App unbundling, search and discovery

"There’s only two ways I know of to make money: bundling and unbundling."

-Jim Barksdale

There’s been a pretty clear story for the past couple of years that the dynamics of smartphone interfaces drive unbundling of services. 

The main reason is that single-purpose applications have an advantage on smartphone interfaces. On a desktop website, you could always add a feature as a new tab on your site navigation, and clicking that was easier than going to another web site. But on mobile, all apps are just two taps away, while there is very little room for links to more features in any given app’s home screen - they quickly get shoved into the ‘hamburger’ menu, and soon after fall into the ‘hamburger basement’, down below the bottom of the screen. This gives startups scope to unbundle features from larger companies' products, and also means those companies themselves unbundle their apps into separate single-function apps. 

This effect is even stronger for social applications, where a lot of the unbundling conversation focuses, since the smartphone is itself a social platform quite unlike a desktop web browser. All apps can access the address book, giving them a ready-made social graph, and the photo library, reducing the hassle of uploading to different sites, and push notifications, removing the hassle of checking lots of different sites. This is why we have dozens of social apps that have passed the 1m, 10m or even 100m user milestone. 

Finally, of course, driven by all of these dynamics, all smartphone apps by their nature unbundle services from the web browser itself, 20 year after Netscape launched. 

All of this seems inevitable and inherent in the nature of the platform. It’s the reason why Facebook has moved from a de facto monopoly of social on the desktop to being merely the leader on mobile, buying the occasional breakout unbundler while experimenting with its own suite (‘constellation’) of different apps searching for new use cases

However, looking at the Chinese internet market is always a great way to challenge your belief in what’s inevitable, and in this case both Baidu Maps and WeChat, amongst others, are thriving on exactly the opposite approach - bundling multiple services into one app. (WeChat and Baidu Maps are a bit different, so I'll focus on Baidu)

Hence, in Baidu Maps you can not just find a restaurant but book a table or order home-delivery. You can find where a film is showing and choose a seat. You can order a cab, with cabs from all the services in your city shown together on the map and the order going to be best offer. You can book a hotel room or a cleaner. Most of of these services are provided through embedded third-parties - Baidu is doing deals with other Chinese internet companies to provide this. WeChat (which now has close to 400m MAUs), is doing much the same, but of course starting from a chat experience rather than a maps experience.

In both cases, the aggregations make a fair amount of sense on their own terms - finding a cinema, taxi or restaurant are location-based activities and they’re also social activities. But the functionality is being flipped upside-down: in the USA a taxi app or a restaurant app will have an embedded map, but in China the map has embedded restaurants and taxis - and lots of other things. Equally, on Yelp you can message friends, but in WeChat you’re already in the messaging app. 

This changes the layer of aggregation. By default, the iOS and Android interfaces aggregate at the app launcher level - you have a screen of different icons for different services, though to some extent notifications are becoming another OS-level aggregation. Anything that hasn't been unbundled from the web is in the browser, where it is in turn aggregated by Google. Apps like WeChat and Baidu Maps move the service aggregation layer up the stack from the home screen, bundling it within a single app. Conversely Siri and Google Now (and arguably Microsoft's Live tiles) are in different ways about moving it down the stack into an OS-level service. But the interesting thing is that changing the aggregation model also enables new types of discovery.

In the app launcher layer, any direct service discovery (as opposed to marketing, word of mouth etc) is limited to the app stores. When they launched this worked adequately, but with over 1m apps in each of the iOS and Android app stores this model is clearly now broken. Just as Yahoo’s hierarchical directory didn’t really scale after 1996 or so, so too app stores have not scaled as a discovery platform. Further, though there are lots of ways that the execution of the app stores could be improved, this still ultimately boils down to saying ‘Yahoo’s home page should be better’ back in 1996. Indeed it should have been, but the answer was still PageRank. And we have no analogue of PageRank for mobile apps. 

Google Now and Siri, moving down the stack, address discovery by removing the whole question of ‘which app/site/service addresses this need?’ Now tries to infer from your general activity what you might want to be told about, while Siri tries to use natural language to answer your question. Both are focused on an answer rather than a specific service - ‘sports scores’ rather than ‘ESPN’ (even if ESPN proves the actual answers). But the problem with both is that underlying answers are provisioned in an essentially manual process. Google Now, in particular, looks like Google’s most automatic and magical product, but in fact it's all editorial. Someone has to decide that Now will support cricket, and then someone has to sit down and code the cricket module. And If no-one has done rugby yet, you won’t get rugby - Now can't work out what rugby is from watching your web browsing. Siri has similar constraints: HAL 9000 didn’t have a list of things you could ask him. This does not scale to the whole internet. 

The general issue here is that any aggregator of services that offers a set of essentially hand-crafted approaches will inevitably run out of room in the metaphorical hamburger menu, just as the app store home pages can only showcase a few dozen apps at once and Yahoo's home page could only showcase a few dozen services. Or, indeed, Apple's Sherlock ran out of tabs. This doesn't scale to the internet. PageRank did and indeed the web did. 

The Baidu Maps model addresses this problem by narrowing scope to a specific domain, location, such that the discovery (hopefully) flows out of a natural pre-existing context. You don’t need to know about restaurant booking services or download the app, but you also don't need to know if Baidu Maps has an AI that can do restaurant bookings (as you would need to know that Siri connects to OpenTable): bookings are an organic extension of the context you're already in. You'll already be searching for restaurants on the map, but now ‘make a booking’ will appear.

That is, searches within maps are naturally constrained - there are only so many use cases (a dozen? two dozen?) where a service integration might make sense. So Baidu can make a maps app that’s an order of magnitude better because it contains within itself every possible maps-based use case, and nothing more. And those use cases can be hidden until you search - you don't need to find space for a 'cleaners' button - you can just show that module when people do that search, as they will.

WeChat comes at this discovery issue from the other direction: the social foundation of the app means that services can come to you thRough friends or through subscriptions to channels, but the point is the same - you don't need to find space on the screen for every single service for people to find them. For Facebook, the model would be to drive traffic out into stand-alone apps (or a web view), with the partnership being only an advertising one - for WeChat the dynamic is perhaps closer to the original Facebook platform on the desktop. 

One way to look at this, I think, is that you trade app discovery challenges for feature discovery challenges. That is, in an app with only one or two features it's easy to see what you can do, but hard to discover that app in the first place. Conversely in an app with hundreds of millions of users but also lots of integrated features, feature discovery becomes the challenge, and the opportunity. 

Changing discovery is one thing, but the other interesting thing is how much happens inside the app. There are two structural things that the desktop web has that the mobile app ecosystem lacks. On the web:

  • You can link to any resource
  • That resource will be there - you don't need to install anything.

These aggregator apps are tackling both problems: they have discovery models to allow you to get to the resource (e.g. social or local context), and the resources are loaded within the app - no need for a new app install. This isn't happening on the web, as such, but it has some of the same dynamics - within, obviously, those limited domains. 

It's interesting to contrast this with the deep linking initiatives from Facebook and Google. Facebook offers a way to drive traffic to your app from social and Google from search, and Google is experimenting with, for example, a link to Uber from within Google Maps. The problem is that if the app isn't there then you fail back to HTML, and if HTML was good enough then you wouldn't have bothered deep linking in the first place. But on the other hand, you can't integrate everything  - we already have a way to do that. Hence, the Facebook and Google moves (and also Line's) disaggregate apps, but use identity, deep links and search as an aggregation layer that is closer to the web and might be more scalable. 


That is, you can bundle things into an app that has distribution, but there's a limit to how many both for practical and discovery reasons. If you use context (maps, social) you can address the discovery problem but this works best in constrained contexts. If you don't bundle, your features are easy to discover (because your app only has one feature) but you give up those opportunities for discovery and fall back on raw search, social sharing or the app stores. 

And for all that we talk about the problems with apps and the lack of PageRank, one cannot really describe the desktop web as a prelapsarian paradise of simple and easy discovery. The web is a neutral, transparent and unmediated platform, and smartphones are not, but neither are Google or Facebook. 

Hence, as Lenin put it, the question is 'Who, whom?' Who has the traffic, and to whom do they give it? Mobile shuffled the deck, but it doesn't alter the problem. 

The other interesting comparison here is with Yahoo and AOL. The portal model failed pretty comprehensively in the 2000s, with category killer sites springing up and AOL and Yahoo and the other would-be portals falling far behind. So when smartphone apps came along we already had a dominant unbounded model that transferred pretty directly: we unbundled services from the browser, but the services themselves were already unbundled from Yahoo.

In China, though, we see a model in which the portals didn't disappear. For a lot of different reasons, the Chinese internet is based on much more horizontal than vertical competition: where we have vertical category killers (Google, Amazon, eBay) China has giants that are each leading in one category but strong in others, and competing aggressively with each other across every category. On mobile, Baidu Maps and WeChat are showing a systematic concept of the app as, well, a portal. Is this what Yahoo might look like on mobile now if it hadn't spent a decade asleep? 

Foursquare traction

This post is simply a sequence of Google trends charts looking at search volume by cohort for 'Foursquare'. It's unscientific and needs to be taken with caution (you need to think carefully about exactly what consumer behaviour you're looking at). But it's interesting. 

First,  global search. Stable, perhaps, but not growing, it seems.   

But what happens if we zoom into the USA? Suddenly the picture changes - a significant reduction in consumer interest.  

And the same in the UK. 

The implication is that a decline in interest in the USA (and UK) is being offset by strong growth somewhere else. Where? The regional view (which requires several handfuls of salt) point to two of the usual suspects for mobile social services- Indonesia and Turkey

Switching to a trend line suggests that Fourquare is in fact off the boil in Indonesia but growing fast in Turkey. This chart (and the map above) is relative within each country, not across countries - 4SQ has a larger share of Turkish search than of US search, but Turkish search volume for 4SQ is not necessarily higher than in the USA. 

If we switch to Turkey and add Whatsapp for scale, we see that 4SQ looks like it's getting real traction in that market. But of course it grew fast in Indonesia too...

Google Trends does not let us compare volume directly across countries. But a search by platform is suggestive. Look what platform it shows as most popular for 4SQ - the one that dominated in Indonesia, not the USA. 

All of this leads to the question - where, exactly, are Foursquare's users? And what are they worth? Google Trend lets us see a decline in interest in the USA, and a rise in Turkey, but we can't really see where the overall interest lies. 

Where do maps come from?

Maps are hard.

Google mounted the event below (by a strange co-incidence, a week or so before Apple went public with its own Maps project) to show just how much hard work goes into Google Maps, and followed up with an artfully placed piece in The Atlantic.

Many people seem to have forgotten now, but laughably bad results used to be a reliable part of the Google Maps experience, and still crop up from time to time. My favourite was earlier this year, when I searched for '42 Dean Street' while standing on Dean Street in London, and was given 'Deren Street', in a small town in rural Australia. Google Maps is a lot better than it used to be, but not anywhere near finished. Of course, no map can ever be finished, almost by definition. 

Moreover, though Google has indeed mapped most of the world itself, from scratch, there are maps and maps. The comparison below, between 'Google Maps' for major stations in Tokyo and New York, is intriguing. 


Why is the data so much richer in Tokyo? The answer, of course, is that this isn't all Google's data. Google has the roads and complete streetview photography and that's a great asset, but a meaningful part of that incredibly rich metadata is licensed from Zenrin (whose copyright appears in small text at the bottom of the browser). What exactly would Google's maps look like, without all of the third-party data that it buys in pretty much every country?

This, of course, begs the question of whether Apple can license the same: at the moment its maps for Tokyo are embarrassingly sparse. In contrast, the situation is China is... interesting. Apple is licensing data from AutoNavi for China, but this data is only available IN China. And the data it has for the rest of the world ISN'T available in China. So when you step off a plane in Beijing, suddenly your iPhone has access to fantastic maps for China and no-where else, but before you start the trip you can only see the basic data Apple has from other partners.  This doesn't really matter to very many people, but it's not ideal. 


On the other hand there are real practical problems with POI data even where Apple HAS done the licensing deal. Apple is taking feeds from a variety of partners with wildly varying data standards, and the integration isn't smooth yet. The most visible teething problem is with Yelp. Apple is licensing POI data from Yelp, embedding reviews and listings onto the maps. However, Yelp's data is far weaker outside the USA, and Apple isn't even taking a live feed - there are both reviews and restaurants near my house in North London that are in Yelp but not in the iOS maps results. 

This is part of a broader execution problem in Apple Maps. In London, Underground stations, which are absolutely crucial, appear and disappear on a daily basis. This blog post goes into some detail as to some of the things that might be at issue (though it has perhaps excessive certainty on these). These will get fixed, of course, just as Google fixed identical issues - it's just a question of when. 

Meanwhile, almost completely ignored at the moment are the 3D 'Flyover Maps' - which are sometimes quite astonishing. Personally I've found it far more useful than Google's Streetview, which after all only really shows you what your destination looks like, not how to get there. 


When this fuss started to blow up, my immediate reaction was that this was another 'antennae-gate' - a major fuss that turned out to be largely artificial. Actually it's a little more serious - it's more like the fuss around the absence of Flash, which was also a real issue, until it stopped being a real issue.

One of the underlying dynamics is similar, though: Apple deciding that it needs control over a key part of the experience. In this there's a certain irony: Apple wanted to own the maps experience instead of relying on partners, but is suffering at the hands of imperfect data and imperfect integration of data from... partners. 

Why hasn't Yelp worked outside the USA?

This chart shows the total reviews created for the top 25 venues in each category in London and in New York. It’s a pretty clear picture: people in NYC use Yelp and people in London, by and large, do not. I’ve combined Yelp and the European rival Qype and even so the gap is huge. There’s a similar picture if you look at other European cities. And if you look on the forums for Yelp UK, a substantial part of the traffic is clearly from American expats. 


I should point out that this isn’t a coverage gap - there are lots of places reviewed - but an engagement gap - places just have far fewer reviews. 

Yelp launched in the UK back in 2009 and smartphone penetration in the UK is not substantially below the USA. So what is it? ‘Cultural’ is easy, but a cop out. Certainly, people in US cities do eat out more, but is that enough? Has Yelp just not pushed hard enough? Or maybe Americans are just keener to share their opinions (cf Zagat). 

Apple Maps

Promo video for the tech behind Apple’s 3D maps. This product was spun out of Saab Aviation into C3, which Apple then bought.

“100 square km modelled per hour of flight time”. Has Apple been buying planes? Wondering if I should be looking at ATC/FAA data. They might be launching with entire states under coverage…