Multi-homing, a strategy where a developer is publishing products for multiple platforms, is studied in this paper using descriptive analysis. The data of over 850,000 applications and their developers were collected from Apple App Store, Google Play, and Microsoft's Windows Phone Marketplace. The developers publishing for several software ecosystems were then identi ed. The ndings show that multi-homing is utilized by only a small set of developers. In addition, the applications available at several platforms do not seem to di er by the type or the popularity from single-homing applications.
During the last few years, software ecosystems have been emerging as a signi cant part of the mobile domain. The marketplaces of these software ecosystems | e.g. Apple App Store, Blackberry App World, Google Play (formerly Android Market), Nokia Ovi, and Windows Phone Marketplace | o er currently hundreds of thousands of applications from tens of thousands of developers and the ecosystems are in a tight competition. Not surprisingly, the size of the marketplace and the application o ering have been used as an argument for the superiority in the struggle of mobile ecosystems. For example, Apple used the phrase \There's an App for That" in advertisements of its products in 20091.
Application stores are working as two-sided markets, where the ecosystem's
orchestrator is enticing both the content vendors and users to commit to the
mobile ecosystem for achieving a positive feedback loop [
In this paper, we present the rst results of a study focusing on multi-homing
in the mobile software ecosystems. A multi-homing developer here is understood
as a developer who publishes its products or services for multiple mobile
ecosystems. This kind of publishing strategy is not new and it is known to be utilized
in the mobile domain. However, very little is known for multi-homing, and its
e ects on applications, developers and ecosystems. Although Idu, van de Zande,
and Jansen [
Thus, the main research objective of the on-going research is:
How does the multi-homing a ect the mobile software ecosystems?
The mobile ecosystems are a high competition area and the application o ering
can be used as a tool to distinguish from the competitors. However, if
multihoming is a common strategy for developers, the di erentiating is hard. To
understand the popularity of the phenomenon, we use quantitative analysis to
study the three major application stores. Especially, in this paper we address
the following questions:
{ How many application developers currently use the multi-homing strategy?
{ What type of applications are published for multiple platforms?
{ Are multi-homing applications more popular (in the number of downloads and
the number of users' ratings) than applications without multi-homing?
We assume that multi-homing strategy might increase the popularity of an
application. The e ect might be bi-directional, since either an application is
multihomed and therefore gaining more downloads or, due to popularity in a single
platform, it is ported to other platforms. Whatever the reasons may be, it is
expected that it would be an e cient strategy for a developer to publish the
product in multiple marketplaces. This argument is based on the low cost of
porting applications from platform to another [
In order to assess our research question, we collected data from three dominant mobile application marketplaces: Android operating system's marketplace Google Play; iPhone, iPad, and iPod touch's Apple App Store; and Windows Phone Marketplace. As a total, we parsed the data of more than 850,000 applications. The gathered data was then analyzed for identifying the developers who publish in more than one platform. Furthermore, we identi ed the multi-homing applications and, from these we investigated the types of these applications.
The ndings show that a relatively small set of developers utilizes multihoming publishing strategy for a small set of applications. Moreover, the application types and popularity are similar to the single-homers. As well, the results serve as a starting point for future research on the topic.
The paper is structured as follows. Section 2 presents brie y the related literature of software ecosystems and multi-homing studies. The research method used is described in Section 3 and the results are reported in Section 4. Section 5 contains the discussion and the last section concludes the study with suggestions for future work.
Although software ecosystems existed already in the `90s [
In addition to the software ecosystems, the concept of `mobile ecosystem'
has been recently discussed [
The application stores in the studied ecosystems are following the theory of
two-sided markets. The concept is de ned as an economic platform where
bene cial cross-group network e ects exist [
Two-sided markets can be extended by the concept of `multi-homing' [
Multi-homing describes a situation where several competing two-sided economic platforms exist in the same market and the two sides of the market are free to exist in several platforms. As an example, a software developer is utilizing multihoming publishing strategy when o ering products in both Windows Phone and Android ecosystem. Similarly, a user is multi-homing when he adopts mobile devices from several platforms.
Multi-homing in two-sided software ecosystem's markets has been studied
before. Burkard et al. [
We chose a quantitative research method to study the phenomenon of multihoming in mobile software ecosystems. First, we collected data from all marketplaces. This data was used to identify multi-homing applications and developers. Second, we analyze the types and popularity of multi-homing applications. Based on this analysis, we further discuss how the publishing strategy a ects the ecosystems.
In order to collect the necessary data, we implemented a set of small scripts (with Java programming language) that gathered the application data in two phases. In the rst phase, the scripts collect the unique identi ers of the applications. In the Apple ecosystem, the identi ers were available directly from the marketplace. From Apple's store, we included all mobile applications found. The script went through all categories listing and stored the applications identi ers into text les. In the end of this rst phase, duplicates were removed from the list.
In two other cases, third parties' listing services had to be utilized in the rst phase due to the lack of easily readable lists in the stores. For Google Play, we used AndroidPIT2 and for Windows Phone Marketplace, we used Windows Phone AppList3. Furthermore, we collected directly from Google Play the most downloaded applications of each categories to ensure that the most popular applications are included into the data. These lists were then combined and the duplicates removed. We consider it noteworthy that the list of Google Play's most downloaded applications did not add new identi ers into the dataset. For 2 AndroidPIT | http://www.androidpit.com 3 Windows Phone Applist | http://www.windowsphoneapplist.com/ the Windows Phone ecosystem, we used only the third party's listing when collecting the unique identi ers.
In the second phase, for each identi ed applications the scripts collected various attributes from the applications' public web pages at the marketplaces. The gathered attributes varied depending on the stores, but for all marketplaces, at least the name, price and the developer of each application was collected. When possible, the US versions of the stores were used. If an application was not available at a US store, other the stores of other countries were investigated too; however, the scripts were run in a server located in Finland, and thus some applications may have been ltered out by the marketplace due to the IP address used. If an application was not available, the script wrote a note to the text le and the application was assumed to have been removed from the marketplace.
The information was gathered in the end of March (Google Play and Apple App Store) and in the beginning of April (Windows Phone Marketplace), 2012. The data was stored separately for each marketplace to a CSV le. In the analysis, we used IBM SPSS Statistics Version 20 and Microsoft Excel 2010. In order to identify which developers publish on several ecosystems we implemented a script (with Python programming language) that compared the developer or publisher elds from di erent marketplaces trying to match them against each other. The algorithm utilized exact and approximate matching. In exact matching `Rovio' (used by the publisher of `Angry Birds' in Windows Phone Marketplace) and `Rovio Mobile Ltd.' (used in Google Play and Apple App Store) are considered as two distinct cases. In approximate matching a certain amount of di erence in the names is allowed. The reason for using two matching rules is that the exact matching forms the lower bound whereas the approximate matching is closer to the actual number of the popularity of the multi-homing applications.
The evaluation of the multi-homing application types was done by analyzing their categories in the marketplaces. All studied marketplaces o er a categorization for the applications, however, each one uses their own set of categories. For this study, we formed a general categorization and mapped the marketplaces' categories to the general one. The general categorization tries to capture the type of the applications, not the exact content or the target group. For example, a majority of top `Sports' applications o ers news and information of a certain sport, league, or club | the same basic functionality is found also in `News' and `Weather' applications. It should be also noted that Android is the only platform with the speci c `Personalization' category due to its highly con gurable interface. The general categorization and the mapping are shown in Table 1.
In total, we gathered information of 857,411 applications from the three application stores. The descriptive statistics | number of applications and developers, General Categories Games Health Books Music News Business Travel Social Finance Finance Education Education Entertainment Entertainment Lifestyle Lifestyle Photography Photo & Video Personalization
Apple Google Windows Phone App Store Play Marketplace Games (19) Games (8) Games (14) Health & Fitness, Health & Fitness, health + tness (3) Medical Medical Books, Catalogs, Books & Reference, books + reference (4) Reference Comics,
Libraries & Demo Music Media & Video, music + video
Music & Audio News, Newsstand, News & Magazines, news + weather (2), Weather, Sports Sports, government + politics (4),
Weather sports Business, Tools, tools + productivity,
Productivity business Business, Productivity Utilities Travel, Transportation, travel + navigation (8) Navigation Travel & Local Social Networking Social, social
Communication Finance personal nance Education education, kids + family Entertainment entertainment Lifestyle, Shopping lifestyle (7) Photography photo Live Wallpaper Personalization
Widgets percentage of applications that are available for download for free, and percentage of developers with only one application | of the ecosystems are presented in Table 2. Apple's ecosystem is the oldest and clearly the largest one. Windows Phone Marketplace is rather young and small with less than 100,000 applications o ered by only a few thousand developers.
It should be noted that Windows Phone Marketplace contains built-in
support for so called Free to try -model where a user can install and use the product
for free, but the premium content is a subject to charge. In App Store and Google
Play, this feature is not directly supported and thus the developers often o er
both a free and a paid version of the same product (see e.g. [
We normalize the Windows Phone Marketplace application count by
calculating `Free to try' applications in the marketplace twice (i.e. once as a free and
once as a paid one). With this, the total number of applications in Windows
Phone Marketplace increases to 72,748. Despite the di erences in the
measurement methods, it is clear that Windows Phone Marketplace is, at the moment
of data gathering, far behind its competitors in terms of number of applications.
However, the marketplace is rapidly growing [
The applications distribution into the general categories (see Table 1) is presented in Table 3. The similarities are noteworthy and the small di erences may be due to the di erences in the marketplaces' original categorizations. For example, Windows Phone Marketplace has the smallest share of the `Music' category, however, it is clearly leading in the `Entertainment' category.
We approach multi-homing by identifying both developers and applications present in at least two ecosystems. In developer matching, we utilize exact matching, i.e. the names have to be exactly similar, including additional words such as `GmbH' or `Inc'.
When identifying multi-homing applications, we are pairing the applications from the di erent platforms and calculating the unique applications. In application matching, we use two matching strategies for both applications and developers: exact and approximate matching. Exact matching requires that the application names and the publisher names are the same (case-insensitive). Approximate matching requires that the application names are exactly the same, however, a certain amount of di erence is allowed in the publisher names. In the approximate matching we used Python's sequence similarity function4 in the similarity metric. The exact matching gives the lower bound of the multihomers, and the actual value is estimated with approximate matching.
We tested the approximate matching with several sequence similarity thresholds. The correctness of approximate matching was evaluated only by visual examination; however, we clearly found pairs with small di erences in the names such as the lack of post xes like `Inc'. Based on this analysis, we found the threshold of 50 % was the lowest value with no signi cant false positives.
With exact matching of developers, we found 14,451 (6.8 % of all unique) developers publishing in at least two marketplaces. Figure 1 presents a Venn diagram of application developers in the three studied ecosystems. Most of the multi-homing developers (12,338) publish for Apple App Store and Google Play. Approximately as many multi-homers publish only for Android and Windows Phone 7 (796, 0.37 %), and only for iOS and Windows Phone 7 (725, 0.30 %). Only 592 (0.28 %) developers publish for all three ecosystems.
The results of the application matchings are presented in Table 4. With the exact matching, we found 15,205 (1.83 %) multi-homers out of the 828,763 unique applications. Only 268 (0.032 %) applications were found in all three ecosystems. Although the number of found matches increased when the similarity requirement was loosened in the approximate matching, the growth was relatively small and the overall ratio of multi-homers out of all applications remained low. Even at the most relaxed case of approximate matching, without clear false positives, the multi-homing applications present only a mere 3.2 % of all unique applications. As a consequence, multi-homing publishing strategy is used only by a few developers and typically for a small set of applications.
When comparing the results of the developer and application matching, it was noted that several multi-homing developers had no multi-homing applications. A small set of these were examined by hand. Some developers o er clearly di erent applications for di erent marketplaces and some developers had published the same application with a di erent name. In few cases, it was not clear whether the developer was the same or just the name, e.g. two John Smiths, one publishing in Apple App Store and another in the Google Play.
The distribution of multi-homing applications into the categories is rather similar than that of all applications (Table 5). We assume that the di erences less than 2 percentage points can can be explained with statistical error. Thus the only signi cant di erences were in `News', `Entertainment', and `Lifestyle' categories. It was surprising that multi-homing is not popular in any single category, e.g. we were assuming that the `News' applications are easily multihomed.
The popularity of multi-homing applications is assessed in gures 2 and 3. Figure 2 depicts the distribution of the installation proportions of multi-homed applications in Google Play. Because Apple App Store and Windows Phone Marketplace do not reveal the installation numbers, the popularity of the multihomed applications is illustrated through the distribution of user ratings in Figure 3. From the gures, it can be seen that the application popularity proportions are similar for all and multi-homing applications.
Although the overall number of applications available at the mature
marketplaces is large, the number of multi-homing developers is fairly small. This was
expected as the ecosystems seem to contain lots of developers who are
publishing just for fun or are lacking a clear monetization plan [
Based on the gathered data, it seems that a tiny fraction of applications forms the majority of the content used by the consumers in the marketplaces. These applications are usually published by professional developers and they are often o ered for several operating systems. These developers are becoming more and more critical to the whole ecosystem. Recently, when it was reported that the sequel of a popular mobile game might skip the Windows Phone 7, market analysts suggested that this might hamper the orchestrator's e ort to attract gamers5. Therefore, the lack of popular developers and applications, such as `Facebook' or `Angry Birds', might be a major handicap.
The application ecosystems are primarily tools for ecosystem orchestrators to leverage their main businesses. The intentions are not openly discussed by the orchestrators; however, understanding them is bene cial in assessing the ecosystems. The main objectives seem to be di erent in all three studied ecosystems: App Store seems to aim at increasing Apple's mobile device sales that have particularly high margins. Google Play seems to extend Google's on-line advertising business into mobile devices. Windows Phone Marketplace and ecosystem more broadly seem to aim at protecting Microsofts software sales in the ongoing mobile-desktop convergence. Generating revenues directly from application sales seem to be, at least for now, a secondary objective. The primary objectives of all three ecosystem orchestrators seem to be to maximize the number of users in their mobile ecosystems. The application stores can contribute by maximizing both the number and quality of available applications, thus making the users value proposition more lucrative. The key to this is boosting the developers' and content providers' interest on the application store.
In the interviews of commercial application developers, the nancial aspects
were highlighted as one of the major reasons to select an ecosystem [
Control and governance practices are the tool to manage this balance. In practice, the orchestrator can help the developers with various monetization opportunities, distribution means with broader or more focused target segments and application discovery mechanisms, other sales funnel management and analytics services. On the other hand, with content selection and orchestrators own and partner's service o erings, the application ecosystem can di erentiated for certain user markets.
We believe that data on multi-homing can be used as an indicator on how di erent ecosystems are evolving with respect to the con ict of loosing strength while trying to gain it. For instance, a situation where all popular applications are available in all ecosystems, is a clear sign of reduced signi cance of the application markets as di erentiators, whereas strong di erences in the distribution of application categories would indicate that the application ecosystems are pro led di erently. To our surprise, we found no evidence of increased multihoming with application popularity and no signi cant di erences in application categories. We were expecting that the multi-homing applications are popular in all marketplaces, and that the majority of multi-homed applications act as front-ends for other services. However, the results suggest that the application stores, as for now, are strongly di erentiated with respect to developers' value proposition or that the cost to developers for multi-homing is prohibitive in many cases.
In order to really gain insight on how the ecosystems are controlled and governed and what are the developers' and content providers' varying business objectives in the ecosystems, more sophisticated indicators are needed. As an example, data on where applications are rst published and how soon they migrate to other ecosystems gives insight on which ecosystem provides the most innovative new applications for users, which is certainly one possible way to enhance the user attractiveness of an ecosystem. Re ecting this level of insight on application ecosystems back to the ways how di erent orchestrators leverage their main businesses could give us understanding on the health and success of the application ecosystems, as well as trending data as the indicators are analyzed over time. These topics will be considered in our future research.
The rst results of our work-in-progress research on the application platform
multi-homing are presented in this paper. There are some factors that might
a ect the generalization of these results. For example, the third parties' listing
services were utilized when collecting the application identi ers for Google Play
and Windows Phone Marketplace. Therefore, it is likely that the gathered set
does not contain every single product of the marketplace. However, in these cases,
the number of missing applications should be rather small. It should also be noted
that the collected data only presents the situation of one day. For example,
during a period of one month, Google Play lost over 7,000 applications while
almost 19,000 new application were published [
Recently, Apple's press release6 stated that the marketplace has more than 550,000 applications. Our dataset of over 428,000 applications is not necessarily complete and it is most likely missing applications that are not o ered for our geographic area. We also have no knowledge on how the number of 550,000 has been computed. However, we believe that the most important applications are included. Finally, since the marketplaces are growing rapidly, we believe that the exact numbers should not be as interesting as the trends and the overall picture.
In this paper we presented the rst results of our on-going research on multihoming in application platforms. We studied three dominant application platforms by collecting the data of over 850,000 applications. The results show that the multi-homing publishing strategy is utilized only by a small set of developers (6.8 %), and that there is no signi cant di erence in the multi-homing vs. all applications' popularity and type. Based on our nding, we discussed the potential e ects of multi-homing for the ecosystems. Further work includes a qualitative study of multi-homing developers and applications to explore the di erence to single-homers. A survey study could also explain better the motivations and strategies of the companies working in the multiple ecosystems. Furthermore, this paper presented a general view on the ecosystem; an analysis from the user's, the orchestrator's and the developer's point of views are another important topic for further research.
Authors thank Mr. Peter Virta, Dr. Tero Jokela, and Mr. Jesse Alho for excellent technical advices. In addition, Hyrynsalmi is grateful for the grant of Nokia Foundation.