Personal Story on Leiki
I was interested in how computers work and studied them before I got access to one. In the beginning of eighties my uncle introduced me to his Apple II and in 1983 I made my first real program which calculated prime numbers. When I went to seventh grade our school had a connection to the university mainframes, and I also heard of the first internet-connected public unix service in Finland. So I managed to get online every day on the internet from 1986, while BBS networking (FidoNet) with 300-1200 bps modems was the mainstream with clued-up geeks.
Writing programs and learning new things from the clever people on the Usenet was great, but I felt computers were more of a tool than a career in themselves. I wanted to understand how the world really works and went to study theoretical physics.
From academia to society
In early 2000 I was working in Copenhagen doing research in theoretical particle physics. It sounds like a dream job; good salary, no boss telling me what to do, and full day available to investigate secrets of the universe that no-one else had ever found before. But there were some aspects of the job that I was starting to feel uncomfortable with. It was a little Ivory Towery. You get grants every two years, mostly financed by taxpayers. Then you work, mostly with yourself and partly with a couple expert colleagues, with very little interaction with the rest of humanity. And like a lot of science, the practical work was quite tedious; in my case most of the time was spent programming supercomputers with archaic tools (like Fortran 77), with a single run of the calculation sometimes taking weeks. Then you looked at the results and tried to figure out if they were due to some bug in the huge amount code, too many approximations in the computational model or perhaps even a real physical feature.
I’m a people person and a part of the job I enjoyed a lot was going to conferences to present your work and meet fun colleagues. In March 2000 I went to Japan for the first time which was amazing. I was looking at the mobile phones people had and noticed they were more advanced than in Europe. Some of them had little animations with cute characters, not functional but for a nice show. When I came back I started thinking why couldn’t these characters be functional with some “intelligence”, so they’d help the user find information which was really hard to look for using the phones.
The internet hype was at its peak and I thought it would be fun to start a company. So I started writing a business plan, based on an intelligent platform with animated characters as user interfaces. The system would follow the user’s actions and present interesting content and other users automatically.
Start of the company
I needed to find a name for the company. I thought it should be something short, unique, easy to remember and pronounce. Beautiful and positive wouldn’t hurt either. Finnish words are often like that, it’s a language that’s totally different from any other language except Estonian, with lots of vowels and a phonetic pronunciation. “Leiki” came up quickly. It’s the imperative form of the word for playing around, or engaging in playful interaction. Something you could say to a child to encourage them to have fun and be creative. While in Finnish you say “lay-ke” many English speakers pronounce it “like-e”, expressing a fondness for the company immediately. Most new companies add a word like “technologies” after the first word of their name, but we didn’t. Short and simple!
When I visited Helsinki in June 2000 I filed the founding papers with my friend and started thinking about funding.
I sent the business plan to various friends for comments and ideas. One of them showed it to his flatmate in London, whose family friend was an investor in Italy. They liked the plan and called me to come over to Lake Garda to discuss it. I’d never been there but arrived at the airport late at night and got a rental car. The map was full of tiny twisting roads, it was dark and raining and there was no GPS. Somehow I arrived at the right place and met a very friendly family. They cooked me dinner and we talked about the plan. I was asked how much I would need and I named a figure, not sure where I got it from – I wasn’t a business person!
In good spirits and without any shareholder contracts – our business is founded on trust – we received the payment on the Helsinki account in a couple of weeks and allocated shares to the investor in return. I was already sharing an office with another company, and we got most of our starting team from them.
First product version
Our fresh team had a meeting to discuss where to start. The basic ingredient seemed to be the intelligent recommendations system, so we needed to make a system to analyse both content items and user interests in detail.
The developers started to work on a Java application that would analyse content with an ontology and follow user actions to build detailed individual interest profiles. I was in contact with a friend at University of Oulu to get some academic studies on semantic content analysis. On user interest analysis based on content analysis there was nothing though, so I made up a formula for that. We started to develop the ontology based on an International Press and Telecoms standard, which was too general though so we needed to add many more detailed categories to make it work for content analysis.
To get some content we met Reuters in Helsinki, who gave us their content feeds. We plugged them into the first version of the product and in January 2001 we were up and running with a system that personalised the news selection based on your (and your friends) clicks in real time. This included the first version of Leiki Targeting, the engine inside Leiki Focus.
Original mobile focus
Our first commercial delivery was in 2001 to Sonera, an operator, for their research program on services of the future. Together with Sonera’s research people we didn’t aim low but created a system that was more than 10 years ahead of its time. As smartphones weren’t available the system used a PocketPC PDA hooked to a mobile phone with an internet connection to simulate one. Leiki Targeting was analysing content on the server side, but was also working inside the handheld device to do user interest analysis and personalisation. The content profiles were sent over the air to content applications, which used a shared, on-device Leiki personalisation service to find out which items were most interesting to the user.
In 2002 we started talking with Nokia and spent the beginning of 2003 delivering a virtual character application to be preinstalled on Nokia’s first gaming phone N-Gage, later known for its taco shape. It was the first full standalone implementation of Leiki Targeting on the device, with the character smartly picking lines based on the history of topics and preferences analysed from the user’s chatting. We also presented our mobile Reuters news application to Nokia, who liked it and we ended up making a framework contract with their new media unit to preinstall our applications on upcoming phones.
We were busy making applications for Nokia’s media unit when we received a letter stating that the entire department would be terminated by the end of 2004. Apparently they were stepping on operators’ toes. As they were our biggest customer by far this wasn’t great news to us. Also the mobile applications market in general seemed quite small, and we needed to think of something new.
Three years later things had changed again and Nokia started a new media unit called Ovi, and the mobile applications market today is quite a bit larger. But we’re not thinking of going back, as what happened in 2004-2005 didn’t kill us, but ended up giving us a stronger focus.
Move to internet focus
We had great technology but using it for mobile apps only was pretty limited. Where was the money online? In advertising of course, and there we had a lot to contribute. Mobile advertising was tiny though, and mostly in the hands of operators who were notoriously difficult customers.
So I set about selling the product to the biggest media sites in Finland. To make it easy for the customer, we’d offer it as Software as a Service. This wouldn’t be used only for advertising, that is campaigns booked by the sales department of the media company, but just as importantly for automatic content recommendation.
Every large media company had its content ridiculously scattered online. A magazine publisher would have 5 different women’s magazines, each under their own URL, with their own, different recipe database and associated editorial team, making recipes that use the same ingredients. But the user only sees content from that specific magazine, and pretty much the only way to bump in to the other magazines – all owned by the same company – is to happen to find them in a Google search.
Or take classifieds. Publishers had high-volume newspapers reviewing cars. The same company had a classified site sellling similar cars. And the easiest way to move from the car review to the car classified was to go the URL of the classifieds site, fill out a car-search form and try to find the right result from the list. A simple functionality that we had taken for granted in the early days of the company, find similar content across different content types, was proving to be very valuable to large publishers. With Leiki Focus content items didn’t need to have the same words to be contextually recommended, which made a crucial difference as products and editorial articles often shared only topics, not words.
Note above it was “I” who went selling. As a typical Finnish technology company, we were 90% product and 10% sales and marketing. So basically I was the only salesman bringing in revenue.
We started with a couple of well-known online sites in 2006 and by 2009 our customer list included most of the largest sites in Finland, including all of the top 3 ones. And we started making good profit again – it was clearly the right focus to have.
We started from the observation that user interfaces were mostly asking the human to adapt to the machine. The computer engineer had a cool logic, such as an SQL database to store features of apartments: Size, location, number of rooms, price etc. People using a service were presented with just this database layout as a form to fill in. Very logical but not very human, in other words not pleasant to use or effective in getting us the results we wanted.
We wanted the machine to adapt to the human. If you called a real estate agent to ask if he had anything for you, you’d tell him in your own words what your family was like, how much space they’d need, which facilities they’d wish are nearby for their hobbies and what kind of area it should be in terms of parks, neighbours and transportation. For example – humans think freely and cannot fit on a form. We wanted services to be like telling your best friend what you needed, not like filling out your tax forms.
A prime manifestation of the machine-centric paradigm is the position of searching as a way to access the internet. Since the start of world wide web around 1993 it’s been a lot about searching, and it still is. Want to find something? Type it in Google and crawl through the result list. But there are many fewer people who want to find something so much that they Google it, than people who are interested in it if they see it without searching.
So the search-based internet only caters to a small part of people’s interests. Automatic recommendation caters to the rest. It shows people content they don’t have time to search for, and content they didn’t realise they were interested in.
What was true in 2000 about humans needing to adapt to the machine is still largely true today, and it’s very motivating to be helping the world become more human.
Join in the fun!
Since 2010 we’ve been busy expanding our customer base outside Finland. Publishers in Europe have been surprised to hear of our new way of thinking and have signed up rapidly. We’ve been also connected with the advertisers, that is eCommerce sites whose products are being recommended. The idea is that in addition to recommendations of its own content each publisher shows the personally most relevant products from a variety of catalogs, greatly increasing traffic and revenues for both the publisher and the advertiser. For the reader basically untargeted, annoying promotions are turned into a personally useful service. So it’s a win-win-win. And unlike existing contextual ad networks, no manual work needs to be spent inserting and targeting each product.
While we continue to improve our product, our main focus going forward is to spread the message in other countries. That will keep us happily busy for a few more years.