Latest / Talk Commerce / AI Agents Are Transforming E-commerce Customer Support and Operations with Luca Borreani
Transcript
- Brent Peterson (00:02.22) Welcome to this episode of Talk Commerce. Today I have Luca Borreani Borreani Borreani. Yeah, Luca, go ahead. He's the co-founder of... Yeah, yeah, sorry. He's the co-founder of ZipChat. But Luca, go ahead, do an introduction for yourself. Tell us your day-to-day role and one of your passions. Luca Borreani (00:07.276) Hello everybody. Borreani Borreani. Luca Borreani (00:21.334) Absolutely. So I'm Luca Borreani, I'm the co-founder of ZipChat AI. I'm also serving as the CMO. So taking care of everything marketing. Thanks to AI, actually today is a much easier job than when I used to have the same role in my previous company, where I just added to hire more people every time. I love doing sports. I love traveling. I'm Italian. Maybe you can find it out by my accent. But I mean... I've been living all over the world and right now I'm busy in Dubai. Brent Peterson (00:53.196) Yeah, that's great. And what brought you to Dubai? Just out of curiosity. Luca Borreani (00:57.836) Yeah, during university, actually after my second master, with my business partner at the time, we were running affiliate marketing campaigns. know, at that point in life, you're young, you're making some money, you work from your laptop. I mean, it sounds like the dream lifestyle everybody was kind of looking for. We could literally live everywhere. Our main goal was actually to move to New York, but it was very complex not being ourselves Americans. So yeah, we tested a few places at the end of the day. Dubai was kind of close to Italy still, just a five, six hours flight. Very easy bureaucracy, good connection between Europe and Asia. So it just made sense at the time and I didn't move back. Brent Peterson (01:41.42) Wow, exciting. Good. All right. Well, so before we get into content, we're going to talk about AI agents and your chat. I'm going to tell you a joke. I did this thing called the Free Joke Project. I'm going to tell you a joke. You just give me a rating 8 through 13. So here we go. At the airport, my friend suggested we disguise ourselves as luggage. I said, let's not get carried away. Luca Borreani (02:06.35) Nice. 11? Brent Peterson (02:09.898) All right, perfect, great. So Luca, us, give us the 10,000 foot view of your past and how you got into what you're doing now. Luca Borreani (02:21.152) Absolutely. So I've always been very ambitious. I've started reading like, you know, self-help books, leadership books, like what at the time were the gurus of the time books about, you know, how to be successful in life when I was like early in high school. So at some point it just made sense for me to start doing something online was the easiest way while studying at university to make a living. I mean, some extra money, not a living at the time. So I started affiliate marketing with my best friend and partner at the time. Things moved quite fast, to be honest. After a few weeks, we were already super profitable. mean, it was another game, it was another time. Not many competitors, would say, especially in the Italian market, so things were quite simple. And we started from that. So at some point we said, okay, but if you are good at promoting other people's stuff, why don't we promote our own stuff? So we started e-commerce. we started running some dropshipping campaigns at the time. Super easy, dropshipping from AliExpress, nightmare with payment processors, lots of complaints because shipping were never coming in the right time, stuff like that. So our idea was, how can we solve this problem for all the growing dropshipping industry? And we started a company called UDROPPY. UDROPPY, we started, we got some funding from a family office in Hong Kong. Then we went to Silicon Valley. We got humiliated by what was at the time our setup. So no VC's were interested in investing in us. So we kind of flipped the game. We added subscription. We flipped the company to US. We raised some funds from Sequoia Capital, from Jason Calacanis, and a couple of other interesting investors. Then after a few years, we were kind of tired of the business. We sold our shares. And in a period of, OK, what's going to be next? One of our previous customers, a very big one, he also sold his e-commerce business and he was starting playing with AI and solving what was his own problem during his e-commerce period and he needed a co-founder, so we joined him. And yeah, the rest is history. Brent Peterson (04:37.378) That's awesome. now you're doing AI chat. So tell us a little bit of, give us the background on agents and what that means for a user. Luca Borreani (04:45.077) Yeah, absolutely. just to give you, let's say, a background, we started as a simple generative AI chatbot. So people were giving an input and they were getting an output, basically a reply, product recommendation, or some information about the policies of the stores. This was basically the phase one of pretty much every AI application out there. Then at some point, AI is much more than that. It's not just a reply. It's potentially an autonomous entity that can take action on your behalf or on behalf of the merchant. In our case, we are not doing the same thing Shopify is doing right now with OpenAI, allowing customers to get product inside charge GPT recommendation and then going on behalf of the person buying these products for them, for example. But it's more on the merchant side. So allowing merchants to automate that kind of task, that kind of repetitive things that usually need the human. can be like customer support or customer success or whatever. So this kind of task can be like anything from following up on questions, recovering cards, edit orders. issue refunds or product recommendation and then upsell, bundling. Basically, every kind of action you could actually do as a salesman or a super agent if you go in a normal store, but online. Brent Peterson (06:16.95) Okay, that's great. is there some safe, I'm assuming there's some safeguards because I'm a big cloud code user and I realize that AI isn't always as smart as it says it is. Give us some workflows there, but how about guardrails around those workflows? Luca Borreani (06:35.404) Yeah, so of course, it's the merchant that decides what are the guardrails. We have some basic ones, but in general, what I think is an approach that is not so common, especially right now when it comes to customer support automation or marketing campaigns, is that everything in our platform is promptable. So you literally use normal language to describe what you want it to happen. and the eye will interpret it and act on it. So you tell it exactly where you want it to act, when, what kind of action, should it sound like a WhatsApp follow-up message, should it send one every 10 minutes, should recommend this kind of product or this, or even negotiate with the user, or tomorrow when users will be using agents to buy. should it negotiate with the user's agent? And how much is allowed to negotiate? Of course, you don't want to provide a 99 % discount. Would it make sense? So you basically set, in this case, how much is the coupon code that you can provide on a one-on-one basis. And then, of course, the AI will try to not get there, but in case that's going to be the limit. And I mean, this is like 360 on our platform. Brent Peterson (07:52.962) Yeah, that's interesting. I think one thing that I've discovered, and there's a lot of big companies now using autonomous agents, some of them are making the mistake of not telling the user that they are an AI agent. And then sometimes, just like you used to get caught into a call loop when you're calling in to somewhere, you could caught in this AI loop where you recognize it's an AI agent. They won't let you out. They just keep responding with some questions and How do you get around that and helping customers who get frustrated with AI? Luca Borreani (08:25.678) So I would say that the funny thing is that what I noticed is that when we started two years ago, everybody was trying to trick the final user this way. So they were trying to never say it's an AI, because they felt like that people would not like to be talking to AI. Of course, the AI experience is different from these three flow chatbots where you press buttons and then you proceed in the flow. So they were trying to trick. Right now, I would say the trend we see at least across our merchants is to declare it's an AI, like a small disclaimer or just to say, I'm the AI of whatever the brand name is. In this case, so they make it clear it's an AI. But for example, in our case, I'm not sure about how everybody else is handling this. What happens is that you basically escalate to humans what the AI is not able to handle. And this can be like either the human, the customer asking for it, or you set up rules for a specific kind of question. You just try to go and escalate to a human-like and opening a ticket in a ticketing system or escalate inside ZipChat itself so people can take it over. Brent Peterson (09:34.904) One use case I talked to a founder about was he had an AI agent that would respond or he would SMS customers who have left their cart abandoned and they would interact with the customer and they didn't disclose that in the beginning that they were an AI agent. It was just an SMS message and they had a customer who interacted with the agent for the entire weekend. They talked about college sports and all kinds of things and it ended up It ended up costing the merchant quite a bit of money because of the SMS costs and the amount of back and forth. And so one of the things they used was they used an AI agent to guardrail the AI agent. So if something happened or a certain amount of time happened that they would get cut off. Are you seeing the same type of thing where maybe even as a secondary step, instead of escalating to a human, you would escalate it to an Luca Borreani (10:11.694) soon. Brent Peterson (10:32.982) Maybe a guardrail agent. Luca Borreani (10:34.914) So to go back to your example, I would say this kind of behavior where people were starting to talk to the, for example, to the chat, the AI chat. Like if it was ChatGPT as you mentioned, bringing up the building cost for the merchant, it's something that we were seeing at the beginning. There were also lots of people trying to trick the AI, asking what was the master prompt or whatever, these kinds of things. We don't really see it lately, to be completely honest. I guess more people have access to ChatGPT directly or Perplexity or whatever other AI. AI chat, so they just put these kinds of questions there. With that being said, our AI is already super guardrail from our side. So if you try to talk about something that is not coherent with the brand, it's going to just stop you, like say something like, I'm sorry, brand, I don't have information about the thing, but they can help you with whatever brand name products are or brand name is. So it's going to already limit this. But we had cases of customers, again, especially at the beginning, that they were teaching the AI to be superhuman, almost like a psychologist, especially for selling some kind of products. So in this case, conversations were a little bit longer. But still, we have lots of anti-spam filters and stuff like that. So if the AI sees the conversation is going nowhere, it basically starts to cut off. Brent Peterson (11:59.69) Yeah, I saw one fun example was when, you know, one of the older, one of the things that people would do to try to fool an AI agent is say, ask, ask for a recipe for my favorite pizza or something. And I saw somebody that said that it disclosed, you know, I am an AI agent. can't give you a recipe for that, but here's a, here's Tuesday tacos, you know, as, just a fun thing to do to say, here's a recipe to. Luca Borreani (12:22.186) That's good actually. Brent Peterson (12:24.824) And the link went back to their site as a post they did on the site. So it also brought them to somewhere where they could see something that's of value. But also then disclosing that they're, you know, that it's making some fun about it. And just, you know, as a follow-up, they weren't trying to fool it. They're trying to make their agent as much of a human as possible. So they gave it some latitude to have conversations with people without, and the person wasn't trying to fool it. But anyways. Luca Borreani (12:34.04) That's nice. Brent Peterson (12:53.666) How about the, this is from the customer service side and you had also said something around the operational side. There's a lot of operational things that you can get out of AI agents. Maybe that's even more important than the customer service side. Luca Borreani (13:07.566) So yeah, in that case, we're really forwarding ourselves a step at a time because we want to be very careful of what the AI can do on an operational level on specific things because it can be also, if it's not handling that perfectly, it can make some mistakes. So we have some beta programs where we do stuff like that. But yeah, the ability to edit orders, especially for example in Europe, I don't know if you're super familiar with that, we are super big with cash on delivery, especially in certain countries. Asia, Africa, stuff like that. And before what happens is that people get called by call centers to confirm the order. And this kind of operation, of course, is quite expensive for brands. So what we do, for example, is to send an automatic WhatsApp message that, again, in this case, most of the time feels very human, and to try to confirm that. And it's going to keep following up based on the order and the conversation that they had already. Of course, the message will be really one-on-one based. And based on the reply, we'll update the order, upsell the client, add more stuff in the order. We are doing some tests right now to work on, for example, handling refunds. Somebody can send a picture, and depending if the product is damaged. But again, for this kind of stuff that are more sensible and may require the customer to actually get money back from the merchants and stuff like that, we still want a human final decision because again, AI is getting there, is getting super powerful, but there are still a few things that even if it's a mistake only one time, still it's something you don't want to happen. Brent Peterson (14:46.444) Yeah, that mistake gets very expensive. think one thing, especially maybe in the email space, I started using an email AI agent that they don't actually send the email, they leave it in a draft so you can make sure it looks good. Again, another safeguard. As we're going into this Black Friday season, what are AI agents doing for merchants who are going into Black Friday, Cyber Monday? Luca Borreani (15:03.212) except Brent Peterson (15:14.444) What's something new and innovative you've seen that's exciting? Luca Borreani (15:18.67) I believe there's lots of insights to get from, thanks to AI, analyzing what was going on inside your store, inside your AD accounts. I think what Triple Whale is doing is very, very interesting, providing lots of insights about what kind of ads or potentially trends, let's say, merchants should double down and where a better crack is... is potentially, is like a real potential. When it comes to us, one of features that we see, especially the biggest merchants right now, all of them asking and we are providing them this, is basically the ability to get deep inside about what the conversations are about. So they can get to the basically Black Friday. I'm making this joke that now it's like Black November and Cyber December. At the end of the day, it's not even like a weekend anymore. Where basically... They know already what kind of offers or upsells people are expecting based on the past six, nine months of conversations, understanding what can be a good offering because people maybe were talking about discounts in the chat, understanding which marketing angles could be better for them. And all of this is basically an agent analyzing the conversation, drafting trends, stuff like that. Brent Peterson (16:41.068) Yeah, I think that you've really hit on it that the most important part or things we could be using with LLMs is that analysis of the data and collecting all that data. I think, you one thing I've heard, especially for the Cyber Monday, Black Friday period is that there's so much volume. Make sure you're collecting that data during that volume to use that data for not just the next year, but just as you're going forward to see how your customers behave. Luca Borreani (17:09.28) Absolutely, absolutely. And generally speaking, if you talk about volumes, of course, one of the biggest problems of Q4 in general, and of course, that month and a half or whatever, is you're going to get so many orders, but at the same time, you're going to waste so much money on traffic that will not convert. Everybody is going to be selling kind of the same offers, very aggressive offers, so your margins will be very thin. The ability to have an AI agent proactively engage customers on the website, talking to them, understanding what they want, recommending the right products straight away. Whenever they have a pre-purchase question, just reply and be on point. At the end of the day, most customers just want to be heard. They want to feel like on the other side, there is an interaction and a real entity or something real happening. So as soon as they see it, they're ready for closing. the purchase or buying the product. So all of these kinds of things are very important. And then of course, after the heavy discounts, everybody will come back asking like, okay, where's my order? I bought the present for my wife, must come before Christmas. So all these kinds of stuff you want to automate as much as possible. Brent Peterson (18:25.43) Yeah, perfect. Luca, we have a few minutes left in the podcast. As I close out the podcast, I give everybody a chance to do a shameless plug about anything they'd like. What would you like to plug today? Luca Borreani (18:36.212) I mean, I would say that everybody right now is finally looking into I seriously. It was something like two years ago, one year ago, everybody was, even like when I was going to conferences, like shop talk. They were asking questions, trying to understand if AI could actually be good for their business. Now everybody's sure about that. The only thing is that they kind of don't know how to apply it properly. They see all these complex workflows on LinkedIn for post-engagement farming. They get excited, but then they feel it is very complex. At the end of the day, AI, whatever it is, It's much simpler than that. And somebody already figured it out how to apply to your business in a easy and kind of plug and play way. So just dive into it, make experiment at the end of the day. It's one of the first technologies that really compounds over time. So regardless of the application you use, it's being for whatever you do in your business, it's going to just compound over time, learn more, understand more your customers, understand more about your operation, business, team, brand identity, everything. So the longer you're going to be using it, the better it's going to be. Brent Peterson (19:43.81) That's perfect. And how can people get in touch with you? Tell us how they find you. Luca Borreani (19:47.023) Yeah, absolutely. They can find me on LinkedIn, Luca Boreani. They can find me on Instagram as well. If they want to send me an email, LUCA@ZIPCHAT.AI Super simple. And if you want to try it, it's 7-day trial, 30-day money back guarantee. And also we have a discount, 10 % off forever using coupon code podcast10. Brent Peterson (20:10.344) Awesome. Great, Luca. It's been such a fun conversation. Thank you so much for being here. Luca Borreani (20:15.129) Thank you, Brent, for inviting me.