By PETER LUNDEGÅRDH | peter.lundegardh@di.se | Torsdag 30 juni 2016
He is the mathematician who traded ideas to Richard Branson in exchange for lunch and invented algorithms that helped Barack Obama to success on social media. And he happens to be the only one who can predict if commuter trains will be delayed or not.
I solve real problems, says Wilhelm Landerholm.
» You have probably, at some point, got stuck on while boarding an airplane, wishing it could go faster. Guess what, Wilhelm Landerholm has an algorithm for it.
With queue theory as his forte he can determine the most efficient way to fill the seats in a short period of time.
Either you put passengers in the right order already at the gate, or you can do it randomly. My suggestion is to give the travelers colored tiles and drop one colour at a time, he says.
FACTS | WILHELM LANDERHOLM
Age: 38 years.
Family: Married, two children, 6 and 4 years.
Living: In Täby, north of Stockholm.
Background: Mathematician, statistician and political scientist. In 2003 he published the e-book About Opportunities, and two years later the book Forget the customer with his own publishing house. Runs the company Queue. Last year he was well known after launching a forecast service for commuter trains via an app called the The Commuter Forecast.
Spare time: Play with his kids, preferably in nature. "I’m a Gaisare at heart and soul" (Gais = football team from Gothenburg). Loves licorice and is most comfortable in fleece sweaters.
Media habits: Twitter. "News junkie", Dagens Industri, Washington Post and The New York Times.
FACTS | 7 PROBLEM SOM WILHELM LANDERHOLM HAR LÖST:
1. A tool for aid organizations that predicts refugee flows.
2. An algorithm that reveal delays in commuter traffic – two hours before it happens.
3. Evaluating the success of a online dating date, based on how fast the woman logs back on to the site.
4. Maybe not a real problem, but he spends his free time predicting who will pick up trophies at the Academy Awards.
5. Sorting out what triggers people to retweet messages on Twitter. Used in Obama's reselection campaign.
6. He helped Richard Branson's Virgin Airlines to simplify the check-in process.
7. A system that alerts customers when products bought online get stuck in transport.
Wilhelm Landerholm is one of Sweden’s few real so-called data scientists. An expert in analyzing large amounts of data and in how to get a clear picture and draw concrete conclusions from statistical noise – through mathematical algorithms.
As the world us generating more and more data – just think about everything you share with your mobile or on Facebook – there are more opportunities than ever to turn this into business.
Retailers and global corporations are turning every stone for that kind of expertise. With his knowledge and ability Wilhelm Landerholm is sure to be one of the people that will help businesses grow in our data driven society.
Working in his "man-cave"
He is proud owner of the company Queue, a laboratory that tailors business concepts through data analysis. The office, or his "man cave", is located in Täby with two rooms and seating for eight people. Yet, he works alone.
The first thing you notice walking in are five clocks, showing the time in Los Angeles, Stockholm, New York, Mumbai and Seoul. But it’s dead silent, no tick, no tock.
The idea was to keep track of the time in my customers timezones, but they ticked extremely loud. I got crazy after two days and unplugged the batteries, Wilhelm says.
Along the walls there are shelves with neatly organized document boxes and large cartons full of sweets - coffee caramel is the obvious favourite. There are even more boxes of chocolate next to some of his favorite books. It is quite a mix of literature, everything from Douglas Adams The guardians of the galaxy to Vilhelm Moberg's The new builders. There you will also find the Start Your Own-handbook
Swedish companies are bad at taking advantage of big data’s opportunities, according to Wilhelm.
There is a high demand from large American groups or companies that work in that environment, such as Schibstedt and Kinnevik, he says and bites his tongue.
If confidentiality agreements could speak, the ones right behind him are screaming.
I have customers who basically are completely dependent on me and they would sue the shit out of me if I told who they are. I have to check who I'm allowed to talk about, but Aftonbladet and Tele2 are probably fine, it's been a while since I delivered algorithms to them.
A quialified guess would be that some of his clients are major retail chains in the UK. Some are perticularly nervous and require that their data is analyzed only on a computer that has never been connected to the Internet.
It's no problem. I'm working on good computers and with a computer power many people only dream about.
For his part, he has kept a low profile so far. Talking about himself, like now, is rare.
I'm really flying under the radar. It's very rare that I meet the end customer. My private zone is pretty big. I'm almost not ready to talk to you, really.
However, if the stereotype of a mathematician is an introvert and square person who loves numeracy, then Wilhelm Landerholm is the opposite.
You will notice him walking in to a room, but in a warm and charming way and he prefers fleeced sweaters over a suit.
Laughing loudly, sometimes in falsetto, and chatting like a popcorn machine. When a grain pops, a chain reaction is in motion and he can talk for hours. Shake the pan with a question or two and he will continue to pop. His answers could really go in every possible direction.
I know that I am like that, and it annoys me. Mathematicians are usually very stringent, and in some way that´s the ideal for me.
He lacks enough ego to admit that he received 0.1 points in the national university test.
There were two columns to fill in the answers. I wrote my answers in the wrong column. You should probably not get a higher score if you can’t read the instructions ...
But he is way more accurate when he describes what a data scientist actually does.
There are many out there who use the title without being close to using the right methods or analyze correctly, he says, pulling out a small circuit board from the jacket.
I usually bring this to customers, like food for thought. It's good for them to hold on to while I’m talking.
Big data isn’t necessarily ones and zeros in server halls in the size of hangars. The valuable part of the information can be handled on a small Raspberry Pi computer that costs a few hundred pounds at your ordinary electronics store.
Much of what I do is very abstract. It's about automated systems, but nothing you and me as customers notice. Sure, you can see the price tag, but from there to the underlying pricing strategies, it is a big, big leap.
I depend on combining a number of techniques, including something called Random Forest.
When I ask him to describe what Random Forest is – with the addition that he should explain it as simple as if I were his grandmother – he answers sincerely:
I've had many questions, but never that one! Lucky you did not ask me to explain to my mom, because she has a licentiate degree in inorganic chemistry.
The big point with Random Forest is to randomize observations and variables and make a decision tree on your sampled data. It allows you to get more trees – a forest. The point is that most of the trees will rate the data correctly, and that the errors appear in different places. This allows you to get a fairly robust estimate compared to just doing it once.
Delayed mail and web dating
But a deep diving in mathematical terms is not really necessary here. More importantly, pretty much anyone who reads this article have used, exposed, or benefited from a technical solution that Wilhelm Landerholm made.
He’s been innovating since 2003 and is so far up to 1,291 models, of which 250 are used right now. Here is a small selection from the models he throws out (at a pace only a stenographer could catch up with):
"The train brain". The train delay model to Stockholmståg. | "I built a model that can predict train delays two hours in advance. As far as we know, the model is world-unique when it's actually implemented and ticking and going. "
Centiro. | "It’s a model that alerts when a mail shipment, delivered from an online store, will be late."
Tele2. | "Several different things, but perhaps the best example is the model for creating customer clusters / product development segments."
Function C2. | "A simple method of measuring how successful a online-date is based on the time it takes before the girl logs in again. But it's so old that I don’t know what's left of it. "
Obama and Twitter
The commuter train tracker became a global news story when released a couple years ago and Wilhelm Landerholm was invited by BBC’s technical magazine Click to describe it.
After the program was broadcasted, my phone was in chock. I suddenly had 591 unanswered calls. One of those who called was a rail operator in Japan.
He made a taxi optimization model at the same time as the people behind Uber began to launch their idea and has carved together features that predict refugee flows, a tool for aid organizations.
In connection with Barack Obama's presidential campaign, datawizards were flocking to cast their spells. Wilhelm Landerholm was one of those who contributed with a feature linked to Twitter.
It was about sorting out what people respond to and what triggers them to retweet messages. If you know that, it's possible to have them send out a similar message, though from your own candidate.
In addition to the juggling of equations and analytical tools in his entrepreneurial business, he has also been involved in the start-up of Ecoguard, a energy measurement company, which currently has around 40 employees and has a turnover of 6 million euros.
The model can also be used as a negotiation weapon – how much to pay the actors in proportion to how much the movie will record.
The model has 21 parameters or measurement points such as action, concept and genre. The secret lies in how these estimates affect each other. He does not want to tell who use it, but the movie records hanging out at his office point to the fact that it's probably 20th Century Fox.
And who would walk home with a statue from this year's Oscars gala, Wilhelm Landerholm already knew in advance. Kind of.
Me and some friends have a game where we try to create as simple models as possible. We count the twelve most important statuettes, and everybody usually has ten-twelve-ish correct. The one who can build a model that is exact as possible with as little data as possible wins. My best performance so far is 20 correct winners out of 24 possible.
I make the mistake of asking how it works – it's like opening the gates to a dam. Out comes industry slang like bagging, logistic regression, classical regression (which seems to be a bad prediction technique, but good for capturing variety), Kalman filters, time series analysis and Markov chains (something Wilhelm Landerholm is obviously particularly weak for).
A bad businessman
In a more simple way, it seems to be about estimating and evaluate what other prizes the films or actors have won, or how many views the videos got on Youtube.
Personally, I often combine several techniques. But I really appreciate creating clusters. With good grouping of data you can create functions that have lower variations and deviations. I also like distance, in other words, vectors, where we describe an object's characteristics and then consider the difference, the distance, between these objects. That's basically what most cluster technicians do, but what I mean is the correction. If you are based on expectant model, you can add a bias to change the conditions.
He always carries a notebook full of ideas. Or improvement suggestions that he calls them. In plain text: equations with more letters than numbers. All written down with very neat writing and sober drawn diagrams and graphs.
I constantly think of new models that can solve problems of all kinds. If you ask my wife, she’d say it's tiring. I am always looking for a thought behind everything and can think about death date or how best to arrange a buffet breakfast in a hotel.
Wilhelm Landerholm tries to discover patterns. He structures reality and tests with different assumptions: "What would happen if you ..."
Even though he already in high school tried to sell biscuits and coca cola from his locker, he sees himself like a bad businessman.
I have not earned any big money. One of the models I have drawn has earned 23 million euros to the company I collaborated with. I received 2800 euros for that algorithm.
However, his brand new Volvo XC90 is parked outside the office. The company is hardly going bad.
Classed as superintelligent
Wilhelm Landerholm grew up in a middle-class part of Mölnlycke, outside Gothenburg. His parents were academics. Parents of some of his friends called him "Q" after the tech genius in James Bond since he rebuilt his bikes and designed soapbox cars. Math was no favorite subject. It was not until he went to Örebro University his love for math flourished. A school were Sweden’s National Statistics Agency happened to be involved with.
I wanted to go my own way after high school and chose Örebro because I did not know a single person there.
A teacher at the statistics institute discovered his potential and made sure that he improved his jedi skills in the subject with the help of a teacher at a university in Wales.
I received an introductory letter where I was described as superintelligent - which I’m really not - but I received special treatment. I have accumulated a lot of university credits. The most difficult of all courses was five points in management policy.
When he set out to find a job, it was more of an uphill battle.
I applied for many jobs as analysts in logistics and optimization. But I was probably too honest in my answers when they asked me what I'm good at, because I always replied “write exams”.
During the study period in the early 2000s, Wilhelm Landerholm sent a list of suggestions for improvement to Richard Branson with the offer that if any of them were introduced to the supercontractor's company, he would be invited to lunch.
There were three ideas, one of them was about how to make check-in functions better at Virgin. It was a long time ago, I do not even remember what we ate. Branson was barely present.
Now Wilhelm Landerholm has the privilege of choosing assignments.
I say no to about two hours of three. Often it is a project where there is no possibility of finding a solution in the given time span.
Very few companies have problems so big that a small computer can’t solve them. Most of the times it does not have to be complicated at all. A dispatching company that turned to him for help with arranging the queues at the loading dock in connection with loading and unloading. That was fixed through a webcam.
If the drivers in their mobiles could see what it looked like on the site, they could plan better and park only when there was a free cargo area. All problems do not require complicated solutions.
Underneath the kind and warm surface there is a more sticky side. Wilhelm Landerholm is getting angry with companies who treat their customers bad. He has even written a book about the phenomenon called Ignore the Customer. The thesis he thinks is that companies must look at their customers the same way rock band treats their fans.
Some companies he made fun of on Twitter are PostNord and Broker Statistics. He sees Facebook as careless in some respects, but has no problem with Google.
I've seen how Facebook – in real time – can sort out which travelers on bus 611 to Danderyd who are dog owners based on information about how people moved and what pictures they shared. To me, Google is more sensible. But maybe that's because I´m in contact with good people in that company.
The most exciting experiment
Some are exposed to his sense of humor. Prior to the last Swedish election, he could not keep his fingers away and started two Twitter accounts. One with a social-democratic profile and one Liberal profile. He let Twitter bots generate tweets and copy the partisans' posts.
That was the funniest experiment I did. The Social Democratic Account received 2,000 followers, the Liberals 800. Then I changed the Twitter profile so that the Liberals got social-democratic tweets and vice versa. I regret that I deleted those accounts without documenting the traffic.
The experiment managed caused some annoyance for several politicians, not least a Liberal in Solna.
Twitter data is fascinating. For a statistician, there is much to do with what is said there. It is possible to chart very much and try things.
Wilhelm Landerholm has, among other things, a weekly competition on his Twitter account where people present problems that need a solution. The best problems he writes on a whiteboard in the office. Right now you can find "Throw less food" and "How to like yourself?" on there.
I like those questions. It can’t be impossible to identify which ingredients are in the food that is thrown away most for example.
What is the relation between what is served and what is being thrown? Wilhelm Landerholm is able to find patterns in everything and thus an opening for a solution.
But what can you do with a emotion like what you feel about yourself?
The question is so abstract that it becomes mathematically interesting. Are there any data you can use to find a solution? Are there any data in what people post on Facebook or anywhere else? And how do you measure it?
A few years ago Wilhelm Landerholm tried to create a standard for statistical surveys. The idea was to have some kind of five-star rating system. Inexperienced investigations reveal a special glow in his eyes.
When I see things like ’One in three women get better hair from this shampoo’ I wonder what happened to the other two thirds. Did they drop their hair?
Can reveal Ikea's secret
In a world where large amounts of data can be exploited, there are huge profits to make from automation. And with more and more being available online opens up opportunities to find things out. The furniture giant Ikea's best kept secrets are how much every department stores sells. But to figure it out is basically simple, according to Wilhelm Landerholm.
The price of the products is known and online, you can see inventory balance. I just need to follow how the stock changes for hundreds of products to get a true picture. If I also want to put sales in relation to the number of customers, just count cars entering the parking lot. You don’t even have to do it by hand, you can set up a small Raspberry Pi computer and a sensor at the entrance.
For his own sake, he reluctantly uses bankcards. But than aging, he doesn’t have a wallet.
I'm probably one of the last ones who carry cash in my pocket. Cash is good. You get the right sense of how much something costs when pay with a five hundred note compared to the vague feeling of scrambling a code when shopping with cards.
If I had a lot of money, I would invest everything in companies dealing with zip technology, compression of large amounts of data, and how it can be done safely with encryption. There will happen a lot with that following years to come.
In addition to the consultancy assignments, Wilhelm Landerholm has an idea of creating "self-running companies".
I want put everything I made together and start data-driven companies. Each Raspberry Pi on this table can become a limited company. Many aspects of a company is very possible to automate.
He uses the Thai food shop near the office as an example. How many customers, what portions do they buy, which dishes are the best, what ingredients are needed? Everything can be entered into a data model. Get customers on Thursdays? Use the customer information available and send discount offers or inform about a special dish. Follow up. Build on.
One problem is that legislation in this area is lagging behind. Because what happens if the company doesn’t pay taxes?
I don’t always think patents are the best solution. It's better to tell everybody what you are doing and then hurry to get ahead of any competitors ready to copy the idea.
Wilhelm Landerholm sees a galaxy of possibilities in front of him, with one crucial detail.
It does not matter how good models I build in my chamber if they can’t be implemented in reality.
Other values in life
After spending the last fifteen years learning how to build models that handle large amounts of information, it is now time for Wilhelm to start capitalizing on the knowledge. His company is profitable, but in recent years, turnover has been about 100 000 euros.
He has formed strategic alliances with Kärnhuset, Novus, Norstat and Prime / United Minds.
I have a very comfortable life where others sell me as part of their projects, and I become part of their solution.
But as well as attracting big money, there are other values in life. For example, to spend a lot of time with the children and to hear his daughter stand up for herself in school:
When I picked her up, there was a little guy there – that kind who seems to need a hug – and says his father is a boss. And my daughter replied, “My father has no boss!”
