We have started developing the technology that estimates app download numbers in autumn 2010. This technology is part of a larger question we research, namely how to build an AppRank similar to Google’s PageRank. We opened it up in beta in late January and started communicating the data in a standard report version from the end of April onwards.

We made a number of custom reports to a variety of clients ranging from carriers to device makers, chipset makers, game and app developers, investors and finance companies. For app developers and analysts we also provided a lot of our data for free, via search.xyologic.com.

This is all to say that over the course of a year we have striven very hard to make the app download data valuable to you all and still continue to do so.

As of September, this is the statement of errors we hear about from our users and clients:

Download estimates of individual apps

- Android: very good
- WP: very good
- iPhone & iPad; good. We hear about a statement of error of up to 50% for app downloads for monthly app downloads below the 50k range. So if an app actually has 20.000 downloads, our estimation may be from 10.000 to 30.000.  This statistical error  decreases for apps with more actual downloads.

Download estimates of individual app publishers

- Android: very good
- WP: very good
- iPhone & iPad; good. We hear about a statement of error of up to 30%.

Download estimates for platforms and countries

- Android: very good .
- WP: very good.
- iPhone & iPad: very good.

Sometimes we get asked why we have chosen to use the method of estimating app download numbers as opposed to other methods (for example the tracking done by Appannie). The key reason for it is that we want to provide a compass to understanding the dynamics of app economy publicly and in a transparent way. By using estimated downloads we are able to provide data for discussion without disclosing data provided to us by individual users. This way we are also able to estimate downloads of all existing apps, rather than only for a small group that would sign up to our program or would use our analytics API.

If you want to share cases where our estimates diverge from factual numbers, do send me an email (matthaus at xyologic dot .com). Every such case helps us tune our algorithm and improve accuracy. Alternatively please leave your comments below or find us on Quora.