Powering India’s largest Voice-Driven Car Retail Operation

Cars 24 is like India's largest used car platform. Um you know we do a lot of buying of cars. You know we transact thousands of cars and you know helping people sell their cars in India. Tens of thousands in fact we also send a huge amount of cars that we sell at our 300 plus retail hubs across India and not just India we also operating in the in the UAE as well as in Australia. So it's like a wide scale of you know [snorts] consumers that we are serving right and at the same time we also help them with financing and just helping them you know make the used car journey good u in terms of why voice matters is 70% of our consumers who are buying their cars are like first-time consumers right it's it's [music] a big decision for them and um you know our motto is like better drives better lives we have like countless emotional stories about what it means for a person to buy a car as well as sell a car there's been numerous you know people who come back to us say you know can you just give us our car back it's just like a part of our family u majority of our sales and conversions happen over someone from our team guiding someone over calls right we do tens of millions of conversation a month right whether it be from helping people discover their perfect car whether it be sales and negotiation on helping people find the right price maybe explaining them the terms of their loans right it's every touch point of our journey is built on conversations voice is what is built our entire business right and it's been a crucial crux of it and it's I would say as important to us as our tech and product. So it's just driving our entire business. I think roughly 30% of our permanent workforce is engaged in these direct consumer conversations, right? Multiple stakeholders but direct conser conversations. But [music] um what we have seen is you know if you're doing so much transactions there's a lot of fatigue that happens. There's a lot of regional nuances that are coming into the play. You're not just serving a local region, you're serving different villages, you're serving different regions of India, each with its own, you know, nuances. So the problem happens is when you're small, you're a startup, it it works out great. But when you scale up to multiple countries, multiple regions, [music] it just becomes very difficult for us to manage everything, right? It just becomes very difficult for us to ensure that, you know, every person that's involved in the process is giving a quality experience to our consumer, right? And how do we ensure that we're actually listening as well? I think that was very important for us. How are we actually listening to our consumers? They're talking to us so much. Are we actually listening? So these are the two challenges that we face. Ensuring that our conversational quality is always up to the mark that every conversation is polite, [music] it's insightful, it's helpful, it's not too draggy. And second is whatever our consumer is talking to us about, it's something that drives us like we're consumer obsessed or customer obsessed. So these were the two challenges that we have faced is how do we actually solve these two problems at scale 70% first-time car buyers they come into the store you know they meet some of our car advisers like we call them on the sales on force that [music] help people make a decision help them do test drives we do like tens of thousands of test drive a month across different cars and the biggest challenge is we don't have uni unique cars every car even if it's the same make model it's different because it's a used car right every car has its you know, maintenance, every car has its different damages, nuances, everything. So, we said, you know, we have 300 plus retail hubs that, you know, people come in to have a conversation with, you know, do test drives that last like 60 to 90 minutes. And the first one, what we wanted to solve is we want to generate insights. We just want to listen to our consumers. And the the challenge became how? So, the one thing that we did notice is all of our car advisers have tablets in their hands. So that was the first unlock that hey we don't need to deploy mics and expensive setups at each and every hub they already have it let's just turn on the mics and record whenever they start a test drive we have everything and we just started recording that conversation now can the second biggest challenge is when we started streaming it a the conversations were 90 minutes long second is these are open mic conversations so it's not just picking up a conversation because in a test drive the people are moving around they're going around the car there's a lot of distance there's a lot of noise there's people talking in the background Sometimes there's like a radio sound as well. Then at on top of that there's uh you know accent nuances. People can switch languages quite often. It was just like a mess. Like even if I gave it to a human the biggest problem is I cannot listen to you know tens of thousands of hour conversation a day and with so much mice. So we looked at 11 lab scribe um you know we explored a lot of these speechtoext um you know providers out there because we knew the answer is going to be AI. While a lot of the providers worked some some way or the other but nothing worked blanket right we had we would have had to customize everything but with the release of scribe what we did is we started u you know transcribing all these conversations and we noticed two things a it was super fast like processing a 90minute conversation um and these conversations are constantly coming to us it was incredibly fast right and it just kind of [music] gave us these insights very quickly the second thing was uh it used to capture you know transcription of a consumer identifying which consumer is speaking even with a lot of noise and it was beautiful and and one of [clears throat] the best thing worked so well that it was actually even capturing you know conversations in the background and labeling them as background noise [music] so and when we passed all of these transcriptions to an LLM we actually [music] it actually found it more useful right it stopped hallucinating less um the error rate was very significantly low compared to even [music] such a noisy environment so all of that was how we kind of solve it so as soon as we added transcription uh you know we asked business hey you have all of this transcribe what do you need an AI or an LLM to kind of generate insights out of it and they went crazy with it they started generating like hundreds of different touch points out of it that was the you know first biggest part that we did the second biggest part is like I said um our sales happens like 60% of sales happens to some sort of assisted person you know [music] getting there problem was stat was very long you know people used to get tired it because of incentive incentives. A lot of consumers wanted to do a zero sum game. There's like a lot of moving players that happened there. So we said you know what doesn't complain what works really good what is like is going to follow your instructions to the tea. It's just an AI voice agent. So that is where we started leveraging um you know AI voice agents from 11 Labs. We found that the realism in the voice and the voice modulation was something that our consumers received very well. Uh and of course you know we started deploying them very quickly. Um and what we also found is that um you know since we do so many different types of calling like 50 plus calling teams different doing different types of callings we figured out that you know just one or two engineers are not going to be able to do it blanket across the org. So we when we started rolling out the agents builder platform to our business folks right it was intuitive enough that they were also able to build calling agents themselves. So that just helped us also u you know deploy these agents across the board. So we went wide went quick everyone was able to experiment and we actually saw really good you know um numbers. So what we did is when we deployed um you know AI agents in those what we actually saw is that our lead qualification percentage improved by 10 20%. [music] Um you know our consumers average duration of conversation increased as well that means people were actually engaging quite a lot. So I think these were the two bits that we saw right that we saw some cost reduction and you know operational efficiency out of it but at the same time we also saw our business grow it as well and we saw that hey this is just very natural and you know of course we are trying to kind of replicate it for us what matters is quality right like I said we're serving so many regions so many accents so many languages in the subcontinent and even outside India right you're doing so much when We looked at other providers that were providing voice whether it be for the speech to text part or TTS part. What we saw is they excelled at some subset of it. So what you'll have to do is you'll have to build 10 different solutions work with 10 different folks to kind of build one solution that kind of works. What we saw with 11 Labs is that um 11 Labs used to give us a product that worked across the board. Right? So that was one. Second is when we are choosing a partner to work with our majority of it is can they keep up with our velocity because we so much investing in growth we you know a hyperfocused team that just wants to move move hyper experiment run 10 different experiments in parallel. If our partner has to kind of you know go through a red tape every time to help us to kind of [music] uh you know has to take 10 different decisions to just help us out. It's not something that's going to work for us. Even though if they can have the best product in the world, it's slowing us down. So that's one of the problems that we first see it is can they keep up with our pace. And what we saw with 11 Labs is that it was very proactive like working was very collaborative and it was not that hey we are the client we say and they do. But it's like what we found is we co-built quite a lot that even if we were not kind of reaching out they were saying hey do you need some help? You know I I noticed that there's a spike in your traffic. are you guys scaling up? So, so those kind of nuances, you know, really helped us kind of choose across the board. And um the third is the quality of innovation, right? Um the AI world is evolving so rapidly. You know, it's very hard to keep up. Every week there's a new model. Every week there's some new breakthrough [music] that's happening. So, you also need a provider that has their own velocity that's able to deliver on at quality of the pace at which the industry is moving. So, those were like these three parameters, right? quality, velocity as well as innovation that based on which we kind of decided that hey 11 Labs is the right partner to work with and you know it's it's been great working with you guys for our on ground intelligence uh because we could like listen to customer conversations and we were able to kind of you know figure out a lot of these problems. So one is like I said we were able to detect mechanical issues in particular cars. So we were able to identify hotspots uh where refurbishment was a problem. So you know we got 9 10% improvements in our refurbishment costs. Um second is now if you're listening to consumers and making decisions backwards on that our customer satisfaction did go up right like 20 30% in certain cases we saw the sales improve when we made good decisions based on consumers as well um you know and growth as well like I said um you [music] know our salesman our car advisers that had 18% conversion versus some that had 28 30% conversions individually the people who were back right we actually could give them directed feedback and actually help them move from let's say 18 not to 28 because that's our best folks but you know we could help them move from 18% conversion to 20 22 23 24 gradually so those [music] kind of things do compound it if you have the best people on the floor your overall average conversion does go up like 10 20%. So um you know it's not just that we need to listen to a consumer it just directly gave us impact in our business as well. Would I recommend LeSol right I mean [music] um if if I was a betting man in this AI game for voice I would definitely bet on 11 Labs. It's just that uh you know 11 labs is a trend setter. They [music] have the best models um and you know they're not following trends that hey here someone beat a benchmark and they're following up. They're actually setting these benchmarks for us. So you know of course if anyone kind of really wants to build successful voice AI solutions I think 11 Labs is the partner to go for.