Gustav’s Mindshare

If you’re anything like me (which you probably are since you got here), you’re constantly interested in how stuff works and why it works the way it does.

Taxi systems interest me, they work very differently in different countries (everyone drives their own car, 100% on provision, big firms with drivers on commission, government run etc) and I think you can tell based on the kind of service you get.

I’m also interested in the general supply and demand problem of planning taxis.

Therefore, the other day, going home from Arlanda airport, I took the chance to get the skinny on how Taxi Stockholm really works, who takes the financial risk, how much they make, how they match supply with demand etc. Pretty intesting I think.

THE SYSTEM:
There are three basic parts to Taxi Stockholm:

– Drivers:
Unlike what most people think, the drivers usually do not own the cars they drive, more often, 2-3 drivers, are scheduled on a car that is owned by what in Swedish is called an “åkare”, loosely translated as a “garage owner” or “car owner”

– Car/garage owners:
These guys usually own several cars, they take a fairly small amount of the money that their drivers charge the customer, so this is really a scale business, where the money is in having many cars making money simultaneously (almost no drivers actually own their own cars since the revenue from owning a single car is not worth the hassle according to the drivers I asked). The car/garage owner makes sure his cars are scheduled and rolling 24/7. There is a central digital board system hosted by Taxi Stockholm that handles supply and demand between drivers and car owners, where drivers can click the hours that they want to drive a car. Usually a driver always drives the same car. But the car/garage owner doesn’t have anything to do with supplying actual customers to his drivers, this is the job of the dispatch.

– The dispatch:
Taxi Stockholm dispatch is obviously what supplies the actual customers to the drivers. Interestingly, Taxi Stockholm is not a share holder’s company (an AB in Swedish) but what is called an “ekonomisk förening” losely (and lossy) translated to a financial partnership, a cooperative owned jointly by all car/garage owners (about 950 across about 1600 cars).

The dispatch is obviously the whole value of being a part of Taxi Stockholm vs driving on your own since they connect demand for taxis with supply of taxis and apart from waiting outside night clubs at closing hour or Arlanda airport (more on that below), it is very hard to predict where this demand will be.

However, while many people think that the taxis are centrally controlled by the dispatch and told where to go in order to spread out and follow demand, it is actually much more bottom up and controlled by the drivers. All cars have a computer that shows three columns and a bunch of rows. Each row is a zone (e.g. Östermalm, Södermalm or Kungsholmen) and the three columns shows:
1. The taxis in this zone right now.
2. How many outstanding reservations this zone has right now.
3. How many pre-orders this zone has coming up.

Based on this demand info, drivers try to match supply themselves by going to the zone they think will be the best bet. When they enter a zone, they log in and get a queue number based on how many cars came there before them. There are a lot of different strategies out there (kind of like fishing in the words of the driver himself) based on avoiding drunk people, dangerous areas, getting long rides (to Arlanda) vs short ones, predicting demand ahead of time etc.

THE MONEY:
So what does the money look like?

While I couldn’t get all the numbers, I got a few. Out of what the customer is charged the division looks roughly like the following:

– VAT 6%
– Driver 35% (about 26% net), i.e. 100% performance based salary, no fixed part.
– Car owner ?% (unknown, but the drivers say it is a pretty tough business, let’s assume maybe 35% as well, which needs to cover the vehicle cost before income tax)
– Taxi Stockholm dispatch staff and orgnaization cost ?% (maybe 24% then)

On a good weekday 12 hour shift, a driver makes about 3000 SEK
On a good weekend 12 hour shift, a driver makes about 6000 SEK
On a good month a driver makes about 30 000 SEK gross/20 000 SEK net (which implies working at least 6 nights a week and probably several full weekends)

Some observations: The incentives to drive illegal cabs are really high as you can charge customers roughly the same per minute/mile but make almost four times as much net per minute/mile. On the other hand, you have no dispatch to feed you business, which means you can really only do this successfully around night clubs at closing time where demand is obvious (most hotels and airports have pretty effectively gotten rid of illegal drivers). The lack of a supply/demand function to make them competitive thus means that illegal cabs aren’t really a big problem for the taxi industry financially.

THE CARS:
Actually, Arlanda airport has a bigger impact on the type of car that a car owner buys than the general car economy like repairs cost of fuel etc.

Arlanda airport is obviously a very big supplier of customers and they have a taxi system that very strongly incentivize environmentally friendly cars over others. Basically, when drivers arrive at the airport (or have dropped customers off) they have to queue at a site a bit away from the airport instead of driving up to a gate. There, they get a queue number based on the environmental points their car has. A car can get up to 100 points (the passat is apparently the top scoring car at something like 85 points). If you have a low score you can be sitting at the queue site for many hours and might as well not bother (i.e. drive back to Stockholm empty to find new business).

That’s it, I hope you learned something you didn’t already know…

The model that is quickly emerging as the de-facto model for location updates is the so called check-in model.

Contrary to the  always-on, automatically updating friend tracking model used by services like Google Latitude and Loopt, that has been prophesized and expected to take off for years,  the check-in model used by services such as Gowalla, Foursquare and MyTown lets a user manually and deliberately “check in” to an exact/discrete location and actively update friends and others about it.

You could argue that this is not really a new model at all, that it is in fact exactly what Dennis Crowley (now of Foursquare) did over SMS and Wap with his previous company Dodgeball (acquired by Google in May 2005). But for one reason or another, Google did nothing with Dodgeball and it seems like everyone went back to focussing on the prophecy of the always-on friend tracking model for the subsequent years.

Why haven’t the friend tracking services taken off?

• Until recently, you could argue that it was because there was not critical mass of technically capable devices to get the network started. With the adoption of GPS enabled iPhones, Androids, Blackberries and Nokias etc, that argument doesn’t hold anymore.
• That leaves the argument that people are simply not prepared to share their location with a bigger group of people, it’s too private. Well, the screaming adoption of these new check-in services, that are actually very public, prove that this was not true either.

To me, it is clear that it was the location model that was wrong. So what is the difference?

What the check-in model solves:

1. Privacy: With the tracking model, as soon as I friend someone, I will have to go through the mental exercise of thinking if I REALLY want to share all my future locations with this person, which is very hard to do since I don’t know what they will be doing in the future. Alternatively, I’ll have to set different resolutions on different people and remember to turn some or all of them off when I go certain places. The check-in model obviously solves this by way of the model itself, pushing an update is always a deliberate event.
2. Accuracy: It may seem intutive at first that location information doesn’t have to be very accurate to be useful, This also rhymes well with the fact that you expect people to be more concerned about their privacy the more accurate it is (the reason Google Latitude offers city level-only filters). In reality though, I find it to be totally the opposite. Actually, to be useful at all, the location information has to be super accurate, down to single meters and it has to include altitude. Looking at a dot on a map with an error of  +/-100 meters or so (which is best case, when you’re outside in GPS view) doesn’t tell me very much at all. I want to know if you’re in a ski slope, a particular café/pub or a department store and which floor  and shop of the department store you’re in. The technology to automatically deduce this with sensors and local directory services simply isn’t there yet, and won’t be for a long time.As low-tech as it may seem, the check-in services solve this in a beatiful way, by using the device’s sensors to present a list of adjacent exact/discrete and canonical locations and then letting the user do the “last mile” positioning manually, by choosing one of these. This beatifully solves the problem of getting infinite resolution in both x,y and z axis. It obviously requires a huge amount of discrete locations to check-in to, but as these services let’s you quickly create such locations, if you find yourself on the corner of Broadway and 5th and it doesn’t exist as a discrete location yet, you can simply create it.
3. Context: Even if the resolution deduced by the sensors would somehow be good enough. If I don’t know that particualar spot myself, an X on a map somewhere doesn’t tell me anything about where that person really is. Is it a hotel, café, store, ski slope etc? Of course, clever mapping to local directory services can attempt to solve some of this, but it will take a loooong time before they will automatically tell me that “Eric is at the skate board ramp in central park” (as opposed to him being at the ice rink 10 m north of it). Not to speak of the complete loss of  Z axes when you go into a 10 story building. The fact that the discrete locations are man made, (but still typed, making them machine readable) means that I get clear text context to that specific location apart from just it’s coordinates.
4. Update frequency: A constantly tracking service has the generic problem of not knowing when a friend of mine does something interesting or news worthy that I might want to know about (which is likely a tiny fraction of the time that I’m tracking him/her). That results in me constantly having to check the service and all my friend’s locations, making the likelihood of any actual serendipity extremely small. The model of deliberate updates solves this by it’s design and allows for both pushed updates to friends and posting to social networks.

As the fit with services such as Twitter and Facebook are so obvious, I’d expect to see some acquisitions fairly early in 2010