Ilya Segalovich
Yandex

Ilya Segalovich is co-founder of Yandex, the leading search engine in Russian-speaking countries. The roots of Yandex trace back to a company called Arkadia, which in the early 1990s developed software featuring full-text search supporting the Russian language. In 1993, Segalovich and Arkady Volozh came up with the word “Yandex” to describe their search technologies. The web site, Yandex.ru, was launched in 1997 and in 2000 Yandex was incorporated as a standalone company.

In May 2011, Yandex raised $1.3 billion in an initial public offering on NASDAQ. It was the biggest IPO for a dot-com since Google went public in 2004.

Pedro Santos: How did Yandex actually start? How did you start with the idea of a search, and how did it evolve to become Yandex.ru?

Ilya Segalovich: The idea wasn't mine. It was my very close friend from the school, Arkady Volozh. He was a software engineer in a Soviet kind of institute. They were doing something like building pipelines, and one of the things that he was appointed to do was create a patent database. At that time, he had a friend working at the Academy of Sciences Computer Center in Moscow. They were working on language processing software for Russian. Actually, it was a thesaurus that would include allof the words in the Russian language. He came up with the idea of applying this linguistic software to the database he was creating.

The friend's name, incidentally, was also Arkady, so calling their start-up Arkadia was only natural. Arkadia is a land of happiness in ancient Greece, or something like that.

Actually, I wasn't part of the very early efforts. I came into it all on a permanent basis in 1990, though. I started by doing small things in '89. But in the late 1990s, I was already part of this team and was helping with shipping the first product they made. It was a search program for patent databases. It was very nice software for MS-DOS. Actually, probably the first software that was able to perform search on a small MS-DOS computer using a hundred- thousand-word dictionary.

It was quite successful. We were selling this software for about two or three years, until probably 1993 when we ran out of money and we couldn't sell the original product anymore. We were developing new versions of that product, but we couldn't sell it, because by that time the industry had collapsed.

The patent industry of the Soviet Union was huge, because in the seventies and eighties, Russians were actually inventing everything from scratch. Everything from shoes to rockets. From 1991 to 1994, the country's industry had been disintegrating, first in engineering, and then in patent ability, so we were able to sell next to nothing.

We almost dissolved. Luckily, Arkady is a great entrepreneur, and he started two companies in '89. One company was started with his computer genius, and the other one he started with his friend from Boston, who was helping him to import hardware. Actually computers, PCs.

Later, they changed from simple PCs to more advanced things like equipment for computer telephony and wireless networks. They were trying to promote new technologies in the mid and late nineties. I joined this larger hardware company, along with some other people who worked on the software, when our companies were kind of acquired, so we didn't go away, but stayed with Arkadia and became a small part of this bigger company—which wasn't all that big of course, but it was steadily growing.

My idea was to make a demonstration product. Take some classical text and use it for a nice shell for Windows. We started to sell the Russian Bible in '95. It was a nice, very well-designed, product to search, and it also showcased our capabilities in language processing and search.

That product got us some orders to create similar products, also for Windows. I think that in '96, after we had got the internet in '95, we understood that it was time to go wider than that and start making real search for large data. In '96, we created software to search web sites, and we called it Yandex. That was the launch of Yandex. Not Yandex.ru, but Yandex as software to sell, to use as a search tool on a web site.

In '97, it was still in CompTek, a large hardware company. In '97, we launched Yandex.ru, and the first idea for Yandex.ru was to promote the software and increase sales. The idea was, “Okay. We are so good in search, we have relevance, we have language processing technologies. We have great, deep, nice, very fast, very clean results. Let's just show people how good we are so that they start buying.”

Santos: The software.

Segalovich: The software, right. Actually, we had been using this model, I think, from '97 to '98 at least, and only in '99, did we really switch to an advertising model, and started to seek venture capitalists. I think late in '99, we found venture capitalists. They were running around in all markets, because then there was a huge boom, as you remember. Maybe NASDAQ was at about five thousand or something like that. We found investors late in '99, and officially got the actual investments in March 2000. A few days before the first NASDAQ crash. And then a new story started.

Santos: One question about the venture capitalists—where are they from? From Europe, from Russia, from the US?

Segalovich: It was a mixed team led by Elena Ivashentseva. Baring Vostok Capital Partners and ruNet Holdings. Actually, there were two bodies that included different people and different groups of people. It wasn't a single person. It was two groups of Americans, Europeans, Russians, all kinds. It was two teams. But both leaders were Russian. The person who made the decision and led this investment was Elena Ivashentseva, she's Russian.

BVCP is Baring Vostok Capital Partners. It's a very successful investment fund in Russia. One of the most successful.

Santos: What were the main lessons and challenges that you learned in just initial phase of Yandex?

Segalovich: That was quite a hard time, and several things are important, and I think every small start-up experiences this. I think one of them is when you create a software product, you have to learn how to sell it, you have to learn how to make it a product. It's a very basic skill. I think every engineer has to try that at least once, to sell the software he created, regardless of how bad it is. No matter how unpolished your product is, you have to try to explain why it is good for someone else.

I was doing all kinds of things, like technical support, installation, distribution. Lots and lots of different things, including MS-DOS, Windows, then FreeBSD. And if you do all this stuff, it helps you understand what the product is and how to make it better.

And then the important thing when we started selling was that we had a very, very small audience because the Russian internet was quite small. But very soon, I think maybe in a year or half a year, we understood how important it is to learn from interaction with users. We started a web forum and people started commenting on each feature we launched, and we worked hard to create the feeling that we're constantly changing. We were constantly updating the product.

I think during the first six months, we changed our search about every two weeks. We had a very small team, maybe six people or something like that. But we were so eager to interact, try it and listen to people. In the early stages, you have to constantly change your product and see how people react and how they like it. I think that was very important for us.

And I think later, in 1999 or 2000, we had to choose between doing only technology or doing only web search, artificial intelligence, language processing and provide search capabilities to other portals, or alternatively market ourselves as a portal. It was clear for us that the portal model has a huge advantage when you have servers, like, service interaction servers.

So, without user feedback, we understood, we just couldn't make it perfect. We couldn't make it a really good product. So that was one of the things that we learned in a very early stage.

Santos: And how was the move from the MS-DOS and Windows platform to the internet?

Segalovich: Fortunately, we were trained to be quite perfect. We didn't want to be dependent on Windows. We never had a code style that required using direct Windows system calls, something like that. We were writing C++ and we believed in algorithms more than in integration with the operating system. So when in '96 we had to port from Windows and DOS to UNIX, it wasn't a big deal. It took us just a few days.

Santos: Days?

Segalovich: Believe me or not. Several days to port everything to a FreeBSD in '96. Yes, seriously. We were writing quite clean C++ code, very, very focused. Another example, I think it was '98 or something like that, I don't remember exactly. Because, as you remember, we were trying to sell search software for web sites, right? So we had Yandex.ru as a …

Santos: A storefront.

Segalovich: Yes. There was a company that had a server site more popular than Yandex. It was called InfoArt. They asked us to give them a search feature for their site. So we were trying to give them a site search, but they were purists. They were running Suns, okay? Sun Solaris. And they were saying, “No, no, no. No Linux, no FreeBSD, we accept only Sun code.” So it took me, I think, a couple of days to port everything into Sun, though Sun has that weird bit order. So with this weird bit order, it took me another couple of days to understand it. It's really weird.

I think we kept a big part of the code constantly ported to Windows. I know some people at Yandex aren't happy about it, but we keep the Windows version, we keep the Sun version, we keep our code quite task centric, not operating system centric. If you do task-centric code, try to avoid dependency on lower levels; that's our kind of motto. This might not apply to everything now, because now we have eleven hundred engineers writing code, so I cannot speak for everyone. But I believe a big part of it is still about writing algorithms and writing good code, not writing for a particular platform.

Santos: Where does the name Yandex comes from?

Segalovich: I think it was in late '92 when I finished the first version of my own Linux processing code. It wasn't good, so we tried to sell that with no success. We were quite depressed and it was a very low moment, when I got some inspiration reading an algorithm book or something like that about how to do it. Very quickly, with the help of Arcady, who helped me with some code, we made a new version in two weeks. It wasn't doing everything, but the ninety percent it did was amazing, really nice speed, with one thousand words per second on MS-DOS, which was amazing. It was really good.

We were so happy about it and very quickly, in two or three months, we were basically finishing all essential components to start text searches with this new core. I think it was in May when I told him how good what we had at that point was. He was very inspired and he asked me to come up with a name.

At that time I didn't have an apartment in Moscow. I was staying in the office. The office itself was also a rented apartment. So I slept at night in the office. And in the morning, I didn't sleep. I was writing down the names.

I wrote a huge list of names. The first one was Yandex. Then Arkady looked at it and suggested that the first letter should be Russian or something like that. So he liked it from the very beginning.¹

Generally, it was the inspiration in how good a product we had and finding an appropriate name.

___________

Santos: Was Yandex hit by the internet bubble when it burst?

Segalovich: Fortunately, no. I have to praise Elena, who was extremely tough. She was always on top of things, very focused, and didn't let any of the smaller investors run away. She kept us from this trouble. We had the money secured and we were slowly burning it, of course, because in 2001 and 2002, we still had to spend. But we were working hard on getting to the peak here. Up and up.

I think in 2003 we already became profitable and it was good. But at some point in mid to late 2000, it was very critical. Because, for example, our competitors, Rambler and Aport, were hit very severely by the burst of the bubble. Investors ran away and they were cut out of money. Aport had no money. Rambler was really bad. They didn't have money for development.

In our case, we had that money in the bank and we could do whatever we wanted with it. We were very smart in not spending just like that, but spending really carefully and trying to get profitable and develop as many services as we could to create a server farm. It wasn't a farm. It was like a rack or something like that. Several racks.

I think in May or June we decided to spend some money on TV advertisements. The idea was that no internet services had been advertised on TV in Russia until we did a commercial in May of 2000. That was a cool move, just to show that there is such a thing as the internet somewhere out there.

People didn't even really know what it was, but they already knew that name. The advertisement was very fun, very good.

Santos: When you entered the internet phase of Yandex, you started as a storefront and then you went into the advertisement model. Did the business model evolve even more or did it stay with the advertising model throughout its history?

Segalovich: No, the original idea was to have banner advertisements, all big brands, and we hired people whose job would be specifically to attract, bring big brands to Yandex. And it didn't work well, actually. Banner advertising wasn't working well. At the same time, I think in late 1999 or maybe late 2000, when GoTo became Overture and started serving ads on all big search engines. Later, I think Google started doing it on the side. That was a big inspiration for us because I think we started with copying GoTo and then Google, not in the sense of actually copycatting them, but understanding how important it was to use. The power of advertising, right?

I think in the summer of 2001 we already had very simple search, keyword-based advertising snippets. Then I think it took us about two years to understand that we could do an auction model. Also in 2001and 2002, there were almost no advertisers. The internet was very small.

In 2003, that was the year when we switched to the auction model and that was a different model for us. In about two years we started context advertising, which was also very important, that was in about 2005, I think.

Santos: How did the search algorithm and how did the technology evolve with time? How did you measure accuracy and improve the search results?

Segalovich: We started, as you remember, very early. When we launched web site search in '96, we had very simple relevance measurements. I think in '97 we made the first big effort to improve search quality. We created a corpus and had something what we could call assessors. We didn't call them this at that time, but we had some ratings for each document that we did, that our systems could find. Based on those statistics, we made our first search function that was more or less purely text-based. I think it incorporated maybe five different features. Later, we added link analysis. I think in 2000 and 2001. Then we added some other features.

I think in 2004, we started collecting a database of assessor ratings. We started using it effectively to tune our algorithms. But probably until 2006 or 2007, we didn't have that level of focused development for different features or machine learning on top of that.

Why is that? It is, I think, one of the reasons we were better than other Russian search engines. None of the important search engines, none of them were incorporating Russian language until 2006.

So we didn't care, because they just couldn't find fifty percent, at least half of the pages that we could, just using simple queries. Typing one word on Yandex you could find twice as many pages as on any other search engine.

Google, I think they made their first attempts in 2006. Then in 2007 they added Russian morphology and Russian linguistics. It hurt us quite significantly because we knew that in many search aspects we were worse.

So we started to effectively develop every missing aspect of our search technolgy, including user behavior analysis, and increased our effort in extracting many other factors and features out of texts, links, URLs, user behavior, general statistics, site statistics, everything.

From 2006 to 2008, we were working hard on improving both the number of features and the way we 'taught' our search function to return the results we needed. I think in 2009 we made a big step forward when we introduced something we call MatrixNet, which is a very robust and very quick machine-learning algorithm that is highly pluralized, both in learning and in computing.

Currently, we are performing machine-learned search rankings using, I think, one thousand features and hundreds of thousands of matrixes. It is the equivalent of trees, if you are familiar with decision trees. But this is decision matrixes.

Actually, that's how we now rank. We work hard to look at search, not only purely in terms of relevance ratings, but also at others of its many aspects, such as freshness, actuality, authority, originality. We work hard to improve every aspect of search, basically using machine learning and incorporating new features, new factors, and so on.

Santos: You mentioned that Google, when it entered, was a threat.

Segalovich: Well, they entered our market in 2001, actually. They conquered the whole world in 2001 and 2002.

Santos: Yes, but they failed to conquer Russia, even today.

Segalovich: Yes, they did fail. Because we were quick enough to respond with all of thesearch quality that we have now.

Santos: Did you ever consider going after Google in their own territory? To attack the American market?

Segalovich: Well, one of the biggest problems here, as I see it, is brand perception and user perception of the brand. We are unfamiliar to users outside of the CIS². We actually, of course, want to be outside of Russia sometimes. We have made several small steps. We are big in the Ukraine. Also, we are big in Kazakhstan. We want to try something that is totally not Russian. For example, what we did recently was we launched Yandex in Turkey in September. This will be a very interesting experience for us, because we want to offer high-quality search and high-quality services, other services, and see how people will react, if they like it.

Will we be able to advance in that market? To what degree will we be able to do it? If we are successful, it will be a good sign for us. Of course, we want to be bigger than just a Russian search engine, definitely.

Santos: Yandex has an office in the Netherlands. Why? Do you have development there? Do you have sales there?

____________

Segalovich: We have a data center, but it's quite new. A longer story is that we are incorporated there. Our lawyers are there. It's because they have so many different investors from all over the world. They considered an option to make appropriate legislation in Russia, but some Russian laws are a little underdeveloped, especially in terms of shareholder agreements and shareholders' rights. So that's why. The Netherlands are not better in terms of taxes. Not better than Russia at all. The only reason to have a mother company in the Netherlands is just to have better protection and better transparency internationally. Even in terms of IP law, it was more important to stay in the Netherlands.

Santos: It was a matter of credibility to the foreign investors?

Segalovich: Right.

Santos: I actually thought when I saw it that you were thinking of expanding and maybe using the Netherlands as a trial base.

Segalovich: No. We are thinking about some European countries, but currently we only have Yandex in Turkey, and we'll see. I think for us, Turkey is a model [for] expansion to new, unfamiliar European countries. If we are successful in this country, then we can go somewhere else.

Santos: You recently IPO'd on NASDAQ. Why not in the Russian stock market, or in a European stock market? For instance, the Dutch one?

Segalovich: There is a simple explanation. There's better visibility on NASDAQ—more investors, more people. We will be reviewed better there. It's a question of business. Higher valuation and all this kind of thing. Very technical, not political.

Santos: What are the major differences for the company, between being a private company that it was until now, and now that you're a public-traded company? Did you have to make any big changes?

Segalovich: There are several things that have changed. One of them is that we had a relatively high compensation plan. We were giving options to about ten percent of our employees, and many of them became wealthy people now. With the market price that they have now after the IPO, we have to change options to some other instrument. Because now everything is transparent, the price is transparent, everyone knows the price of Yandex. Everyone knows the option price. It's not established once a year or something, at the board of directors meeting or something like that. It's all transparent, so we can use it more freely to stimulate our people.

Another thing is, of course, acquisitions of companies. It's very transparent now, having this price established and everything.

There are some negative impacts, such as me having to sacrifice three weeks [of my time] for the IPO in May, and maybe a week or two later in the year or something like that. That's not good. Fortunately, I do not look at the index at all. I have an app for that, but I never use it. Honestly, I don't know what Yandex costs now. I don't know the price. I believe some people are a little bit worried about this. It's not good if they're just looking at this thing. You have to work, you have to create a product, work on the core services and make them better. That's the main purpose, not the index or something like that.

Santos: What were the hardest challenges of Yandex's history that you faced, in your opinion? How did you overcome them?

Segalovich: Actually, because the story of Yandex is long, there are different periods. Each period has its own hard time. In the early period, I think one of the hardest times was 1995, when we already had that Russian Bible search, and we had product orders, so we could develop things. But the mother company was much more successful. There was some pressure to stop development of this low-margin software and move everyone to hardware sales. That was a very critical moment, and we managed to keep that part working.

Santos: How did you convince the rest? What was the selling point?

Segalovich: Just because I didn't want to go into sales. I had good credibility from Arkadia and Arkady. I think he defended me from all other members of the board and all the other people who were really hesitating, really. They were not believing in the future of search at all back in 1995. They just didn't know it would work out.

Another moment was I think in late 1999, when we were close to getting investments. At that time, we knew the race was on. For example, Rambler got investments, and they had one of the best language processing companies in Russia joining the regional Rambler team.

That was a very strong team that we were quite afraid of, and they were number one in search. We were probably a distant third or something like that. We were small.

But we managed to grow from ten people to one hundred people in three months, and create maybe ten services in half a year. Very important services. From email to hosting, parallel search, lots of things. That was very intense development, extremely intense, and also a very inspiring time.

In late 2000 we became number one. I think another critical moment was in 2006 or 2007, when Google finally added Russian-language development. They started focussing on the Russian market. They created a great Russian spellchecker, with a great search term suggestion [mechanism], everything.

They were really focused on Russian search. At that time, we had to expand from maybe fifty search engineers to maybe three hundred. We did it in a year and a half and really reorganized the team because the original team had just me and some other guys. We were unable to develop and hire at the required pace.

Everything had to be restructured. Without hesitation, we had to expand several times more, and really work hard without knowing where to go.

Santos: When you went from ten engineers to one hundred engineers, how did you keep the culture in the company? With such growth, how do you keep the quality of the engineers?

Segalovich: I think it was from ten people to one hundred people, so maybe from five engineers to fifty engineers. I think that was an interesting moment. Unfortunately, at that time, we went from one culture to three different cultures in the company, and we stayed with three different cultures through all these years. We were very liberal. We were very tolerant. Too tolerant. We did create the style guide for our search code, though, and we were quite cautious about keeping the team working on one code base that was universally compiled on many platforms and everything. In some other parts of engineering, they were more loose, more loose and more flexible. They were using different languages and everything.

More or less, we got three different cultures, and they were mixing and going together. Portal services and email have one culture, banner system has a different one, and there is a special culture around our search. So, there are three different cultures. Yandex is a multi-cultured, multinational organization.

Santos: If you could do one thing differently, what would it be?

Segalovich: I have about five. And some bitter remorse every day. Why did I do that exactly like that? One mistake at an early stage creates these huge trees that you cannot merge at all. In engineering, I know my mistakes that are now deep in the infrastructure. They split everything. As for organization, yes, that moment in 2000 when we were maybe too loose in some aspects of the company, maybe it wasn't good. We should have been more unified in a way. It's hard to say now. It's history. I don't know.

Santos: What would you advise a new entrepreneur? What advice would you give him?

Segalovich: Actually, life now is very different, because I see how people are managing to create companies in a very thin way, with five high-profile engineers. And all they do is just the algorithm. Everything else is outsourced in a way, from finance to lawyers, and more importantly, database system administration, everything is in the cloud. People are trying to use other companies' APIs, other companies' layers of services, as much as they can, concentrating sharply on what they think is good. Very rarely, is this thing a product. Sometimes it's just some part of some weird product, some weird part.

But sometimes it's a standalone product. And if people stick to this goal, and find that others start using it as a complete product, they start expanding, and start putting more value into that, I think that's just a great way to go now.