Tomi Ahonen has collected a lot of good data about the iPhone app ecosystem and applies solid analysis to reach the conclusion that, from an economic standpoint, on average, it is a waste of developers’ time to build iPhone apps. The data is good but the conclusion is dubious.
This is a case study for the classic “How to Lie with Statistics,” in this case by accident. It’s like asserting that because, from an economic standpoint, on average, startups can’t raise money and startups fail it follows that entrepreneurs should pack up their bags and stop starting new companies and stop trying to raise money. Or that, because so few make it to pro sports and even there the injury rate is so high, high school and college students should stop trying to go pro. Or, as Freakonomics told it, drug dealers should quit the biz.
Well, there are many who feel this way but the ranks of founders, pro sports wannabes and corner dealers are growing. There are several things going on here…
Almost everyone learns about basic stats metrics such as averages and medians in the context of commonly-observed and often symmetric distributions such as the normal distribution. They make a lot of sense there. Applying them to a highly asymmetric distribution, such as iPhone app revenues or startup founder returns, without even knowing the shape of the distribution because granular data is not available, is likely to mislead the common reader. I know this because I’ve spent a couple of months studying angel returns and trying to separate reality from urban legend. More on this later.
From startups to pro sports to gambling, it is very human to pursue large statistically unlikely returns. Differently from gambling, as one invests more time in startups and sports, one hopefully gets better. And differently from sports where one gets only a couple of chances to go pro, startups don’t have this restriction. A serial entrepreneur can do many startups. In the process the entrepreneur hopefully becomes smarter, learns from past successes and mistakes and developers a bigger & better network, thereby improving the odds of success next time.
Repeat play is particularly important in games where the chances of success in any given game are very small. This includes both starting companies and joining companies as employees. If you are hot stuff, you are move likely to land a job at a hot startup.
Here is an example from angel investing based on data I’ve been looking at data recently: 68% of all angel investors lose all their money, primarily because they do too few investments. A change in portfolio size from 5 to 10 investments and 5 to 25 investments increases return at the beginning of the top quartile by 54% and 200%, respectively. (The angel at the beginning of the top quartile has better returns than 75% of angels and worse returns than 25% of angels.) The distribution of angel returns is surely not similar to similar to iPhone apps so the example is purely illustrative.
Then there is the fact that many iPhone app developers build apps while having other jobs that provide cash, in many cases more than a typical developer salary because mobile development skills are in short supply. Therefore, we should be comparing expected returns from startup equity to expected returns from iPhone app development. It is a lot easier to start an iPhone app company than many other types of companies. Therefore, we should not be comparing the return on founding equity of a startup backed by high-quality angels and VCs. Instead, we should be comparing the differences in net cash together with the return on equity that non-founders get. I don’t have great data on this but from what I know I’d argue that, outside of lasting bubble markets, non-founders and non-execs don’t make that much on equity.
Last but not least, economics is not the whole story. Some people fall in love with code and startups. It gives them a sense of purpose and a way to express their creativity. With iPhone apps where it often takes a developer or two to build a basic app, entrepreneurs also get a lot of control and a higher likelihood of calling the shots as a CEO or an exec. I have several friends who are doing iPhone apps because they can be their own boss through a combination of consulting and self-employment. They could be founders and even execs at other startups but it’s unlikely that’ll be CEOs.
To recap, Tomi’s analysis uses good data and good reasoning but it misses the forest for the trees. This doesn’t mean everyone should pile on the mobile app bubble. My point is simply that this type of by-the-numbers analysis misses the point of why so many have gotten into mobile development.
Let me know what you think in the comments or at @simeons.
I think the main mistake he made is thinking that the earnings curves linear. If you look at any other business niche there are many that fail terribly (most), some that do a sustainable business, and some that are spectacular. It would be interesting to take another industry (restaurants say) and see what the distribution is. Could be something like: 1% spectacular? 15% sustaining? 84% fail? If you did this then you would divide up the $1B in developer’s earnings at the same ratio in some kind of exponential curve going up towards the 1% group. This puts a lot more money in the 15% group. (we wont talk about the ‘lottery ticket’ 1% group). The point being that the earnings number is not evenly divided or that the earnings are not linear and thus that the mean is not interesting.
The question is, is there any reason to think that iPhone app publishers have any more or less likelihood of failure than in any other business? Seems like opening a restaurant is insanely risky since the vast majority seem to fail. Likewise for most other business types.
Anyway, maybe I’m just trying to rationalize my own behavior/choice to be an iPhone app developer…! 🙂
Yes, the return curve is neither linear nor normally distributed.
The difference between iPhone app and other startup businesses primarily has to do with the benefits and constraints of the AppStore. The balance will vary from company to company.
I am not sure that he is saying anything to VCs or the angel investor community. Simply put his message to developers and those managing developers seems to be: if you only target the iPhone you are reducing your audience significantly (<5%) AND in that market the average cost of development is so much AND the average pay-out is much less.
Most of the analysis is good. The conclusions are unwarranted.
Another interesting post. Perhaps this is just another way of saying what you and Dad said. If the cost of an entrepreneurial effort is 10 but the return is 110, you can have 9 failures and 1 success and still make money. Just because 90% of start-ups fail, does not mean that the proposition as a whole is not sensible.
I think Tomi’s point was to warn everybody from thinking: I f I write an app for the iPhone, I’ll get millionaire.Most of the application developper are no way to become millionaire.
As Simeon says, it’s true you can earn a lot of money from iPhone apps, providing your app is very attractive and good.
Both point of vue are valid. The conclusion is probably more: As in any other industries some will make a huge amount of money from a very clever idea, with a good execution and a good marketing, but this is not an eldorado where everybody will win. So careffuly think in which basket you want to put your eggs in.
… and take good care of the basket. 😉
Apply an exponential or a powerlaw distribution and you’ll get similar results, at the border of the 3rd and 4th quartile you have 6k in revenue before apple takes 2k.
So what do we have? Exponential distribution (assumption based on sales distributions of similar kinds of markets (books and music)), 1B in app sales distributed across 225000 apps. Expected value? 4444 per app. But that’s not useful, the median is 3000 if we go by these assumptions because our lambda for exponential is 1/lambda.
Scale thing up, if you’re in the top 10% of apps you might crack 10k. But to get to 60k sales revenue you’ll have to be 1 in 1 000 000 (there aren’t even a million apps out there right now).
So ignoring the free apps:
1*ln(1 – 1/4)/lambda = 1278 # first quartile
1*ln(1 – 1/2)/lambda = 3080 # median
1*ln(1 – 3/4)/lambda = 6160 # 3rd quartile
1*ln(1 – 9/10)/lambda = 10232 # 90th percentile
1*ln(1 – 99/100)/lambda = 20465 # 99th
1*ln(1 – 999/1000)/lambda = 30698 # 99.9
1*ln(1 – 9999/10000)/lambda = 40930 # 99.99
1*ln(1 – 99999/100000)/lambda= 51163 # 99.999
1*ln(1 – 999999/1000000)/lambda= 61396 # 99.9999 percentile
Add a fudge factor of an order of magnitude (+ or minus 10X) and you still pretty awful numbers.
Go google sales and powerlaw and then google music sales powerlaw and book sales powerlaw and videogame sales/adoption. You’ll see classic fits to exponential and powerlaw distributions.
Even if the distribution was log normal or pareto the results would be pretty bad.
Pingback: Finally revealed. The easiest way to make money as an iPhone Developer « excapite