How Didi Works

Zhang Xinyu

“Sometimes when I get off work and hail a ride, the driver is a product manager from one of the big companies nearby [Didi, Baidu, Sina, Netease], because I use my real name.”

This is Yu Jun, one of the earliest product managers working in China’s internet industry, and formerly VP of product at Baidu. As the idol of many of China’s product managers, Yu’s “product methodology” has long been admired, and his thinking has influenced an entire generation of product managers. More than a year ago he joined Didi as a consultant. Now, he is Didi’s senior VP of product development.

Over the last two years and as the mobile internet has matured, there have been fewer and fewer breakthrough products, rendering marketing, branding, and operations all the more important for internet companies. But for truly successful—one might even say great—companies, the quality of the product itself remains paramount.

PingWest sat down to speak with Yu Jun about his work at Didi, and the evolution of his thinking about product management.

Didi’s work, Yu Jun says, is organized into a system that is at once both similar to and distinct from that of Baidu, Alibaba, and other large tech companies, something that is hinted at in how the company searches for and recruits its professional talent: “An ordinary product manager sets down roots in one field, while the outstanding ones will accumulate different experiences in different fields,” he says.

Yu admits he did not used to spend much time on casual reading, but his office is now littered with piles of books like Principles of Economics, Economic Development Theory, and Statistical Psychology. It reflects a part of his own ongoing growth, as well as the needs of his work at Didi. Yu believes that Didi’s product design and business model are not entirely that of a “free market” and third party platform, though they obviously don’t fall under the heading of a planned economy, either. Hence the economic theory: he, like so many others, is trying to work out where a ride-hailing company fits in the “new” economy.

Below is a transcript of our conversation, edited for length and clarity.

Q: How do you determine Didi’s product direction?

Yu: Everything starts with user value. Where are users unsatisfied? What do they want?

There are many channels for this, a lot of feedback. We use Didi ourselves, including our product managers, engineers, and other high level staff. [We also look to] our outside partners, our friends on WeChat, and we can gather a lot. You just have to filter it, but there’s always a lot of needs laid out there. And then you make a judgement: this is a pain point, that is a need. Then we analyze each one, how many people they affect and to what degree, and then we rank them. We look at the technical resources we have to invest and which ones are relatively easy to solve, which will be longer term projects. We just list them out like that.

Of course, this is commonplace, but for a product manager, you always have to be thinking "what is key, where is the greatest benefit, what could be most transformative?" Because a tech company can’t rely on ordinary optimizations to sustain growth, it has to think about what is truly transformative, what pain point it is or where a new factor can be introduced.

We also do feedback, because that’s where you get inspiration. It might be a detail that’s key, but if you didn’t do all of that [at the start] you’d never find it. Of course, you might go to all that trouble and find nothing, and then you still have to do the routine optimization work. But you always have to be looking for that transformative thing, the next main growth point.

Didi currently is mostly working on carpooling, but that’s a longterm [strategy]. It isn’t to say that I’m doing this today and I’ll see an obvious change by tomorrow. I know it will lead to change, but it might take half a year or a year of gradual improvements. Market penetration and acceptance take time.

Q: Does Didi collect rider feedback on social media?

Yu: We do have that. The userbase is so large, there’s always someone mentioning us.

With carpooling, there is a social component, like with drivers picking riders. Some drivers just do it when commuting to and from work to pick up some gas money, and they don’t care who their riders are. But some drivers, like in Zhongguancun or Xi Erqi [two Beijing neighborhoods where a lot of tech companies reside] the driver wants those tech company employees, so they can socialize with them.

Like for myself, sometimes when I get off work and hail a ride, the driver is a product manager from one of the big companies nearby, because I use my real name. It’s a way of socializing.

The ease of that in carpooling is different from that with [on demand] hailed rides. But we can’t focus too much on it, because we also have to tend to the needs of other users. So we do work on the social aspect, but it’s not an especially big point. There’s less need for social networking with the hailed rides. In our first year, maybe there were some drivers who got into it specifically so they could talk to people, and I can sometimes still run into some old drivers who like to chat, but they’re a minority.

There’s another direction for social. We do carpooling, and behind that we consider carpooling within companies. So right now, carpooling is open to anyone, and although that’s very efficient, some people are reluctant to do it. But say we’re at the same company, whether it’s Tencent or PingWest or Didi, and we’re always coming to and leaving work together. Especially at these big companies, everyone lives out at Hui Long Guan [an area on the north side of Beijing], and if we’re at the same company we might want to carpool.

Naturally, the rider and driver can choose one another, but only when there’s a match can there be social interaction.

Q: In the near term, where is the main focus going to be for Didi?

Yu: Actually, it’s still a question of balancing supply and demand.

Because our business model is very particular, other companies haven’t really run into this type of problem before.

It’s a business that’s destined to undershoot demand. If you do a search engine, or a news feed, or even if you do social media and games, maybe an ecommerce platform, you never worry about not having a supply. A small amount of the time you might run a shortage of some good, but for other businesses, the more customers you have, the better. For social media, for games, it’s the same: the more content you have, the better. The more goods you have, the better. For other businesses, the keypoint is finding needs, and then you organize the supply.

But with the ride-hailing business, the user need is unlimited. Living standards improve, and everyone wants to hail a ride [to get around]. As long as incomes increase, the demand increases. But the need increases faster than the supply of cars and roads. So over time, the supply can’t keep up with demand. This is one aspect of it, a major trend.

And then more specifically there’s a question of availability. At morning rush hour the supply is never enough, because there are more and more people. It rains, or it’s especially hot or cold one day, and the demand increases, maybe by 30, 50%. But more drivers can’t just suddenly appear. You might get a handful more coming out, three or five for 30 or 50 additional riders.

That’s time, but then there’s also location. At a given time, over here it might be dead, while it’s bustling over there. There are more people within Beijing’s Third Ring Road, right? But drivers aren’t rich, they can’t live in the city center, they live outside the Fifth Ring Road or even further away. So in the morning, before the drivers have come, there are a lot of people in the city center who want rides, and demand will be far greater than supply. The drivers aren't willing to go 20, 30 kilometers just to pick up one person, so the supply has to be smaller than the demand.

Some cities are more segmented, and supply and demand are more balanced, but over the longterm Didi's model will still be different from those of other businesses.

So we have to expend a lot of energy to solve this problem. Because supply is smaller than demand, with so many people, and in so many different environments, they won't understand. Every day there will be people who can't catch a ride, and they're going to be mad at you. Although some people can understand why it works this way, there will always be others who don't and aren't satisfied.

There's a lot of pressure, and we are very cautious. To solve this balance problem, we have to increase supply, and that includes increasing efficiency of dispatches. Increasing carpooling is another efficiency. [We look at] all kinds of ways of increasing efficiency.

But if we are increasing it incrementally, and there are so many people hailing rides, just what user needs exactly can we meet? For instance with frequent riders, we could have a membership program. Or should we just let everyone queue? Or should we use dynamic pricing? Or should we give favorable pricing to the elderly and disabled?

There's a lot of pressure on Didi. Do we control dynamic pricing? If we do already have it somewhat under control, we have to look for other ways to meet user needs ... Every day there are going to be some people who can't hail a ride, so who is that going to be? How do we reasonably meet user needs, how do we manage to be fair while being economically efficient?

I feel we could work at this for many years, and the more mature the business becomes, the more we need to spend our energy on it. Big data, ethics, economic research—we've sought out economists and psychologists to research this. So the complexity of this business is interesting, and a lot of overseas Chinese professors are interested in it, they're willing to get involved when they hear about this business model, because it's a big challenge.

Specifically, the carpooling is one obvious direction for this, because it has a lot of certainty, and we can explore the pricing model. We are trialling queues in Zhongguancun and Guomao, although for people getting to and from work, they might be left waiting a while, so there's no certainty there. But for now we can do queues, and tell you how many people are in front of you, and before long we'll roll that out across the country. But queueing isn't good for everything. Some people might have something urgent to get to, and you have to give them a way out, some way to identify those people, and that's also something we're looking at.

Q: What riders care about most is still price, so how is Didi handling dynamic pricing?

Yu: Our business is a nonstandard product. Standardization can only be seen in a part of the service you provide the rider, such as in the app experience, the customer service, the dispatch efficiency. But most of the time, the middle part—the way the driver drives, the phone call communication with the driver, the smoothness of the ride and the service attitude, those are what determine the experience. Every driver is different, and their moods are different with each ride. Every order is different, and there's a certain probability of problems arising.

So the reason we ask economists and psychologists to help us is because this is an interesting, complex [problem]. We want to gradually improve the fairness of the rules and make the service more even.

We've already established a pricing cap system, but through the pricing leverage we can detect that some users might be in more of a hurry. Some lodge complaints saying, "I couldn't get a car when I really needed one, you didn't give me any choice, but I'd pay an extra 100 if only I could get a ride."

So how do we satisfy that greatest need? We could work on this for years, and it needs support on every side, not just technical support, but economic, psychological, and big data support.

We hosted one economist who asked us, is Didi a platform or a business? A platform like an ecommerce site [like Alibaba], where it doesn't set the prices, but buyers and sellers set the prices through competition? Or a business, which sets the prices itself?

Thinking about it, this business is very peculiar. There's high efficiency in setting the price ourselves, but if we do that, both sides will feel wronged. Riders always want lower prices, so that no matter how you set them, they'll think they're too high and that you're ripping them off. Drivers always want the prices to be higher, so no matter how you set them, they'll feel they're too low and that you're favoring the riders. The trade off here is very important.

Perhaps in the future it will be a combinational model, with partly the two sides able to freely set prices, but with us in the middle giving a suggested price.

This isn't a platform with completely free competition, where so many drivers and riders have all the time they want to choose one another. It's a short time range, regional, provisional market. You send out a call, there are a hundred other people and thirty rides within three kilometers of you, and you need to match up within a few minutes. After just a few minutes the cars will settle on their fares, and [if they can't get a ride] the user will give up or find another means of transportation. So you have to close the deal within a few minutes, and with every minute's delay the driver and the rider both lose some benefit.

It's not like most products, where you can buy it one day or the next and there's no great difference, because the product is always there. With ride hailing, everyone wants it done within just a few minutes.

It's fine to provide a platform for the two sides to negotiate prices. That's how ride hailing works now, with both sides able to see one another, so that right now they can choose riders or drivers. Later on they could choose prices with each side freely making bids, but the efficiency of that is low. Some people who are used to using the internet might think it's great, and they can choose the lowest price or the best car, but they are few.

For 90 something percent of users, the first requirement is to get a car quickly. You can only recommend a car at this place and time, where the prices are about the same, and then account for supply and demand.

We are exploring the possibility of giving everyone a free market in some cases where certain conditions are met. In those scenarios, we would supply a suggested price and allow it to be adjusted. But it has to start from the rider, because it is the rider who is paying, and if we started with the driver it would cause problems.

We haven't concluded that yet, it's just a direction [we're exploring], and we'll keep exploring. In most situations we can only give one price, for efficiency. Early on it was first come, first serve, with drivers grabbing fares as they became available, but we later found out this was inefficient, and organizing dispatches was better. Even if it wasn't perfect, it was still efficient in most cases, and beneficial for both riders and drivers.

Q: With Didi striving to raise efficiency, can it also establish an emotional connection with users, like some other internet services do?

Yu: It's definitely possible, because I understand that users' values are money, time, health, and emotional connection. But emotional connection is more complex, and we weren't very good at it before, while Uber has done very well on that score.

What I can do is just on the product feature side. I can make the user more satisfied, and make things more personalized.

We're doing some experiments with greater personalization or activity, not just for efficiency but also to obtain data. We're considering, for instance, if a user says, "I don't like this driver," then in the future they can block out that driver as an option.

Because some drivers live nearby and are often within range, there's actually a good chance they might repeatedly receive the same rider, so if a rider doesn't like them they can block them out. Some drivers also don't like certain riders' behavior, so in theory they could also block riders. We're looking into it, and then we'll take the data and market opinion and make a decision on that.

If you give both sides the right to block when they're not satisfied with someone, I think that's definitely a benefit for both.

We can do even more personalization. There are two types of people, those who like speaking with drivers on the phone and those who don't, or those who like having the AC on and those who don't. Some like chatting with the drivers and some don't, or they don't want music in the car. We've already put all of these things into the pool of needs to look into.

That's a lot, but if you want to make an emotional connection between a user and the service, things will multiply. In the first half of the year we did a "baby car" service, cars equipped with child seats. There weren't many, but some riders with kids want that. Some disabled people can ride in cars equipped for wheelchairs, and there are even fewer cars for that, but there has been great word of mouth from that. I think there's a lot we can explore with these kinds of things.