Inside Just Eat's AI Lab: Voice Agents & Agentic Commerce

Gurthie Cooper: [00:00:00] I'm Guthrie Cooper. I head up product for the incubator domain, and that is part of Just Eat Takeaway, which is a company that we work for, and we're obviously within the Process Group.

Nidhi Sharma: I'm Nidhi Sharma. Majority of my career had been in the grocery and retail. So I worked for multiple retailers to build their e-commerce website as an engineer, then moved into architecture role and then into engineering leadership.

Nidhi Sharma: You know, I could just frame the, uh, AI innovation and incubation a little bit just to give a little bit oversight of, uh, you know, what this team is all about and how we are structured. Cool. Um, so currently from the team structure-wise, we have-- we begin with an incubation within incubation. So we have a team called Pre-seed.

Nidhi Sharma: It's

Demetrios: like inception.

Nidhi Sharma: Yes, it's like inception. So we, we-- this was one of our learning as well. So what we have is a pre-seed team, and they are all about just, you know, bringing a prototype to [00:01:00] life in two to four weeks.

Gurthie Cooper: Wow. In incubator, we operate like a startup. We're really-- we are inventing the next wave or waves of this industry.

Gurthie Cooper: Yeah.

Nidhi Sharma: So li- right now in the world of AI, we see that, you know, every other day there is a new protocol and there's a new technology. So we need a bunch of people to just, you know, jump on it, try it out, see if it has potential, and then we add it to our pipeline, and that's why we call it incubator within incubator.

Demetrios: Mm.

Nidhi Sharma: And then rest of the teams are basically working on different themes. As you know, that world, world is evolving faster than we think. So we have-- we just have to, you know, come up with multiple ideas working in parallel. So some of our teams are working on autonomous delivery. Some of our teams are experimenting with all the AI innovation things like evolving our existing AI assistant to a fully agentic commerce, um, pers- [00:02:00] personal shopper agent.

Nidhi Sharma: And Gertrude will explain more about, you know, the vision for the personal shopper. Um, and then, uh, some of our team are also working on our off-the-platform strategy because we work with bold, uh, vision that there is a possibility that we'll not have any apps in the market, or we may have just, just super apps.

Nidhi Sharma: Mm-hmm. We don't know how the, uh, you know, the consumer will interact in future.

Gurthie Cooper: We launched back in the year 2000. Oh, wow. So we were the first, uh, food ordering online platform in Europe. always been about innovation, um, and it continues to be about innovation. With the Process acquisition that happened last year, we, we have Jet Fuel, um, uh, plugged into, you know, that, that, uh, culture of innovation.

Gurthie Cooper: So 2000 on the web, around 2009, we were one of the first, uh, apps, so we moved obviously to the app ecosystem. Mm-hmm. That was the big change that happened then. [00:03:00] Um, between 2009 and where we are now, uh, you know, technology innovation was a big part of what we were doing, but it was also about adding more, um, uh, merchandise for our consumers to purchase, right?

Gurthie Cooper: So the consumer expectation continues to evolve over time.

Demetrios: And in, in 2000 it started as a food delivery service. Because Jet now, I guess for the Americans that are listening who may not be so familiar with it- Yeah ... it is a food delivery service. You can order from restaurants, you can order from supermarkets, you can also order from other places too, right?

Gurthie Cooper: Exactly, yeah. So in the last couple of years we've added grocery. Grocery's been a big growth lever for us. Mm. We've added, uh, electronics. You know, our vision is to empower everyday convenience, and if you really think about what that means, everyday convenience, consumers have many needs acro- over the, the course of their day, over the course of their week, over the course of their month, and we are increasingly there for every [00:04:00] single moment that they need to purchase something, they need something instantly.

Gurthie Cooper: Give you an example. If you're playing PlayStation, uh, you know, with your, with your family or your friends and one of the consoles breaks, you can order a replacement console and you can get it in 20 minutes-

Demetrios: Wow ...

Gurthie Cooper: from Just Eat, right? So innovation is on the tech side, but it's also very important to make sure you have the right merchandise for the customer at the right time.

Gurthie Cooper: Um- Looking at the future, so here we are, you know, early 2026, consumer expectations and the tech landscape is evolving faster than ever. It's one of the most ex-exciting times that I've had in my career, right? So, and to N-Nidhi's points, we are very interested in ensuring that we are exactly where customers are, right?

Gurthie Cooper: So we wanna meet them where they are, and that means, uh, on our app and off our app. So making sure that we're present... We call it beyond the app. That's the name for our strategic theme, [00:05:00] right? So whether they're using chat programs, whether they're using search, and when I say search, I mean people will think Google, but of course, OpenAI is cl- is battling Google for that top of the funnel, you know.

Gurthie Cooper: And I think the metrics at this point are, you know, 2% of global search is done in ChatGPT. The growth rate of that search, uh, uh, of that search, uh, of that discovery, uh, in space is 60% year over year. Google is still the majority, but they're growing at a slower rate. So the consumer landscape is changing, and in Incubator, we exist to make sure that we're one step ahead of consumers, and we're there for them.

Gurthie Cooper: Voice is not a new platform, it's a new mode.

Demetrios: Mm.

Gurthie Cooper: Right? And the voice feature that we've released is built on top of the most innovative tech platforms available right now. It is fully conversational. We can maybe do a small demo later so I can show you what I mean. But it is, uh, it's contextual, it has [00:06:00] personality, you can interrupt it, uh, you can have a conversation with it, and it uses all of that context within the conversation, and it makes recommendations for you.

Gurthie Cooper: Some of the benefits of that are, you know... So if you think about, uh, accessibility, right? So people who would prefer... There's many customers who would prefer to use voice versus use their fingers or some other paradigm. That's why you see companies like, uh, ChatGPT, Gemini, et cetera, um, integrating voice.

Gurthie Cooper: You know, a lot of people are using that. During the course of your day, there are moments when it makes much more sense to use voice. For example, uh, you're driving your car. Yeah. You're on a bike. And the vision there, if we think beyond the app, and we pair that with another strategic theme that we have, which is conversational continuity, uh, maintaining that context, maintaining the context of the conversation across modes and touchpoints throughout your day.

Gurthie Cooper: That's what we believe is gonna be, um, you know, a really powerful lever for our [00:07:00] growth as a company, and it will meet customers' expectations.

Demetrios: Let's talk for a minute about these different touchpoints. And basically what I'm hearing you say is memory. You just wanna be Remembering everything in every mode that folks are engaging with the app.

Gurthie Cooper: Yeah.

Nidhi Sharma: So when we say beyond the platform, we talk about our platform technically, but the context and the intent from the external apps.

Demetrios: Okay.

Nidhi Sharma: Or it could be from our app as well. So for, for, for example, if you are using ChatGPT, you are basically ChatGPT's user.

Demetrios: Yeah.

Nidhi Sharma: So your intent and conversation will remain with ChatGPT as per the privacy consents and everything.

Nidhi Sharma: But the moment you switch to Jet, uh, to pay, uh, to buy something, that's when Jet app will start your context. Mm-hmm. And with the consent, we will maintain that context. [00:08:00] But when we say conversational continuity, what we mean by that is that you are in your car and you're hungry, and you start a conversation to get something, uh, but you have to switch to your mobile later, or you might have to use your laptop.

Nidhi Sharma: The same conversation should be continued. So the context should be there, the information should be there, and the flow should not break. Hmm. Um, that's what we mean by conversational continuity. But when it comes to voice agents specifically, when you start a session, uh, that session context is maintained.

Nidhi Sharma: So that conversation is being maintained, um, in the memory. Uh, and based on tools like, you know, your order history, your dietary preferences, uh, you will get the right recommendations.

Demetrios: Yeah. Okay. If you're populating things on ChatGPT's apps or Claude apps, I guess MCP apps is what they call it now? [00:09:00] How much of that information are you getting back so that you can help the MCP apps give the right things and make sure that you are re- recommending the proper things to the user?

Demetrios: If you don't have any of that information, now all of a sudden it's like you're flying blind or you're just getting the intent from ChatGPT or Claude, right? You don't get to decide what the intent is or try and puzzle the intent and put new things in front of the user that you think they might want or upsell them on things.

Gurthie Cooper: Yeah. It, it depends on the platform that we're working with, right? So, uh, for example, Google might have, um, different policies in place, different tooling in place versus an OpenAI, right? We saw a talk from Google last night where they're really, you know, Google is, is, uh, famously an open platform, right? So that memory, that context might [00:10:00] be available to us, um, in some partnerships more than others.

Gurthie Cooper: The dream is to have the right kind of context, the right amount of memory, the most relevant data available across platforms at all times, across modes and across platforms. And it even goes beyond, you know, uh, ordering agents, ChatGPT, Gemini, et cetera. You can think about wearable devices, so for example, your smartwatch or- Yeah Think about health da- data, right?

Gurthie Cooper: So food and health are obviously- They're

Demetrios: intertwined, yeah.

Gurthie Cooper: Exactly, right. So obviously this, uh, you know, I need to caveat this by saying we take, uh, user privacy, user data extremely seriously. Um, so we never wanna do anything that's obviously unleggislated for the regions that we operate in. And of course, user consent is, is super key.

Demetrios: Yeah. Disclaimer. Let's put that in there just so everyone knows that it's good.

Gurthie Cooper: Indeed. I mean, it goes without saying, but it is super, [00:11:00] super important to us, right?

Demetrios: But it does-- it opens you up to a new world. If you're using health data-

Gurthie Cooper: Yeah ...

Demetrios: I just had a workout this morning, and now you probably know if you're reading my WHOOP-

Gurthie Cooper: Yeah

Demetrios: that I did a workout. Oh, I should order you some-

Gurthie Cooper: Yeah. Well- ...

Demetrios: whatever, protein-rich food.

Gurthie Cooper: Exactly. It actually goes one step further. So I'm gonna, I'm gonna give you a, a story that combines that, uh, health data with some agentic proactive use cases as well, right? So we know that you, Demetrius, order the same coffee from the same partner, uh, regularly, right?

Demetrios: Yep.

Gurthie Cooper: So you've, you've had your workout. We know that usually after the workout you order that coffee, right? So we proactively prompt you to order that coffee. We might even place the order for you, or there might be some, uh, healthy friction in there where you just get a notification on your, on your watch or in the app, and you say, "Yeah, go ahead and buy it."

Gurthie Cooper: On the way to your coffee, you are on your bike and you realize, "Oh, I'm gonna invite a buddy to join me." Mm-hmm. Right? So now you need two, two drinks in [00:12:00] that order.

Nidhi Sharma: Yeah.

Gurthie Cooper: You can pick up where you left off on voice 'cause you're on the bike, and you say, "Hey, the order that's being placed, I need an additional cappuccino.

Gurthie Cooper: Can you add that to, to the order?" That can be done through voice mode, right? Yeah. And then, uh-

Demetrios: Wait, can I stop you there? Please. Because I think that is a beautiful vision, but the friction or the problems that I generally have when I'm in that experience is that I'm on my bike, I can't open my phone and open the app.

Gurthie Cooper: Yeah.

Demetrios: Or I do, but it's a pain in the ass like to be r- riding, trying to focus on the phone and then find the app and then... So what I'll try to do normally, and this is where it fails, is I'll say, "Siri, open the app."

Gurthie Cooper: Yeah.

Demetrios: Um, open Just Eat, and Siri sucks, so it gives me some random app where it says, "Oh, I don't have access to Just Eat," or, "You need to scan your face ID for that."

Demetrios: And I'm like, "I'm on the bike. I can't scan it right now." Or I will have to whip it out and it's kinda that same thing. So I [00:13:00] see the vision, I think it's amazing, but you're almost beholden to how bad the other voice agents are out there.

Gurthie Cooper: Yeah. The ecosystem and the way the ecosystem interacts is super important.

Gurthie Cooper: Mm-hmm. And that, that friction point you just described now is exactly why our domain incubator exists. Mm-hmm. We see those friction points as absolute gold. So that's an opportunity, right? So we would take that use case, um, let's assume, and we could probably quite safely assume that we do a bunch of research, which is a really important part of our process, and the friction point you've raised comes up as, as a common thing.

Demetrios: Yeah.

Gurthie Cooper: Right? We have really tight relationships with many of the companies you mentioned and others, right? And they're looking to solve those problems as well. So we would team up with them to make sure that we can find a way to get that right, and it could be through an integration with your Apple Watch.

Gurthie Cooper: That's the level that we're operating in. Those, those connection points are the key because it's very easy for me to sit here and map out the happy path.

Demetrios: Yeah.

Gurthie Cooper: But the work is, you know- Yeah ... understanding those friction points

Demetrios: and solving them. Exactly where [00:14:00] it fails and if it fails before it even gets to you-

Gurthie Cooper: Yeah

Demetrios: then you don't know why. It's like a silent failure.

Gurthie Cooper: Yeah.

Demetrios: Yeah.

Nidhi Sharma: Yes, and that's where I think Agentic AI commerce comes in play because what we-- One of the thing with Agentic AI commerce is that we should be, we should integrate all the possible surfaces for the frictionless commerce. Yeah. And one of the surfaces, this, uh, you know, scenario where the problem begins with your inbuilt app in the iPhone.

Nidhi Sharma: How do you actually improve that side first? Because you need that as an entry point. Even if you solve the problem-

Gurthie Cooper: Yeah ...

Nidhi Sharma: with building multiple agents for ordering, decision-making, recommending, but at the same time, the basic fundamental blockers are still there. As a customer, you won't be satisfied.

Gurthie Cooper: Exactly.

Nidhi Sharma: So [00:15:00] this is where I think these new protocols where the, where you have system-to-system integrations Are going to add some values and will provide the frictionless commerce.

Demetrios: And system to system, you mean like MCP?

Nidhi Sharma: Yes. Okay. MCP, agent to agent- Yeah ... basically. Um, and then there is, there, there is also, you know, potential and ant- anticipation that instead of having multiple apps, you will only have one lifestyle app, for example.

Demetrios: Yeah.

Nidhi Sharma: And that lifestyle app can actually read when you are bike- when you are riding, when you are free, and then based on that, you know, make sure that f- your phone is unlocked, for example. Mm-hmm. Uh, but that is, like, all these things needs to be think about, and then based on the research, there should be some solutions in place.

Nidhi Sharma: What we know as of now is that there are multiple surfaces, there are variables like, you know, with your Meta [00:16:00] glasses, we can integrate our app, or, um, if you are just using Gemini all the time, or if you are just using ChatGPT. So you did mention about, you know, how this user interaction will work and how would Jetbe benefit with this kind of integration where you're actually using ChatGPT, and your intent and your context and your behavior and everything is with ChatGPT.

Nidhi Sharma: We, Jet is there just to, you know, provide you the, the best recommendations. So I think it depends what kind of user base we are talking about.

Demetrios: Mm.

Nidhi Sharma: So the, the Jet already has the user base. We already have, uh, you know, large amount of customer data, and we do have their order history based on the consent and everything.

Nidhi Sharma: Uh, similarly, ChatGPT has their own, or OpenAI, their, has their own user base. Some of the users are already Jet customers- Mm ... some not. So we need to think about, you know, these things individually and build the use [00:17:00] cases and solution around those use cases. But this, these, th- this is still new.

Demetrios: Yeah.

Nidhi Sharma: The kind of contract, the kind of standards, and the kind of collaboration, uh, we'll have in this B2B scenario, uh, will define how this, uh, you know, relationship will work.

Nidhi Sharma: But we need to keep customers' privacy at top. Mm. We need to take the right consent with them. In this agentic AI commerce, we shouldn't just allow, you know, sharing data between two vendors or two companies without even, uh, letting our customer know that this is what we are going to do with their data, even though it's for customer convenience.

Demetrios: Yeah. I do see a world where-- 'cause I think about how I use three or four different LLMs daily, and Sometimes it's just out of convenience that I'll use one or the other 'cause it's there. It's like, all right, I'm using Antigravity with some, with Gemini, and then I've got Claude, and I've got a hotkey where I just press Caps Lock and I instantly [00:18:00] start talking to Claude.

Demetrios: And there's probably a subset of users that you said, like they're Jet customers and they appreciate the tight integration. And so they're going to be able to configure into their different LLMs like, "Hey, I'm a Jet customer. Here's my whatever Jet customer ID," and you wanna be able to have a very tight coupling so that anytime I ask for anything, it's like a default that it will just go and use Jet.

Demetrios: But then there's another subset of customers where it is like searching for a restaurant on Google and you get the blue links, and some of those are sponsored and some of them are not

Gurthie Cooper: Yeah, there's a spectrum, and I think we need to be prepared for a world where there'll be many, many different types of user appetites, right?

Gurthie Cooper: From your early adopters, your Dimitra style, uh, users, all the way to, you know, we serve all demographics, and we're in 16 countries, right? So [00:19:00] luckily you say there's, there's on a long enough timescale, perhaps the apps, the apps, the front end of the apps disappear, and it's just like a fragmented, uh, uh, you know, ecosystem of, uh, consumer tools.

Gurthie Cooper: Probably the reality will be somewhere in between those two. Um, so we need to make sure that we're able to serve consumers no matter how advanced in technology usage they are or what they want, right? And you should be able to change that over time as well. Yeah. So if you're a ChatGPT user and then you decide, actually, you know, I, I'm not into talking to LLMs anymore, I wanna go back to the app, you should be able to do that, right?

Gurthie Cooper: For sure.

Demetrios: Yeah. Or you switch from ChatGPT to Claude or to Gemini.

Gurthie Cooper: Absolutely. Absolutely.

Nidhi Sharma: Yeah. And, and I- I think we all have our favorites.

Demetrios: Yeah.

Nidhi Sharma: Uh, we still-- Even though there are so many LLMs, we still think like, you know, some of those really understand us.

Demetrios: Yeah.

Nidhi Sharma: And I do have my favorite because I feel like I, I get the right kind of responses, and I've been heard and understood properly.

Nidhi Sharma: And that's, [00:20:00] that's the vision actually. The more you know about your customers, the more you know about their intentions, their feelings, the better you can serve them.

Gurthie Cooper: Mm-hmm.

Nidhi Sharma: And that is why one solution cannot be for every customer. We need to just look at, you know, the kind of behaviors they have and then serve them accordingly.

Nidhi Sharma: And that's what I think in, in Jet, that's what we live by, that customer convenience and meeting customer where they are. Mm-hmm. Uh, not just physically, but also emotionally and, you know, what, what kind of, uh, what kind of things they're looking for, be it food, be it any convenience, any other convenience, uh, from general merchandising in the example which, uh, Gatri just shared.

Nidhi Sharma: So th-this is the kind of convenience we want to provide, but there should be that alignment with you as a customer really need it because we don't want to over-serve at the same time.

Demetrios: Mm-hmm.

Nidhi Sharma: And I'll share some examples where we feel like, you know, this will be instead of customer convenience, it will become [00:21:00] customer con- uh, grievance.

Nidhi Sharma: Uh, and push notification is one of the example where you keep sending notifications for different vouchers and offers and, you know, you buy this from here and you get this. Sometime that is like too much overhead for a customer to keep track of these things-

Gurthie Cooper: Yeah ...

Nidhi Sharma: even though you are providing the convenience.

Nidhi Sharma: So we have to maintain that balance, uh, when we think about, you know, what exactly, what type of problems we want to solve and how we want to come across as a brand, as well as how we want to come across as, as, as an organization who is actually solving, um, and fixing the real challenges people are going to have in AI world and even exist today.

Demetrios: That is exactly what I thought about when you were saying proactive.

Gurthie Cooper: Mm-hmm.

Demetrios: And I was like, oof, you-- it's so dangerous Because as soon as you send two notifications that aren't relevant, I turn off notifications.

Nidhi Sharma: Yeah.

Gurthie Cooper: Yeah.

Demetrios: But if it's relevant, it's like, "Oh, [00:22:00] damn, that's actually kind of nice." I just see it with all of the apps that I have in general.

Demetrios: I will turn off notifications because it's usually just like, "Oh, save 20% if you do this," and it's trying to get me to use the app, which isn't necessarily that I don't wanna save 20%. It's just that in this moment in time, I don't have a use case for what you're trying to push me. Yeah.

Gurthie Cooper: Yeah.

Demetrios: So it doesn't make sense- Yeah

Demetrios: that I would use your app. It's not that I'm not using it because I don't want to.

Gurthie Cooper: Yeah. It's

Demetrios: just that I don't have a use case right now, and so you have to be really smart about when to do that.

Gurthie Cooper: Yeah. Yeah. Very few companies have, have cracked this, right?

Demetrios: Yeah.

Gurthie Cooper: Um, and I really think it is such a huge opportunity.

Gurthie Cooper: The last thing you wanna do is get people to turn off notifications.

Demetrios: Yeah.

Gurthie Cooper: Right. Because then you've lost a chance.

Demetrios: Yeah.

Gurthie Cooper: But if you get, um, if you get context right, and you get the moment right, and you understand your customer base and [00:23:00] their history to the right level, it can be done, right? A- a- if I can pick up this, the story where I left off.

Gurthie Cooper: So- Oh

Demetrios: yeah, I totally cut you off. Sorry.

Gurthie Cooper: No worries. It was a good time to jump back into it. So, so, so, um, coffee story is done. You've had your workout. On the way home, you're on the app, and you wanna place a grocery order, right? So you can use the traditional app to place that grocery order, or you can go to our chat assistance, right, which is a text-based assistant, smash like ChatGPT.

Gurthie Cooper: It has the voice mode. So the voice mode is really just voice, the voice version of that chat app. So we have both of those modes. So you can use that app to build a grocery list for your evening's dinner with your family. So you could say, "Hey, um, I wanna make a carbonara for the family." Um, this is how many people there are, any dietary restrictions, et cetera, and it would pull that list of, of ingredients together.

Gurthie Cooper: It might even surprise you with some extra ingredients that you didn't know were in the recipe, for example. Beyond that, so let's go back to you- you've got a WHOOP on, right? Yeah. So let's assume we have [00:24:00] an integration with WHOOP or, or Apple Watch.

Demetrios: Speaking of shitty notifications.

Gurthie Cooper: Yeah.

Demetrios: WHOOP. Oh my God, dude.

Demetrios: I am... I, I complain about this like every other podcast, but they're so passive-aggressive on me. They're like, "Your sleep, uh, score could be better."

Gurthie Cooper: Yeah.

Demetrios: I'm like-

Gurthie Cooper: Yeah. Yeah ...

Demetrios: I already feel like shit, WHOOP. You don't have to remind me.

Gurthie Cooper: And you've slept already, so you can't

Demetrios: fix it. Yeah, it's not like I can retroactively go back in time.

Demetrios: Yeah. So anyway, sorry, that's just a grievance I have with WHOOP.

Gurthie Cooper: I hear you. That's, that's actually why I don't have one

Nidhi Sharma: of those ones. That's

Gurthie Cooper: what I said.

Nidhi Sharma: Yeah? Uh, it can easily become a grievance. Yeah, exactly. The convenience can be a grievance as well. It's a

Gurthie Cooper: grievance. So agree, convenience can also become delight.

Demetrios: Yeah.

Gurthie Cooper: Right? So, so what I'm getting to is like you order that carbonara, those ingredients, um, and you get-- imagine the assistant nudges, nudges you in the right way with language that is tailored to you so that it won't become a grievance to say, "Hey, you know, we noticed your, [00:25:00] uh, heart rate variability was a little different this morning versus your baseline.

Gurthie Cooper: You might wanna consider skim milk in that pasta."

Demetrios: Mm.

Gurthie Cooper: Right? I think- Does

Demetrios: that actually work or did you just make that up? That

Gurthie Cooper: doesn't work yet.

Demetrios: Okay.

Gurthie Cooper: But well, let's, let's-

Demetrios: No, I mean the skim milk- Oh, there's

Nidhi Sharma: skim milk in pasta ...

Demetrios: from my heart rate variability. '

Gurthie Cooper: Cause I'm not a doctor, don't play one on the internet.

Gurthie Cooper: Sometimes. But

Nidhi Sharma: you can put milk in pasta sauce.

Gurthie Cooper: Yeah.

Demetrios: Yeah, yeah. But anyway, yeah, that integration, like understanding a 360 view of you and saying, "I see this happening in your-- all of these data points that I have for you, and here's my recommendation."

Gurthie Cooper: Exactly. And, uh, I-- we can play, we can test that out in the voice app today.

Demetrios: Yeah.

Gurthie Cooper: We don't have the, uh, health data plugged in yet, but you can give it that data in the conversation. Mm-hmm. So you can say, "Hey, I'm trying to lose weight. Um, can you just tailor [00:26:00] this recipe a little bit for me?" Or, you know, yesterday I was ordering, uh, a burger and I said, uh, "Yeah, I want the healthiest option possible."

Gurthie Cooper: And it made some recommendations around bun choices.

Demetrios: Did it say just no burger? That's not... You

Nidhi Sharma: want healthy?

Demetrios: Do you want a burger?

Nidhi Sharma: Yeah. I'm serving you dinner.

Gurthie Cooper: It offered me a vegetarian option.

Demetrios: Oh, there you go.

Gurthie Cooper: No,

Demetrios: this one's

Gurthie Cooper: good.

Demetrios: All, all vegetables.

Gurthie Cooper: Indeed. Indeed.

Nidhi Sharma: Just

Gurthie Cooper: vegetables. Yeah,

Nidhi Sharma: yeah. Yeah. I think food is the integral part of our being.

Demetrios: Yeah.

Nidhi Sharma: It touches upon the health, it touches upon the mood-

Demetrios: Yeah, social ...

Nidhi Sharma: and the lifestyle and the social interaction, and in some culture, food is a language of love-

Demetrios: Yeah ...

Nidhi Sharma: and connection. So making it valuable and ensuring that, you know, there is a technology as well as the intent clearly defined and balancing out for the, for, for an individual is tricky, but at the same time it's kind of, you know, a social [00:27:00] cause where you feel satisfied and complete that you are actually adding value to somebody's life- Yeah

Nidhi Sharma: by helping them to choose the right options when it comes to eat.

Demetrios: Yeah.

Nidhi Sharma: Uh, and that's where I think the vision for our personal shopper is to ensure that we understand, or this agent understand you fully as an individual and surface the recommendations which are very much tailored for you. We, we know the hyper-personalization is quite a terminology, but in reality, in the world of food and grocery and, uh, you know, providing the customer convenience, the hyper-personalization is that tailored individual approach for, for, for that particular person who is using it and, and availing those, uh, services which we are providing.

Nidhi Sharma: And, and that's, that's the vision. So it could be, you know, the, the persona could be a single mother or you could be a working professional or you could be something [00:28:00] like me. I mean, I'm, I'm health freak for a couple of months and then I completely go nuts and then I'm... So it's like versa- versatile personality.

Nidhi Sharma: But if that agent or that personal shopper still understands me and, you know, tailor those, uh, those recommendations, I will be like glued to it forever.

Demetrios: Yeah. You know

Nidhi Sharma: what

Demetrios: I just realized, and I would love to see data on this. You probably have it, uh, 'cause we're, we're talking to iFood, we're talking to all these different companies in the Process Group that are, are doing and trying this innovation stuff.

Demetrios: And with the iFood, like, magic chatbot that they created- One thing that is difficult with chat is getting more context. But with voice, I just naturally am ready- Yeah ... to give up more information.

Nidhi Sharma: Yeah.

Demetrios: Have you tried both of those modes and you see-

Nidhi Sharma: Yes ...

Demetrios: a big difference? I, I can imagine, [00:29:00] like I explain much more when I'm speaking through voice than I do when I'm chatting because chatting, first of all, just like holding and looking at my phone, nobody wants to do more of that during the day.

Gurthie Cooper: Yeah, I fully agree. Absolutely. So, you know, an ex- a concrete example of that might be, uh, "Hey, I need a burger for lunch," and, you know, the assistant starts pulling those, uh, options up for you. If you're on, if you're on chat and you, you wanna just make an adjustment and say, "Hey, I'm trying to lose weight," like I said earlier, uh, you might not wanna do that, interrupt the flow of the chat 'cause it just doesn't feel natural, right?

Gurthie Cooper: On voice, you can just interrupt the assistant. She'll stop or he'll stop, listen to you, and quickly adjust what she's recommending. So yeah, voice is, it, it, it is about, um, natural language, right? So chatting with, um, your e-commerce provider via text is a big move away from navigating menus and, and GUIs, and that's where we've been for the last 20 years.

Gurthie Cooper: Voice is [00:30:00] even more natural, so I'd agree with that. Yeah. But you know, there's, there's space for both.

Nidhi Sharma: And I can share my personal experience. So when we were testing voice internally, so within incubation when we are done with any of our ideas, uh, we just test internally. Um, so I was-- I had a party at home and, you know, when you test your internal product, you just want to be very, uh, careful that you don't, you don't screw up the product itself, right?

Nidhi Sharma: So firstly, I was giving very simple instruction. Can I have a vegetarian burger? Or very simple and easy instruction which voice can help, and this was giving me a, you know, kind of a satisfaction and happiness that, oh, yes, this product is working. This is a cool product.

Gurthie Cooper: Mm-hmm.

Nidhi Sharma: But I actually had a party on Sunday, and then I said, uh, "Can you actually arrange a party for 10 people, these many adults and these many childrens?

Nidhi Sharma: And don't tell me the recommendation. This is my budget. Just add things into my basket." I was just testing it, and it actually [00:31:00] happened, which was really great, and that gives me, you know, a clarity around when you talk, how much you express yourself.

Demetrios: Exactly.

Nidhi Sharma: Because to your point, if the same scenario, uh, has happened with the text or typing, I might not type this much, like, you know.

Demetrios: Yep.

Nidhi Sharma: I might have said something more like I'm feeling anxious or I'm feeling this and that. Um-

Demetrios: But you're so much more direct when you type because it is so much more cumbersome to actually type, especially if you're on the phone. And I've noticed this just using a lot of these speech-to-text tools that I am F- way more verbose when I talk.

Demetrios: Uh, so usually on a lot of the sp- speech-to-text tools, I'll have to pare it down, or I will make sure that I know what I wanna say before I actually hit the start dictating button.

Gurthie Cooper: Yeah.

Demetrios: But when I'm, I'm talking, I'm giving all this context, and I find that I'm much more lazy [00:32:00] to type now because I've gotten used to the mode of talking.

Gurthie Cooper: Yeah.

Nidhi Sharma: Yeah Yeah, but this voi-voice agent, the moment you want to see, like, you know, once it says that, "Okay, these are the things in your basket," you can switch easily, uh, because you can see the cards, you can swap and, uh, make decision whether you want to go for that dish or not. So it's quite intuitive.

Nidhi Sharma: Does it give you the recipes,

Demetrios: I imagine, too?

Nidhi Sharma: Uh, recipes not in the details, but if you ask for recipe, it can, yeah- It can, yeah ... it will explain. Like you were

Demetrios: saying, like order me or-

Nidhi Sharma: Yeah ... plan

Demetrios: my party for me, basically.

Nidhi Sharma: Yeah.

Demetrios: Order everything I need. Does it also say, "Oh, well, how about some flower arrangements

Nidhi Sharma: too?"

Nidhi Sharma: Yes. Yeah? Yes, it gives those recommendation. Yes. That's cool. And because we do maintain your order history, um, and your dietary information, so it will also say that last time you tried the fries with this burger, would you like to add that in the basket? Uh, things like that. Yeah. So it's quite proactive, uh, to allow-- to [00:33:00] help you to make decisions and react.

Demetrios: And where does it stop? Because as you're talking about this, food is almost like the core, right? But then you start going out and you have tech and you have, uh, groceries, which is food, so that's not that big of a leap. But is there a world where I am traveling, you know that I like exercising, and there's a gym nearby, and so it's like, "Oh, well, we got your day pass at the gym if you want it."

Demetrios: Boom. Yeah. I can now go to that.

Gurthie Cooper: Yeah.

Demetrios: So where-- Do you have a line for what you wanna do, or is that kind of your job as the innovation lab to continue to expand the foot-footprint?

Gurthie Cooper: With the acquisition of Process, we now have such a bigger playing field, um, and we can really start to think about the partnerships- Yeah

Gurthie Cooper: that this company has, um, and the companies that Process owns, and we can start to serve, so we say everyday convenience for food and retail. It can be way bigger than that, right? So like you mentioned, it can be travel, it can be, [00:34:00] um, hotels, it can be secondhand goods, it can be you name it. So absolutely, we see a world where I think there will be, like I said, there'll be a spectrum.

Gurthie Cooper: You know, some people will prefer to use individual apps for those use cases, but the super app concept, and we see this, uh, uh, Tencent and WeChat is huge on this and have been for many years. Um, it works, right? And then if you think about, you know, what's happened in the last couple of weeks with, uh, Claude Bots, MultBot, whatever it's called now, probably changed their name today.

Gurthie Cooper: I

Nidhi Sharma: don't

Gurthie Cooper: know. OpenClaude. OpenClaude. Yeah, yeah. So that, that was my hot take the other night. It's like, I don't know, it, it's a very techy, sort of risky thing right now, proposition right now. But very quickly, you know, I believe that consumer expectations are gonna move in this direction. So what does that mean?

Gurthie Cooper: That means you should be able to either have an agent that can accomplish those tasks for you-

Nidhi Sharma: Mm-hmm ...

Gurthie Cooper: objectives at a high level, plan my next week. You know what's going on. You have access to my calendar, you have access to my email, you know what I like, you know my [00:35:00] preferences, and I can make really highly tailored dec- personalized recommendations and actually do the things for you.

Gurthie Cooper: Book the flights, book the hotel, contact your friends even and set up like a group chat so that you can meet up. All of those things are absolutely gonna be possible. And I think it's a, it's kind of like a chicken and egg thing, right? So we could start building that tomorrow, but are consumers gonna be interested in that?

Gurthie Cooper: Are they gonna be worried about their privacy? It's really encouraging to see, uh, things like, uh, like Claude, um, I would say sort of paving the way because that, that's, you know, this is-- it seems like a little side, sideshow right now, but once consumers get wind of that... And, and again, like AI seems like so much more commonplace now.

Gurthie Cooper: You know, my, my folks-in-law, like they know what's going on in AI. It's not, it's, it's not backroom stuff anymore. Once somebody figures out a way to, um, wrap that in a very convenient consumer packaging, people will start to think, "Hey, why don't I have my own personalized agent? Can I just make my own?[00:36:00]

Gurthie Cooper: What are the other options out there that can do these things for, for me, that can really simplify my life?"

Nidhi Sharma: And the beauty is that technology allows us to achieve all these goals.

Gurthie Cooper: Yeah.

Nidhi Sharma: We have everything in place to ensure that we can actually do these things. So just to answer your question again, like the vision can be broader and broader based on, you know, the, the, the, uh-- there's so much potential out there.

Nidhi Sharma: Yeah. And with this new era, like agent e-commerce and billing agents and like couple of steps, it's all achievable.

Demetrios: Hmm. With infinite options now and kind of all this blue space to run in, how do you weigh what to tackle next?

Gurthie Cooper: It, it's about combining a couple of things, right? So we need to understand what consumers want now.

Gurthie Cooper: [00:37:00] We need to have opinionated bets on what consumers are gonna want in the future.

Demetrios: Mm-hmm.

Gurthie Cooper: And we combine that with what the company strategy is, what the company vision is, and what our business goals are as well, right? So how do you balance, um, building all the things with concrete growth expectations in the short term?

Gurthie Cooper: So how we do that in the incubator is we invest heavily in what we call scanning the horizon, right? Mm-hmm. So we have teams that are really thinking into the future and, uh, talking to customers, looking at all of the tech innovations that are happening in our markets and in, uh, States as well, and in China, just so we can really have those strong opinionated views on, on what we believe is coming next.

Nidhi Sharma: Mm-hmm.

Gurthie Cooper: And then we work in, you know, it's, it's an old paradigm now, but it still really works, the lean startup.

Nidhi Sharma: Mm-hmm.

Gurthie Cooper: Right? So we really think about desirability, feasibility, and viability. And depending on the ideas and the hypotheses, hypotheses are very, very important to how we think, right? [00:38:00] So from scanning the horizon, we'll have a backlog of hypotheses that are outcome of that.

Gurthie Cooper: Um, we'll think about, right, so we'll look at these and we'll decide whether we wanna start with desirability or feasibility or viability in order to get to a stage gate.

Demetrios: Mm-hmm.

Gurthie Cooper: Right? And in some cases it'll be we'll wanna look at feasibility first. So we'll work with a partner, and we would think about what's the leanest thing we can do in terms of a tech POC, just to see what we need to do, uh, in order to validate the hypothesis of can we integrate.

Nidhi Sharma: Hmm.

Gurthie Cooper: Right? So can we get the data that we want um, into our systems and then decide what to do with that next. So that would be our first, first stage gate. That'd be an example of a tech POC. In other cases, we might want to validate desirability on the consumer side. So do customers even want this? Might they want it in the future?

Gurthie Cooper: And there's a few ways we go about that. The, the good old painted door still works, right? So you put something in front of customers that, um, [00:39:00] could just be a question, are you interested in this thing? You can create like a, a button that just reveals a sign up for more information, we're working on this, something like that.

Gurthie Cooper: And of course, like user research is super important as well, so just to get that really to understand like what the appetite is. So we work in really tight stage gates, and we try to validate and invalidate at the shortest possible cycles possible. So we're talking like one week turnarounds, right?

Gurthie Cooper: During that process... So that's the classic way that many startups work. We're in a huge global company, so we need to operate a little bit differently. We operate in that way, but it's very important for us to leverage our, um, maybe the resources that we have, which is we have scale, we have market teams, we have very strong commercial teams, we have stakeholders, um, we have deep, you know, design resources, data, et cetera, et cetera.

Gurthie Cooper: So along that journey, we're taking everybody along for the journey, making sure that we have strong alignment on the storyline, why we're doing what, and we need to drive alignment from [00:40:00] the beginning. The ultimate goal of any of these, um, tests or hypotheses or, or POCs is to get through the stage gates that are unique for each of the ideas, and then ultimately to what we call graduated, right?

Gurthie Cooper: And that means effectively moving it from incubation into scaling, early scaling, or maybe an MVP. And at that point, usually there's a lot of, um alignments and collaboration with other parts of the organization. You know, we're a 25-year-old company. There is a lot of complexity under the hood. So once we've validated the right things in the incubator, we need to be really, really smart about alignment.

Gurthie Cooper: Um, and, and that goes for tech as well. Very, very important to have alignment on what the goals are, what the expectations are. As soon as we have that initial signal, we don't just, um, you know, announce it, you know, three months later or two months later, "Surprise, this is your product." We take the guys along for the ride with us.

Gurthie Cooper: Uh, so there's an art to that. E-effectively, it's, uh, corporate innovation.

Nidhi Sharma: Yeah. [00:41:00]

Gurthie Cooper: And I've, I've done a lot of startup work, and th-this is the thing about-- this is the big difference about innovating in a big company like this. There are so many powerful levers that you can leverage. You need to look at them as opportunities.

Gurthie Cooper: You know, some people might look at them as, as hindrances or, or things that slow you down. You can use them to speed you up.

Nidhi Sharma: Even though this process looks like it's like a, you know, sequential process, and we have different, uh, steps to do it, it's all happens in parallel. Mm-hmm. So all the time, we have many ideas coming from all over the places.

Nidhi Sharma: We have a research team. They come up with the innovation ideas. We have engineering people who have s- all-- some-- most of the time, they have different type of problems, and they come up with the ideas. And then our organizational strategy is the key. So we balance these two together, and we just, you know, look at the best possible bets.

Demetrios: Yeah.

Nidhi Sharma: And then most of the time, [00:42:00] we are playing with uncertainty, and we are traversing with ambiguity. And we have to ensure that we just, you know, have an engineering capacity in place to ensure that we are at least trying all the ideas, even though we fail. Mm-hmm. So failing is a learning, and failing is an experience.

Nidhi Sharma: So we don't wait for everything to become the most perfect, uh, you know, thing. We just, we just-- We have two types of, uh, you know, innovation and ideas most of the time. One is where the technology is evo-evolving. Every day there's a new protocol, every day there's a new tool and new technologies to integrate and to see out of the possibility.

Nidhi Sharma: So we really want our engineers to jump on it and see what we can do, what s- what type of outcomes we can achieve for our business. But at the same time, what Gatrie has explained, we do follow that process in parallel as well. We ensure that, you know, our ideas, even though they're failing, [00:43:00] at least one of them is becoming a, a impactful version within the organization.

Nidhi Sharma: And we have so many examples like that. Last year has been really successful, uh, where our team was able to, you know, fail with some, but was also able to succeed with, uh, many ideas. And, um, that's how this whole process worked. So it, it's like Starting with the real problem and the real challenge. Because idea can be vague.

Demetrios: Yeah.

Nidhi Sharma: You know, uh, there should be some practicality, some tangible action, uh, some real problem to be solved. And that's, I think, our intake process. That's where our previous, uh, our past experience as well as our learning and as well as where the market is evolving, blending all these things together, we decide like which one is the best one to pick and to move.

Demetrios: Yeah, really defining the problem, having

Nidhi Sharma: it very clear. Yeah, defining the problem is the key.

Demetrios: Because yeah, it can be just a gut feeling like, "Oh, I think it's this," but [00:44:00] then you shape it and you're creating that-

Nidhi Sharma: Yeah ... sort of- Because we can't just run after all the shiny things in the market, right? Mm-hmm.

Nidhi Sharma: And, uh, AI is also not a solution for every problem.

Gurthie Cooper: Mm-hmm.

Nidhi Sharma: So defining the problem is important. That's what I think we focus on. So once you define the problem, how you fix the problem is completely different thing. You can easily, uh, you know, sometime you can use AI, and most of the time it just, it's, it's a simple fix.

Nidhi Sharma: Uh, but being incubator, we are not there to fix the near term problems because those problems are already part of our strategic map, and other product teams are already working towards that. It is in their own map. We are just looking ahead at least one or the two years down the line and- Yeah, different view

Nidhi Sharma: and figuring out the moonshots, uh- Mm-hmm ... and the kind of evolution we'll see in future.

Demetrios: Yeah, that's cool. You get to kind of play out all these different scenarios, potential scenarios of where [00:45:00] things are going and how to- Yeah ... strategically position yourself in that.

Nidhi Sharma: Yeah. A, a lot of hypothesis.

Demetrios: Yeah.

Nidhi Sharma: A lot of uncertainty and ambiguity, like, you know, playing with- Yeah

Nidhi Sharma: different things that-

Demetrios: And sometimes I imagine, like you were saying, like, yeah, we can be directionally correct on this, but the market isn't ready for it yet. Yeah. Because of some reason or other.

Gurthie Cooper: Yeah, exactly. So I, I've got an example actually of what, what... to build a little bit of color on what, what Nidhi's saying.

Gurthie Cooper: So we, one of the hypotheses that we had, uh, last year was, so in, in the traditional app, search is one of the most important interfaces, right? Yeah. Um, there's a few different ways that people search. One of the ways that people search sometimes is, is conversationally, but at a certain point last year, we didn't support conversational search.

Gurthie Cooper: So we, the hypothesis was that, you know, a certain subset of customers would like to, um, ask questions in search, much like, you know, they do in, in many other products, including- Yeah ... you know, all of their, [00:46:00] their AI agents these days. So we looked at that and we had to work with, um, you know, incubator had to work with the customer team, which is the team that owns, you know, the customer interfaces, the search team within that team.

Gurthie Cooper: Um, and it was really, you know, to my point earlier about building, uh, strong connections and a shared view on what the outcomes are gonna be, the potential to, the business potential and the product potential, and the customer delight potential of adding that feature- Mm-hmm ... was the story that we had to tell the whole way to really bring everybody on board.

Gurthie Cooper: What we found though, when we were looking at the data This is really interesting is that, you know, so conversational search you could argue is gonna add value in the short term, but it's probably also, you know, a mid to longer term bet, right?

Demetrios: Because the customers will expect it.

Gurthie Cooper: Yes, exactly. They might not expect it now, right?

Gurthie Cooper: So there's only a, a small amount of customers who are just naturally asking questions because we're not prompting them to do that.

Demetrios: I was gonna say like, how did you tell people that you now support that?

Gurthie Cooper: Exactly. [00:47:00] So it's, that's customer education, right? So it's actually in play right now. It should be launched in the UK in the next couple of weeks, is that real onboarding like, "Hey, here are some suggestions on conversational search terms, and by the way, um, you can type whatever you want in here."

Demetrios: Ask a follow-up question.

Gurthie Cooper: Exactly. Right. So, so it's that. Um, when we were looking at the data though, we found that there's a good amount of searches that end in zero results, right? Mm-hmm. It's a common problem across most, uh, consumer apps, right? So people just aren't finding what they want because of spelling mistakes or, you know-

Demetrios: Yeah, that's me

Gurthie Cooper: filling in the blank. I do that

Demetrios: stuff for shit. Exactly.

Gurthie Cooper: That's

Demetrios: so true.

Gurthie Cooper: So, so semantic search is one of the ways that the, you know, we're, we're tackling that. The conversational search h- um, managed... Well, it was actually almost like a side benefit of what, what our initial hypothesis was, is that conversational search solved, uh, a big percentage of those dead ends you could call it.

Gurthie Cooper: Mm-hmm. Right? So which ended up having a revenue impact in the [00:48:00] short term that was significant.

Demetrios: Wow.

Gurthie Cooper: Yeah. So by thinking about conversational search and really driving, uh, you know, um, the short-term innovation and trying to c- and, and getting that into the product, we managed to solve a real problem in the short term, much more related to, you know, the bigger OKRs, which is really around revenue and growth.

Demetrios: What are all the things that didn't make it through this great filter? What are things where you as the innovation lab thought, "Oh, of course, this is the way the future's gonna be. We should put this into play," and then you looked at it and you were like, "Actually, it's not gonna work"?

Nidhi Sharma: One of the example is sending notification based on your location.

Demetrios: Mm-hmm.

Nidhi Sharma: We called it location by notification. So we wanted to, um, notify customers for a collection order, uh, when they are near to a restaurant, let's say 10, 15 meters. And that, the idea [00:49:00] was that our partners will get some benefits because they are doing collection orders and sometime customer don't know that this particular restaurant had this, uh, proposition and, uh, other way helping the customers for customer convenience.

Nidhi Sharma: And we started thinking about, you know, how to fix the problem. But the kind of challenges we had was first the legal, uh, you know, constraints. Uh, you It's very difficult to track customers' live location all the time.

Demetrios: Mm-hmm.

Nidhi Sharma: And if we do that, there is some physical devices needs to be attached to, uh, you know, the stores or the partners.

Nidhi Sharma: And that means there is operational overhead. Sure. Because you need to maintain those physical devices which keeps on taking signal through your mobile device, and then we, our app catch those signal and send it to customer. And in the end, if you look at the value [00:50:00] As we discussed earlier, it could actually become a customer grievance instead of customer, uh, you know, delight because- So

Demetrios: all of a sudden you're looking at this and you're saying, "We have to put this huge amount of effort into putting in devices into all the merchants' shops-

Nidhi Sharma: Yeah

Demetrios: only to send more notifications, which the end user probably- May not like ... isn't going to like.

Nidhi Sharma: Yeah.

Demetrios: Yeah. So that, yeah, I can see how that one died.

Nidhi Sharma: Yeah. But then we also thought about, you know, experimenting, experimenting with the simplest thing. So we, we thought about not to put any physical device, not to even ask our customer to share their live location with us.

Nidhi Sharma: Just, uh, you know, have some reasoning model in the background which will keep learning from the customer behavior, like you order coffee every day at 9:00 AM, or you order lunch at 1:00 PM. So we just send some, uh, you know, blind notification, uh, at that point of time based on your order history and your behavior.

Gurthie Cooper: Mm-hmm.

Nidhi Sharma: Uh, and [00:51:00] we designed it, uh, and technically it was achievable. But then again, when we evaluated, you know, the impact versus value and, uh, as we mentioned, you know, how do we... What is our intake process? We have some parameters, and based on those parameters and all the conversation we had, we just dropped this idea.

Demetrios: Numbers didn't look good.

Nidhi Sharma: No.

Demetrios: You know what it really reminds me of? Your jobs are very much in line with, uh, who's that marketer that's famous in the UK? He, I think he worked at Ogilvy. He wrote a great book. I'm literally was-- I just finished it. Um, I think it's Roy Sutherland. Do you know him?

Gurthie Cooper: Yeah.

Demetrios: Rory.

Demetrios: I've

Nidhi Sharma: heard the name,

Demetrios: yeah. Rory Sutherland. Maybe. Sutherland, maybe. Sullivan. I don't... I'll figure it out. What's

Gurthie Cooper: the book, what's the book called?

Demetrios: Ah, I'll, I have to pull it up on my phone 'cause I, I was listening to it. And he gives all of these antidotes where you think through [00:52:00] something and you think, "Oh yeah, this is the logical thing."

Demetrios: But then when you put it out there, customers and consumers do the opposite depending on how it's framed. So the one that I think he had a Tech Talk or TED Talk on back in the day that I first saw, and then he adds in the book that really captured my imagination was, he said, "The UK government is proposing to spend billions of dollars to make a train, a high-speed train, that will take a trip from London to Manchester or London to Liverpool and make it 30 minutes faster."

Demetrios: So billions are getting sunk into this high-speed train. And he said, "What if we just made the current train more of a- delightful experience so that the customers don't care if they're on a three-hour train or a two-and-a-half-hour train ride

Nidhi Sharma: Yeah. No. Yeah, yeah, yeah. And that's the challenge, uh, being an incubator, [00:53:00] uh, like space, because it's really hard to find that balance between, you know, choosing a perfect idea versus something which is-- which may not work or which is not logical at all.

Nidhi Sharma: This is the real challenge. We sometime get emotional with what we are doing because we are actually putting our energy and our focus into it, and it's really hard to be rational about it and make a decision that this is the right time to stop or pivot or keep going.

Gurthie Cooper: Yeah.

Nidhi Sharma: Uh, we do have, uh, those type of frameworks and, you know, kind of, uh, ceremonies where we challenge each other, we make each other accountable, but it's still like, you know, very difficult that when you are spending a lot of your time into a product or an idea to execute it and see the art of possibility, and you've been asked to stop it- Yeah

Nidhi Sharma: or pivot to a different direction. This is a challenge, I would say. And there's a technical term for that. And it's really hard [00:54:00] to nail it down.

Demetrios: What is it called? Like a-- Is it- Sunk,

Gurthie Cooper: sunk

Demetrios: cost. Sunk cost. Yeah, exactly. And j- you, you feel ownership, and you don't wanna kill your darlings, but you have to have that, like, ruthlessness to you in a way.

Nidhi Sharma: Or park it for some t- even parking for some time is, is also challenging. It's, it's difficult to make that decision, "Okay, let's pause it. Let's do something else."

Gurthie Cooper: Yeah. Well, we find ourselves in that position quite often, actually, right? So sometimes we're just too early, so we'll put something on ice. Um, we did that with our integrations with Mercedes-Benz.

Gurthie Cooper: You know, so we have an in-car ordering app with Mercedes-Benz right now. Um, we worked on that, um, for, you know, probably a year, uh, maybe six months before, um, really landing the contracts and the commercials, et cetera, et cetera. Working with a company that is not natively a tech company, something we've learned is- Oh.

Gurthie Cooper: It's, uh, it's certainly interesting. Yeah. Because, you know, the difference between working with [00:55:00] a, a Meta or a Google versus, you know, a, a automotive OEM, night and day. But to go back to our vision, if we wanna be ubiquitous, this is the playing field we need to operate in, right? So, so in some cases, you need to put something on ice for three months, six months, which in our world is like a lifetime, but in, uh, another world might be, you know, a very short amount of time.

Gurthie Cooper: So it's interesting to balance those two things.

Nidhi Sharma: And for us, um, not just experimentation is, uh, you know, something which we are, we are c- we are focused on, but also Measuring our success is experimentish-- is, is experimental as well. So when- Yeah ... it's very hard to measure your success based on, uh, an incubated idea.

Nidhi Sharma: Uh, it's like, you know, you put something out there like this, uh, NLS. We can't market it when it is not fully ready- Yeah ... to be, to be shared and sh- to be [00:56:00] talked about. But at the same time, we want to learn from it, and we want to improve it. So our success is also like, you know, quite i- quite in, quite iterative and experimental.

Nidhi Sharma: Um, so we have, uh, found out a way to, to, you know, measure our success with different parameters. So we just have-- We just say, when we define the problem, just think about what's the outcome or what's the clear success we are looking here. Hypothesis is one thing, but what will be the real parameters and measuring points?

Nidhi Sharma: Uh, sometime it could be as simple as just customer engagement.

Gurthie Cooper: Mm-hmm.

Nidhi Sharma: Uh, most of the time it could be completely like, you know, end goal, which is order conversion. Yeah. And a few time it is just the exploration or setting up the foundation so that we can enable something else in future. That also happens.

Nidhi Sharma: And in agentic AI world, before we even start building new agents, we wanted to think about how the data will be [00:57:00] perceived, because we need structured data, uh, as a prerequisite to build the potential agents. And that's the foundational work, and that's qui-quite, quite a tech, you know, experiment rather than a clear business objective or a business goal.

Nidhi Sharma: So finding out what kind of success we are looking for is the key thing when we define the problem.

Demetrios: Mm-hmm.

Nidhi Sharma: And then making sure that we have our internal, you know, testing people or the group of people who can test, uh, and genuinely give us the feedback so that we can improve, iterate, and have that feedback loop in place.

Nidhi Sharma: And once it goes to market, then also we go with, you know, the, the lowest risk possible. We just roll it out to 5% people, to one country, and sometime to even one region, so that customer can share the feedback. And then we just go into that feedback loop. Mm-hmm. And once we feel like, yes, it is now ready to be talked about and to market, then we launch it fully.

Nidhi Sharma: [00:58:00] And voice agent has gone through this phase, uh, for a longer period of time. The product which we see today wasn't like this. It was-- It has gone through that iterative cycle of improvement before we have, uh, put it across the consumers.

Demetrios: What were some of the hard challenges creating voice

Nidhi Sharma: agent? I think the first challenge was the idea itself, because when we started, uh, it was not in something which others, uh, you know, competitors or other, uh, vendors were using.

Nidhi Sharma: So finding out the right technology, uh, because there are so many nuances when it comes to voice. We need to make sure that latency is accurate, uh, the quality is there, uh, the conversation is intent-based. Because we were not aiming for a voice mode. We were aiming for a voice companion- Mm-hmm ... a personal voice agent to really understand you and give you an [00:59:00] option of being, you know, switching between the visuals and the audio.

Nidhi Sharma: And that's where the real challenge was to actually technically deciding which is the right path to go for it. And then it has gone through a lot of back and forth. There had been challenges between the UX and the audio quality, because it has both the versions, right? So you can see the cards, you can swap across, and you can choose the right, uh, dish for you, but at the same time, you're on the audio.

Nidhi Sharma: So you have to make sure that the customer experience is balanced and it's, it's basically to the point for both audio as well as your visuals, and that's where the UX part plus the core engineering, the server-based or the backend-based, uh, engineering have been through some challenges, um, where- Yeah, I

Demetrios: imagine you can't take for granted that somebody's gonna be looking at their phone if they're on voice mode.

Nidhi Sharma: Yeah.

Demetrios: Yeah. You have to have the experience be flawless only on voice, and then you sprinkle on some delight [01:00:00] with the visuals. With

Nidhi Sharma: the visuals, and yet the... Then you can go back to voice.

Demetrios: Mm-hmm.

Nidhi Sharma: And you can switch between, uh, so which means you're not just building a voice agent, you're also building an intuitive experience for your consumers.

Demetrios: Yeah. How did you deal with latency constraints? How did you deal with all these challenges?

Nidhi Sharma: So we are using, uh, OpenAI real-time API And, uh, with that integration we have, we have server-based approach. So we just, as soon as you start the conversation, even before you start the conversation, we load the whole context in the background.

Gurthie Cooper: Mm-hmm.

Nidhi Sharma: So that you have the smoother experience and the latency is being reduced at that point only because when you, when in the, when in the background you have the context, so by that time agent actually knows most of the things about you.

Demetrios: And when you say context, I imagine there, that is a very difficult thing.

Demetrios: Like [01:01:00] context engineering is huge, right? And-

Nidhi Sharma: Yeah ... it's

Demetrios: huge because it is difficult. So how do you know what to weigh? Are you just giving them the last 10 orders? Are you just... Like how do you explain that prompt so that it is contextually re- relevant?

Nidhi Sharma: So as of now, it is quite simple- Oh ... and standard. So it is based on the order history.

Demetrios: Yeah.

Nidhi Sharma: And some level of personalization for your dietary requirements. Mm. So context is, the scope of the agent is very simple and concise so that it do not hallucinate and it, you know, sometime prompt engineering and context engineering is being heavily loaded for one engineer, uh, sorry, one agent, and that becomes actually cumbersome for the agent to respond appropriately.

Nidhi Sharma: So we wanted to make it a lightweight, easy to communicate and, uh, easy to e- easy to be consumed. That information which the agent is giving you should be, you know- Very much tailored for you, but at the same [01:02:00] time, not every possible information you have shared with us is going to be in the context. Yeah.

Nidhi Sharma: So it's very simple context, and it's event-based, so there are multiple events in parallel. So if you're asking a complex question like I asked that I want to arrange a party for these many people. So based on my, my query, it just, it just surfaced multiple events in parallel so that there's a speed, there's accuracy, and the, even the UI structure is being ready in the background even before you see it as a visual.

Gurthie Cooper: Mm.

Nidhi Sharma: So a lot of processing is happening in the background for voice agent, and that's why once you interact with it, it's quite fast, it quite accurate, and it gives you that personalized recommendation because it will not just tell you that, "Try this pizza." It will also tell you, "Because you're a vegetarian, you might like this vegetarian pizza."

Nidhi Sharma: Mm-hmm. And you can start the conversation as a simple thing like, you know, "I'm looking for something spicy." Or you can also give a [01:03:00] complex query like, "I'm looking for something spicy which is vegetarian, but I am hungry, so I can't wait more than 30 minutes."

Gurthie Cooper: Yeah.

Nidhi Sharma: And all this processing is server-based processing in the background, which is quite fast, and a lot of information is cached as well because- Yeah

Nidhi Sharma: uh, your conversation is cached, the whole conversation, and that also gives the speed and accuracy.

Demetrios: It, it's interesting you talk about how you almost preload things just in case, because that is a very common design pattern I've been seeing a few folks use, especially when it comes to voice, 'cause you'd almost rather have it and not need it than need it and not have it and then make the user wait.

Demetrios: Mm. Since it is so high stakes, the user, if it has to wait 30 seconds, it might think something's wrong, especially if you're not getting any voice cues that, "Oh, sorry, I'm still searching. Oh, wait, I've got something. Let me check." You could do a bunch of fun stuff with the voice cues, but 30 seconds is still for 30 seconds.

Demetrios: And so [01:04:00] if you can shave some time off of that, because you already have the latency of the model, which is gonna be just-

Nidhi Sharma: Yeah, the two kind of voice-

Demetrios: Yeah, and- ... uh, that

Nidhi Sharma: is given ...

Demetrios: so if you can have all of the data and you can have all of the stuff that it needs there, and then you decide whether or not you show it to the user, or you decide whether or not it's relevant.

Demetrios: Um, I've seen that, and then I've also seen constellation of models, not just one. It's not just hit one model and then you come back and you've got it. But they'll have many different agents hanging out, and they're ready in case there is a moment they need to go and grab something, and they're just o- observers of the conversation, and they know Oh, if I hear that keyword, I should go do this.

Demetrios: Or if I-- But they're not jumping in. They're not the agent that the user is interacting with

Nidhi Sharma: So the similar concept here. So when I say parallel, uh, processing-

Demetrios: Mm-hmm ...

Nidhi Sharma: [01:05:00] so we have tools in this case. So while you're asking for a pizza, so there will be three parallel tools being called, uh, one to load the menu, another to load the partner information.

Nidhi Sharma: And these are cached.

Gurthie Cooper: Mm-hmm. So

Nidhi Sharma: because partner information is something which we are not going to change, we can easily cache it. And that also allows some level of, you know, faster- Speed ... speed and ability to, to use it in the most intuitive way. And this processing makes it unique and better.

Gurthie Cooper: Yeah.

Nidhi Sharma: Um, and also, the biggest learning has been fine-tuning it.

Nidhi Sharma: So starting with very basic structure, which is just solving the basic purpose, which is just to understand the intent of the customer and giving the recommendation. But then fine-tuning the prompt after every iteration and based on the conversations we have, um, that prompts, [01:06:00] system prompting has, had been a key here, because we wrote a lot of integration tests, uh, with so many hypothetical scenarios, uh, based on different kind of conversation customer can have.

Nidhi Sharma: And based on those integration tests, we fine-tune our prompting system.

Gurthie Cooper: Our text assistance has been in market for longer than our voice assistant, right? Voice is brand new. And there are some subtle ways, um, that those two products differ. So there's a threshold on the text assistant where if you speak to it, I believe it's if you have five conversations within the space of two weeks, it will retain more memory than if, if you have not hit that threshold, right?

Gurthie Cooper: And how that plays out for the consumer is you come back after that sixth conversation, and it will actually tell you a little bit about your preferences in text form. Mm-hmm. And it'll make a, a subtle recommendation on maybe what you want to try next. So that's something we haven't taken into our voice product yet, but you can imagine that would be a great conversation starter, right?

Gurthie Cooper: [01:07:00] Yeah.

Nidhi Sharma: Yeah. Yeah, and, and text a- uh, text assistant has been actually, uh, very much useful for voice agent, because those conversation helped us to shape up, you know, the-

Demetrios: Mm-hmm ...

Nidhi Sharma: integration testing and prompting for the voice.

Demetrios: Yeah, what people are asking, what the happy paths are, how quickly- Yeah ... they need things.

Demetrios: That, that makes sense. And also, I can see, going back to I worked out today The voice assistant knows that when I open it, it says, "Hey, how was the workout? Do you want me to order some-"

Gurthie Cooper: Yeah "...

Demetrios: protein?"

Gurthie Cooper: Yeah, exactly. Exactly. Yeah. Yeah. So sometimes we start with, you know, a very limited, uh, feature in a very limited, uh, region.

Gurthie Cooper: Other times we just put a single robot on the road.

Demetrios: Hmm.

Gurthie Cooper: Segue to drones and robotics. Yeah. If you're ready for that.

Demetrios: Yeah, yeah.

Nidhi Sharma: When we think about AI, we have a virtual or a software-based AI solution and a physical AI solution. Mm-hmm. And for physical AI solution, we have drone delivery and robotics in [01:08:00] place.

Nidhi Sharma: For the virtual or the software solution, uh, we need to think about our consumers, our partners, and logistics. And that's where, you know, the solution which we are building currently, they're primarily focused on consumer, but in parallel, we have automation and agentic AI going on to serve our, uh, logistic, uh, teams and to serve our logistic partners as well.

Gurthie Cooper: Mm-hmm.

Nidhi Sharma: And we have sales agents in place, uh, to onboard our partners, uh, in the most, uh, automatic way. Um, now When it comes to the physical AI, we have already launched our first, uh, partner in Switzerland. And as we talked about, you know, how in incubation we start in the most experimental way, we go very lean.

Nidhi Sharma: So I think Atri can a- add more, but we started with the one robot.

Demetrios: Just

Nidhi Sharma: one? Uh, just one robot. Uh-huh. Uh, just to see that of the [01:09:00] possibility, just to see how it gets that end-to-end delivery done. Uh, because last mile delivery is something we want to achieve, but at this point of time, our, our robots are actually doing the end-to-end delivery.

Gurthie Cooper: It was the leanest possible robot pilot you can imagine, uh, in that there was a human walking next to... And it was a robot dog, by the way.

Demetrios: Oh.

Gurthie Cooper: Right? So the partner there is, uh, River, and, uh, their form factor is like your classic four-legged, uh, quad robot dog. It has wheels as well.

Demetrios: Mm-hmm.

Gurthie Cooper: So they're unique in that way.

Gurthie Cooper: So it, it can ride and it can climb up stairs, et cetera. For the launch, we had a human walking next to the robot, um, just to make sure that no, you know, there were no issues. And it's an interesting thing, right? Um, public, uh, int- im- impression of a robot walking through the streets

Demetrios: turns some heads.

Gurthie Cooper: Indeed, indeed.

Demetrios: So- I've seen some videos of people attacking the [01:10:00] autonomous vehicles that are coming through and delivering things.

Gurthie Cooper: Yeah.

Demetrios: Or they try-- They attack it just because, uh- It's fascinating ... they're taking-- Yeah, and they're taking out their aggression on it, or it is- Yeah ... it's like, "Hey, let's see if it can handle a little adversity."

Gurthie Cooper: Yeah.

Demetrios: Or they attack it because they try and get the food that's inside. Yeah. It's like, "Hey, there's some free food here, and let's get that."

Gurthie Cooper: Exactly. Th- th- there's a lot to say about, right, um, job replacement and those kinds of things. Our, our vision with this is really that we're not gonna be replacing our human couriers, we're augmenting them.

Gurthie Cooper: We're making them safer. Mm-hmm. So the vision is to have an orchestrated last mile delivery system orchestrated between humans and robots with seamless handoff, right? Mm-hmm. Depending on things like weather conditions, traffic. Um, in Ife they have this interesting position where- a situation where they have a huge river, right?

Gurthie Cooper: And it takes... Although it's quite close to get to the other side of the river as, as the crow flies, um, it takes an hour [01:11:00] to drive because of the traffic. So that's ha- that's, that's their, um, use case for the drone, right? So the drone takes the delivery across and then hands it to a human.

Nidhi Sharma: Mm.

Gurthie Cooper: Yeah. So it, it's all about orchestration and augmentation, and then obviously expanding our reach.

Gurthie Cooper: That's where the growth piece comes in for the business. There's two parts to that. First is expanding our reach, so you can deliver to rural areas or areas that would take much longer for, you know, a driver to get to. Um, and then the other one is that the, it reduces the cost of a delivery significantly.

Gurthie Cooper: So that's, you know, as many mentioned, we're a three-sided marketplace, consumers, partners, meaning restaurants and grocery stores, and then our logistics arm as well. So innovation happens across those three user types.

Demetrios: Mm-hmm.

Gurthie Cooper: And the most promising thing we're doing right now in, in the third one is drones and robotics.

Gurthie Cooper: We're delivering via drone in Ireland as well. We've partnered with a, a startup called Manna And people love it.

Demetrios: Yeah.

Gurthie Cooper: It's really interesting to see, right? 'Cause we, [01:12:00] again, back to how we work, so very iteratively, very hypothesis-driven. Um, and we really kind of soft launched into it, and we wanted to understand customers' perception in the app.

Gurthie Cooper: We gave them the choice, "Do you wanna deliver this via drone?" It's a new thing. It's a novel thing, right? And that was what we, some of the metrics we were looking at as well. It was like, is this just gonna be an initial burst of novelty and then it'll tail off?

Demetrios: Yeah.

Gurthie Cooper: But it hasn't.

Demetrios: No.

Gurthie Cooper: No. So we- Is

Demetrios: it because it's faster, people would want it?

Gurthie Cooper: It's faster.

Demetrios: Yeah. It's

Gurthie Cooper: faster.

Demetrios: Well, 'cause if you can fly, it's gonna be faster, right?

Gurthie Cooper: Exactly. And I think that the novelty thing is part of it as well. It's just fun to get your order via drone, right? Mm-hmm. So we, we see that a huge percentage of our Burger King orders in our, in the town that we've launched in, Blanchardstown in Ireland, are delivered via drone.

Demetrios: But it's not like I open my window on the fifth floor and the drone is hovering outside of my window. Not yet. It's like-

Nidhi Sharma: No, there are compliances. Uh, we, we have to follow those [01:13:00] compliances, so we have to maintain that distance from, you know, the- The buildings ... the buildings. Sure. And we have to make it safe.

Nidhi Sharma: But I think the biggest advantage is to handle situations like pandemic-

Gurthie Cooper: Sure ...

Nidhi Sharma: or natural disasters, and reaching where human cannot reach. Machines can do what human cannot do when it comes to delivering, uh, last mile. And that is another aim that we still, you know, manage the customer convenience even when there is, the weather conditions are not very promising or when the things are not as natural and normal as they are and there are some, you know, casual calamities and all.

Demetrios: Hmm. Things get delivered sooner. They can get delivered further away. Me living in a rural area, I'm very happy to hear that because-

Gurthie Cooper: Yeah ...

Demetrios: I don't have the ability to use a lot of these food delivery services.

Gurthie Cooper: Yeah.

Demetrios: Since there's just not restaurants that are... I'm very picky about my [01:14:00] food, and also, uh, I live near a city, but it's a 40-minute drive.

Demetrios: Yeah. So if I could get the food from that city delivered via drone, that would be amazing.

Nidhi Sharma: Yeah. Yeah. And there's an operational efficiency as well using drones and robotics for last mile delivery.

Demetrios: How so?

Nidhi Sharma: Uh, because of the speed- Yeah ... uh, because of the amount of, uh, you know, uh, deliveries you can do, uh, parallelly.

Nidhi Sharma: Uh-huh. Uh, so this is, this is actually going to save cost- Yeah ... as well as going to provide the better efficiencies.

Demetrios: Yeah, you can scale up-

Nidhi Sharma: Yeah ...

Demetrios: a lot.

Nidhi Sharma: But by no mean it is going to replace human or our couriers. We are still going to maintain that. It is just an-- As Kathir mentioned, that it is just an augmentation in our existing logistics.

Gurthie Cooper: Hmm. Yeah, so concretely, y- you can imagine that, um, a courier is stuck in traffic jam, unexpected traffic jam, right? So he sees his ETA is gonna be an [01:15:00] hour now. This is not a great ex- experience for the customer, obviously. No. So what we wanna do is have, uh, the ability for him or her to say, "Let me hand this off to a robot or a drone to finish the delivery," or to pick it up from the driver and hand it to another driver on the other side of the traffic jam, for example.

Gurthie Cooper: Yeah. Hmm.

Demetrios: Who gets the tip?

Gurthie Cooper: Uh, the robot, of course. Yeah. Right? And, and Petcoin. I'm kidding, I'm kidding, I'm kidding.

Nidhi Sharma: There's one point we haven't touched upon is about what does AI innovation means to Jet in general. Uh, because the, the innovation is not limited to just one team. It is actually across the whole organization.

Nidhi Sharma: Mm. Uh, we are AI first, uh, organization. So we have our AI champions, the group of people who proactively work on the, uh, you know, the tools in the market, play around with them, and then once we f- see the value, we just [01:16:00] allow our engineers to use those tools. So we can see the efficiency and the velocity is increasing, even though we don't have the real data, the concrete data, which can actually, uh, sh- uh, show the real difference.

Nidhi Sharma: But we know that, you know, whatever you were doing in five days, now with these tools, you can do in one day or two. So there is definitely a scope and opportunity to improve. And we are already kind of ahead or with pace in the market. So we have AI in engineering innovation to use the right tools and the right technologies- Mm

Nidhi Sharma: beyond just the inno- incubation team.

Gurthie Cooper: And the same is happening in products. So the product managers in my team are, are using AI to work faster in the right areas, right? So the craft of product is still super important.

Demetrios: Yeah.

Gurthie Cooper: And I think it's important to do deep thinking. Um, but we now have effectively like copilots to help with that, to speed up the right parts of the process.

Gurthie Cooper: Are they

Demetrios: submitting PRs and shipping code yet?

Gurthie Cooper: Um, oh, oh, yeah. [01:17:00] They do. I mean, my... Yeah. Do from some prototype. Our PMs, we do, yeah.

Nidhi Sharma: Yeah.

Gurthie Cooper: Lucky enough to have PMs that are, are technical, right? Mm-hmm. So they do a lot of vibe coding and increasingly, um, it, it's, it's making things way more efficient. We're also using evals.

Gurthie Cooper: It's so, so important to understand what's actually going on in your conversations. And the PMs are setting those frameworks up, right?

Demetrios: Yeah.

Gurthie Cooper: Um, something else I'm really interested in is, is scaling usability testing using AI, right?

Demetrios: Oh,

Gurthie Cooper: interesting. So I, I can't remember what it's, what the term is, what the industry term is, but basically there's some startups out there that can spin up 1,000 different personas for you-

Demetrios: Mm-hmm

Gurthie Cooper: and provide real feedback on your product so you don't have to, you know, get actual humans. Yeah. Super important to have humans in the process- Human validation ...

Demetrios: but- Yeah ...

Gurthie Cooper: it's a great augmentation, yeah.

Demetrios: Yeah, actually I've seen, um, some project that will help you do that just to check your current app and interface, all the links go to the right places and- Yeah.

Demetrios: 'Cause otherwise you're sitting there and you're clicking [01:18:00] on the buttons and you're making sure that everything works like you want it to work.

Gurthie Cooper: Yeah.

Demetrios: But why can't you just simulate that?

Gurthie Cooper: Yeah, indeed. But you can also get feedback, like behavior type feedback. Like, is this a good experience?

Demetrios: Mm.

Gurthie Cooper: Where'd you get st- where'd you get stuck, right?

Gurthie Cooper: The things that, uh, historically you had to sit with the user and spend the time.

Demetrios: Yeah.

Gurthie Cooper: Um, you know, you can scale that using AI. I think that's really, really interesting.

Demetrios: Wow.

Demetrios: Puedos.

Gurthie Cooper: Puedos, puedo, puedo, puedo