. If bots understand customers really well, bots should handle the repetitive tasks; the situations where you can just get the reply and send the reply to the customer. But there's a certain type of query that should absolutely be escalated to an agent, because a bot cannot do much. And if I escalate to the agent right away, then I guarantee a perfect customer experience. So in that regard, I believe one, the first task of AI is to understand what should be automated and what should not be automated. And then an AI solution should have the flexibility to allow you to automate certain things, escalate others, and do these gradually. Because that's the best way to build trust in AI, and start implementing it.

Most of our customers are not ready to go full automation on day one, but they are ok with automating the top three to five customer queries, especially if that helps them offload a part of the volume to the machine. And that doesn't damage the customer experience. So in order to guarantee a good customer experience, automate certain things, escalate early and often for the things that really make sense. This can be for example, revenue drivers. If I'm asking an ecommerce company for help on buying a certain product, I think there's a good chance if I escalate that when I mentioned, the experience is going to be best if I have a time sensitive query. And I use AI to understand what that is and prioritise it, of course, that will result in a very good experience. So trying to understand what makes sense for each type of request, what's the best course of action to personalise the experience, and not trying to automate everything.

So on the agent side, I would say, the role of AI primarily should be to try to eliminate all the manual tasks that these people need to do. For instances where you're just writing the same thing over and over, or your copy and pasting replies, or you're searching for the right information to send to the customers, AI can do this for you. Models are quite sophisticated today, they can find information for you, we can use the historical data to understand what's the best way to reply to a particular customer. All of these can be done manually. Now, there's complex queries or more sensitive questions that should absolutely be handled by an agent, there's for sure cases where empathy and just the human touch will make a big difference. And then there's lots of cases where the AI has a suggestion of what to do. But the AI is not super sure, in regards to is not very confident about a particular reply or resolution. And the agent should be the one making that decision. So in the CX world, having an assistant that can help agents be more productive, and automate some of these manual tasks, I think is a winning strategy.

MYCUSTOMER: One aspect that stands out to me in relation to that is the ability for agents to use AI to understand what customers want and how they're feeling before the conversation even begins. So is that something you can talk a bit about?

CRISTINA FONSECA: For sure. And look, that has been the foundational aspect of our solution. And the reason why we invested a lot of time and effort building this triage layer is because when we started Cleverly, most AI and CX projects were consultancy projects. So you'd start all over again, every time you deploy a solution to a customer. And on the surface, everyone can talk about the same things, ‘Oh, I can implement your bot’, or ‘I can make your agents more productive’. But these applications, like bots and agency capabilities, are only going to perform well, if I know very well what the customer is asking. And I would say this is the reason why some bots are quite frustrating, because they don't understand the customer. And if you don't understand the customer, well, then you can control what happens next.

So we figured, look, the only way to solve this problem, and to have AI that works is to understand what the customer is asking right away, and do that with very, very good quality. At the same time, when we do these with a relatively small data set. If we do these customer by customer, it's very hard to get these models to be really accurate. That's why we have spent the last five years creating a database of every single thing that can end up in support. So we recognise when people say, I want to update my email, I want to update my mail address, I want to add a new seat to my account. So we have a huge list of intense reasons for contact that we have off the shelf ready for users, for our customers to use.

And that's like the starting point for you then to automate without you having to go and train an AI to recognise, ‘Oh, what are the refund requests? I guess every single customer in ecommerce wants this?’ Why are they asked to train their own model to do this every time? It's a little bit nonsense, right? So we have the data, we have the expertise, and we have the technical talent, and then have these very powerful insights. And they are able to understand, ‘Oh, what are my customers conducting me about?’ And then based on that they can build a proper automation plan and an automation roadmap. And that's usually the starting point for them to implement AI.

MYCUSTOMER: You also mentioned something I wanted to pick up on because you said, “no one really likes bots”. And I wanted to know if you can say something about consumer behaviours and satisfaction rates towards the use of AI.

CRISTINA FONSECA: We still don't have a lot of data on that every year. We run these CX trend reports, and we survey our customers in regards to lots of things but we still don't have a lot of data in regards to their feedback on being served by AI. But we have some data in regards to their expectations. And their expectations are firstly, customers just want their problems to be solved. And they believe AI can really help companies personalise interactions. And when I read this, I'm not sure if customers are looking for personalisation, or just AI to really understand what they are talking about, and being able to offer the best solution for them. So again, I think maybe next year, we can survey our customers in regards to AI effectiveness. But again, up until now our feedback has been that customers just want their problem solved. They believe AI can help in certain ways. And at the same time, we see our customers that have successfully implemented AI, reducing the number of queries that agents need to answer repetitively – improving customer satisfaction. Because you can just answer customers more strategically, you can prioritise what's important, you can escalate what requires human assistance, and customers are totally fine with getting a reply from a bot, or a reply that's automated, if that helps them because that's what they want, when they contact support, right?

I would even say, and this may be a little bit provocative, that the customer service industry is a little bit broken, right? When a customer contacts support it’s usually because something is wrong. So it's not that people are absolutely excited with these types of interactions, they just want help. And at the same time, agents also are not the profession that excites people the most. I mean, most agents are not very motivated – it’s an industry with a super high turnover. And it's just a job they need to have to pay the bills. So there's a little bit of a disconnect. And of course, AI can help make everyone more productive and make everyone happier in this ecosystem.

MYCUSTOMER: Yeah, it would be really nice to see AI empowering staff as well. And perhaps that would help with employee experience.

CRISTINA FONSECA: Look, there's lots of statistics, but the turnover in the CX industry is around like 50%. So 50% of the customer service employees change jobs every year, because it's super manual, because it's tiring, and because most of the time they don't have the right tools. So if you are able to change that training completely. I mean, right now it's very manual, it's old school and takes a lot of time for a new customer service representative to get trained. Like, the job is not easy, right? So I think if we use AI to train people better, so they have an assistant that can help them be better at their job, while making their work more meaningful, I would say everyone wins.

And we've seen – I mean, we still don't have data, but this is something we want to work on – in the long term that agents being onboarded with AI, they get ramped up way faster. And they have this notion of having someone that can help them. And they can get up to speed faster, they build confidence on the job, which helps both the agent and the customer. So big win win. Of course, there's also some scepticism, especially from agents that have been on the job for longer, that are super experts, and that believe the AI cannot get to the level where they are, which I mean, there will be very specific tasks where the human touch really makes a difference. But overall, agents are going to have these assistants that will make them  way better at their job. And every single one of us in the next couple of years will have assistants, so we get more productive at our jobs and so we focus on the high value tasks.

At some point, these technologies will be so embedded in the way we work that we won't even notice. If you imagine trying to give YouTube to a kid with the recommendations off – it will be a frustrating experience.

MYCUSTOMER: Good point. So there's obviously a lot of potential here, but what would you say are the main challenges for organisations to adopt AI?

CRISTINA FONSECA: To me, the biggest challenge is building trust in AI. I think AI, because of how powerful it is,  at the same time it is a little bit scary. So there's some risks, like the fact that these large language models still make stuff up, cybersecurity risks, and so on. And all companies are aware. So there's a little bit of a scepticism in implementing AI. So I would say building trust. And in order to build trust we should allow these companies to go step-by-step. So it doesn't need to be all or nothing. For example, at Zendesk, we really try to build products in a way that these companies can start small and automate one thing and then go from there. And at the same time on the admin side, make the confidence of predictions transparent. So we know that sometimes I'm suggesting these replies, and I'm 99% sure these should be best, and it's going to help the customer. And based on past interactions this is the best way to solve this particular problem. But some other times, I don't have good data, like I don't know, I'm just trying, and the best thing I can recommend is that I'm like 20% sure. So if we make these predictions, the confidence level of these predictions available to agents and administrators, they can easily build trust in these AI systems. Because AI is not perfect, it's never going to give you predictions that are 100% true at all times. And it’s very easy to point at the ones that are wrong. So this confidence level thing is something that we tried to do. And in every single thing we build, we try to find a way so our customers can go step-by-step, and it doesn't need to be all or nothing.

And maybe, if I can add to that, with open AI and Chat GPT being launched, the expectation of customers basically has been to give me an AI that works off the shelf. And the fact that I can just turn it on, see it working, see the AI getting me insights, that's also a very, very positive first step. So the barrier to start is basically very, very low. And I think that's also another very important driver as these companies are implementing AI. So in order to start, it needs to be very simple. Click a button, go, you start seeing the value, and then you can build on top. So getting new insights first, using AI to get you insights you can use to then plan properly. I think it's also a very important aspect.

MYCUSTOMER: And before we wrap up, then Christina, I have one final question for you. And this is something that we ask all our guests on the MyC’d Up with Tech Leaders podcast. If there was one particular tool solution or product type that you think businesses should invest in, which isn't provided by your organisation, what would it be? And why?

CRISTINA FONSECA: Look, it's a very, very good question. And the first thing that comes to mind, especially in this AI boom, I would ask everyone to think about which assistants can make every single role within your organisation. And that will be different depending on the role, what's the best assistant that will improve productivity for the different roles you have in your organisation? Because as I mentioned, I believe in the next couple of years, all of us will have an assistant or multiple assistants to help us get better at our jobs. So I think that should be the priority for companies. What are the AI systems you should be looking at in order to improve productivity internally for your employees.

Replies (0)

Please login or register to join the discussion.

There are currently no replies, be the first to post a reply.