How has Carro put AI to work?

We started using AI on day one. We’d already developed an algorithm at Carnegie Mellon to figure out the best car on the market given certain requirements and used it to build Carro. In the US, there are numerous sources of information like CARFAX and Kelley Blue Book about used cars. But in Southeast Asia, there’s much less information and data. To fill that gap, we use machine learning and game theory for pricing. We use computer vision to extract relevant information from digital images. We also used sound signals to determine the internal quality of a vehicle. Listening to the engine can help us determine, for instance, the vehicle’s age and whether its sound signatures were irregular—indicating a prior accident or water damage.

Have recent developments in generative AI changed the game for you?

We already use AI in our external-facing work. We now see generative AI as something that could transform the way we work internally. For example, we recently built a training AI for sales staff using generative AI. In the past, this type of training would have required one-on-one work: a trainer teaching a newcomer. Now, sales staff can interact with the AI and go through its training anytime without the need for human involvement. This greatly improves productivity and, more importantly, consistency. Externally, say a customer wants to buy a Subaru XV, but wants to know how it compares to a Mercedes GLA. That’s easy enough to answer using a lot of technical jargon—miles per gallon or things like torque, etc. But the generative AI goes way deeper than a superficial comparison—delving into details that are more personalized based on the customer’s needs and that are explained in layman’s terms. For example, it may tell you that one car’s trunk can fit three golf bags, or that the other has a history of engine recall issues. I think of this as actionable information on top of the facts. And we saw improved conversion rates thanks to the added capability of generative AI to our sales team.

Is generative AI changing the way you hire and how you organize the company?

I’m experimenting with that right now. I’ve assigned data scientists to groups like finance and operations, instead of having one larger, siloed data department. In the past, the data science team would’ve been part of a team of 20, 30, up to 100 data scientists. These days I’ve assigned data scientists to departments in groups of two or three. That’s the power of generative AI. You don’t need that many people, just a few of them. That’s a small experiment we’re testing: putting good quality data scientists straight into a department instead of managing them centrally as a pool. We’re going to see whether we can reduce costs in those departments, improve efficiency, and consolidate. I’m also interested in figuring out how to put AI into marketing and finance. Another area where we want to test AI is language. We operate in many jurisdictions and languages. Could we use generative AI to centralize customer service in one country, versus multiple countries? That could help with expansion.

Are there places where you’ve experimented and determined that generative AI isn’t helping?

Generative AI does not work well when it comes to predictions for very recent or real-time datasets. Further, it does not work very well for pricing prediction for vehicles. In my opinion, these are not exactly limitations, but more like the wrong application of generative AI as a type of technique or algorithm on the wrong problem statement. We have a lot of things happening that are still in experimental stages. For instance, we’re using AI to predict revenues and profitability in the quarters ahead. By putting all these details into a large model, we will be better able to predict where we can improve profit or reduce cost down to line items—oftentimes in places we never thought about. Another place we’re using AI is in research of electric vehicle batteries. Generative AI is extremely good at summarizing, condensing, and cross-referencing big amounts of data and literature. Compare that to a researcher trying to access and synthesize across the thousands of journals and pieces of information that are out there.

You’re obviously very comfortable with experimenting. How do you avoid getting stuck in the planning and iterating stages?

We intentionally set up Carro as a place where people truly can experiment. We fail fast and kill a project quickly if it’s not working. What you don’t want is a situation where you don’t try. Or you do try and it doesn’t work but you keep throwing resources at the thing that’s not working. For me, that’s the opposite of innovation.

What’s a good place to start with generative AI?

Reading. There’s a lot of literature on generative AI these days. Your potential use case is probably already in a video or article out there. At Carro, we like to really, really look into our problems. I ask myself, what if I use AI here? What might the outcome be? We may not know the eventual outcome, but the trick lies in a “divide and conquer” approach: breaking problems into small actionable steps before spending too much time and resources. We’re also focused on building a product that attracts brand love. So it’s more than simply implementing AI across what we do. We need to figure out how people interact with the AI so that it becomes useful. We can build the best tool in the world, the absolute best predictive AI. But if the customer doesn’t understand how to use it, it’s useless. It’s about so much more than the algorithms—it’s the interaction with the AI. To stay ahead of the competition, we’re focused on delivering a product that customers can intuitively use, and that builds brand love.

Many people are concerned about downsides of AI when it comes to jobs and other risks. Do you share those concerns?

AI is here to stay, but I don’t know whether it is going to take over the world. It’s a matter of how we harness it. It doesn’t necessarily mean that people will be out of jobs, but that jobs will be different. We’ll need more people to help improve the AI, and improve the capabilities of generative AI. The time it’s taking us to create new processes at Carro is now much faster thanks to generative AI. We’ve been able to get a usable form of generative AI ready in about a week—not necessarily market-ready, but good for internal use. I don’t think that means that programmers will be out of a job, but rather that AI is helping improve productivity for everyone—not just the business side of things, but from the programmer standpoint, too. For me it’s about harnessing AI for greater productivity in our personal lives, and in our work lives. I myself would prefer to spend less time on mundane tasks. AI is already helping me there.

Is there a risk for companies that don’t embrace generative AI quickly enough?

Potentially for incumbents, yes. We just got a major investment from one of the largest auto conglomerates in Asia Pacific. They invested in us because of how we apply AI across the whole used car retail value chain, including aftersales. The incumbents are often concerned about becoming irrelevant since car manufacturers are going direct versus through distributors. Investing in Carro helps them hedge against future risks as they speed up internal push of AI across their company. I believe everyone should embrace AI as it truly helps to reduce cost and improve sales.

English

Q&A

Aaron Tan: How AI is driving used car sales in Asia Pacific

Carro has digitized the entire buying process. Now he’s looking for more opportunities to put generative AI to work

It’s more than simply implementing AI across what we do. We need to figure out how people interact with the AI so that it becomes useful.Aaron Tan, Co-founder & CEO, Carro

Founded in 2015, based in Singapore, and with 5,000 employees across 10 countries, Carro transacts approximately 150,000 units per year across seven Asia Pacific markets. With a revenue run rate of more than $1 billion, it is profitable and was the fastest-growing company in the Asia Pacific region in 2022.

In a broad-ranging conversation, Tan discussed how he and his co-founders—fellow Carnegie Mellon computer scientists—implement and experiment with AI at Carro, and how they constantly look for new areas where AI can reduce cost and improve both the customer and employee experiences.

Interactive Data

CFOs are cautiously optimistic about sidestepping a recession

Explore the full results of our Q2 2023 CFO survey. Compare and benchmark your company.

Feature

Cloudbeds: The art of
calculated risk

Big swings are a top reason for Cloudbeds’ success