As the number of Android users continues approach 1B devices, more and more startups are looking to deploy complementary applications on Android in addition to their iOS applications. Some startups are beginning looking to launch on Android first. But the dynamics of the Android app store are quite different than iOS. In fact, they are much more challenging for startups.
Like iOS’s app store, Android’s app store also ranks applications. But the ranking system is very different in four key ways:
Android seems to tune/control the charts to encourage fewer large movements of applications within the rankings. The rankings aren’t wholly calculated from the previous hour’s downloads like iOS.
As a result, it’s much harder for new entrants to break into the top of the charts on Android than iOS. Existing top applications are much better insulated from competitors.
But, the decreased volatility means paid customer acquisition is much more cost-effective on Android, presuming the app’s engagement and monetization is similar across platform, because if your app can reach the top of the charts, it’s much less likely to be dethroned and the cost-per-install of the paid users can be amortized over a much larger organic user base who discovers the app by virtue of the high rank.
The competitive categories vary significantly from Android to iOS. This means there is probably an arbitrage opportunity to bring the best apps from iOS to Android.
To tell the real story behind climbing these app stores, I’ve been gathering ranking data from the app stores for the past two weeks.
Android Ranking Movements are Much More Subtle than Apple
In a given day, the average iOS app could move up to 50 ranks up or down. The typical Android app moves up to 12 slots.
Below are two charts comparing the variances of all applications on each store in a two week window.
An average iOS application might vary by fifty to one hundred ranks or more in a two week period.
On the other hand, the same application on Android wouldn’t see more than a 50 rank fluctuation.
Volatility Difference is Pretty Consistent No Matter Your App Rank
Now you might point out that apps in the lower ranks (200-300) should move much more than those apps in the top 10. And perhaps that explains the difference between Android and iOS. But that’s not true. iOS has larger variances at every rank bucket.
Compare these two charts that show the variance of applications that reached rank 10, 50, 100 and 200 at some point during my test.
Android apps tend to stay put where they are much more than iOS no matter if they are ranked three, thirty or three hundred.
Below is one way of looking at competitiveness by category for each of these app charts. The more the average application can change rank within a category, the better the quality of competition exists (perfect competition in the economic sense). Where variances are low, there is something like an oligopoly - for example ten apps that dominate the top ten.
Click on the images to enlarge them.
iOS most oligopolistic categories are Catalogs, Social Networking, Photo & Video and Food & Drink. The easiest categories to enter are Sports, Weather, Lifestyle and News.
On the other hand, Android’s most oligopolistic categories are Arcade & Action Games, Tools, Shopping and Racing Games. The best categories for new entrants are News, Medical, Finance, Business.
What Got You Here Won’t Get You There
Android’s market isn’t a carbon copy of Apple’s. Android requires a distinct strategy considering the much stronger tendencies to reinforce the top applications within Google Play. There are many drawbacks for startups when launching on Android but the addressable market may ultimately render those challenges moot.
With any analysis, there should be lots of caveats
- This is a two week study with a sample size of about 25k applications and roughly 1 data point per app per day for 14 days.
- Android may seem to have a less accessible app store because they are fewer quality apps and so the best bubble to the top and stay there. I can’t make a determination either way with this data.
- Looking at standard deviations and variances implies that the underlying data follow a Gaussian distribution and these data may not.
Last night, Elon Musk inspired the audience at D. It is hard to overstate the scope and breadth of his ambitions or the impact of his start ups PayPal, Tesla, and SpaceX on payments, automobiles and space.
Behind the desire to listen to great men like Musk speak about their perspectives is a hope to receive some insight, some pearl of wisdom. Last night, I came away with two of those both about startups.
First, when the inevitable question came along along asking Musk to compare his grand plans with the innovations and ambitions in photo sharing applications, Musk replied in the following way:
He said that there are many ways of creating value. Photo sharing apps create a small amount of value for a huge number of people. Aggregating that small value across billions can create big companies.
Second, when asked about the origins of SpaceX, his first major step to starting the business was to understand the cost structure of existing rocket companies. He and his team calculated the cost of all the materials required to build a rocket. They found that the titanium and carbon fiber and fuel represented only about 2% of the the total cost of launching a rocket. And in his words, the industry must be doing something silly to drive costs to their current order of magnitude. In other words, there was an opportunity to innovate by cost reduction.
There is one thread common to both of these examples: Musk understands the details, the small things - the cost of titanium screws and the marginal value of a shared photo, and how they relate to the overall picture.
His ability to think in systems, understanding how ecosystems react and respond to regulation and subsidies in the case of Tesla and oil and gas companies, and how PayPal fit into the world of payments and how the costs of screws and fuel fit into the overall cost of the rocket is what I admire about Elon Musk.
At the D conference yesterday, Tim Cook said many things without saying much. But one story did strike me. Cook described the product management and strategy process at Apple.
Walt Mossberg asked Cook why the iPhone has only one flavor when the iPod had so many including the shuffle, the nano, the mini, and the classic. Even though both products originally launched as a single model, the iPod flourished into a family of products while the iPhone has remained a single SKU.
Cook responded by telling the story of the shuffle. The original shuffle had no screen, focused primarily on shuffling music and was designed for athletes - a use case that while somewhat well served by the existing iPod, was much better served by a proprietary device. But it was only once the iPod had reached a certain scale of usage that the executive team at Apple could justify creating a child product. By that point it was apparent that a new customer segment had emerged that wouldn’t cannibalize their existing business. The data proved it.
Say what you will about Apple’s phone strategy, and of course that might all change radically at WWDC in just 10 days, but what did come through and Tim Cook’s description of the product management process at Apple is a clear understanding of customer segments: their desires and needs. Throughout the evening Cook spoke often about customer satisfaction scores and used the refrain of “customers tell me this and that” frequently.
It was apparent to me that even at the executive levels of the world’s largest technology company, user centric design is paramount at Apple. But we knew that.
The wrinkle is that Apple combines this design with superb product marketing. customer segmentation, customer studies and use case analysis is at the core of the product development process.
Cook walked through Apple’s disciplined process to first understand the market, justify the need, develop a deep understanding of the market segment and only then release a product.
Like any discipline, it’s easy to describe but hard to maintain. And a disciplined process like this will always be in conflict with product management by epiphany or vision. But I think it’s something we can all learn from as startup founders, product managers and VCs. It’s block and tackle product marketing.
You have just launched your new software start up. The last webpage to go up on the website is the pricing page. Like many other SaaS startups, you decide to employ some version the three pane pricing plan: first the free version, second a paid upgrade costing between $5-$40 per month, and third an enterprise tier with a “Call for Quote” in place of the price.
A few days after you launch, an enterprise customer contacts you asking for a quote. You respond with an offer that significantly higher per seat than the paid plan. Let’s test the waters, you think. And having seen your pricing page, the enterprise customer asks why the enterprise tier is so much more expensive than the paid tier. After all, the cost to deliver the service for each kind of user, whether individual or enterprise, is the same.
Serving Two Customers at the Same Time
This dilemma is called channel conflict. The pricing page is trying to serve two contrasting customer segments: the individual who wants to upgrade and the enterprise looking for a companywide or team-wide plan.
Each of these customers perceive the value of your service differently and each has a wildly varying willingness to pay.
Typically, the individual user is on a budget. They might be paying for the service out of pocket and are generally price sensitive. As the founder, your strategy might be to maximize the number of individuals on the service in order to drive bottoms up adoption. So instead of maximizing revenue from this segment you price the paid tier to generate profit based on the cost to acquire the customer and cost to serve the customer. This is cost-based pricing - charging the customer the your cost to deliver the service plus some mark-up.
On the other hand, the enterprise customer will derive significantly more value from the use of your service than the individual . Additionally, this customer segment has the capacity to pay orders of magnitude more than an individual who upgrades to a paid tier. So quite naturally you want to capture some fraction of that value. This is called value-based pricing.
Rationalizing Channel Price
It’s hard to convince one customer segments to pay based on cost-based pricing and another to pay on value-based pricing on the same page for nearly the same service. So, you have two choices.
First you can distinguish the paid and enterprise products by offering enterprise-level support, service-level agreements, and IT management features that aren’t available in the paid-tier in order to justify the value based pricing to the enterprise customer. Many startups pursue this route like Yammer.
Second, you can create different websites or even brands for different segments. Within each website, one segment is prioritized and the marketing copy, pricing page, and other content all serves that particular segment. In retail, brand name drugs and generics serve different customer segments the same products at different prices (value vs cost-based pricing). This strategy can be harder for start ups because of the complexities managing two or more friends at the same time.
Whichever route you decide for your startup, the most important thing is to clearly determine which customer segments you’ll be charging on a cost basis and which you’ll be charging on a value basis. Clearly separating the value proposition either with feature disparity or by creating different brands will be critical to supporting the pricing differences between the two segments.
: According the US Census, a 5k person organization pays more than 50% more in payroll than a 10 person company. So at the very least, in larger organizations, an employees time is worth 50% more to an employer.
“First mover isn’t what’s important — it’s the last mover. Like Microsoft was the last operating system, and Google was the last search engine.”
I hear this refrain more and more in pitches. The thinking goes the last entrant to the market benefit from mistakes made by earlier entrants.
But being the last mover isn’t always advantageous. The reality is more nuanced.
When Last Mover Works
Technology, platform or behavioral discontinuities. The last mover is the agent of Clayton Christensen’s Innovator’s Dilemma embodied. It’s the company that pursues an incumbent with faster, better or cheaper solution and in particular a solution that cannibalizes the incumbent’s business model typically because of a lower cost structure.
While Google did develop a better ranking algo, Google also had a fundamental cost advantage. Google served queries at more than an order of magnitude less cost than its competitors, who used expensive Oracle database servers, enabled by Google’s investment in running its software on commodity hardware.
Zynga leveraged Facebook’s Open Connect platform to grow its casual gaming business much, much faster than incumbents Pogo and others. The dramatically lower cost of customer acquisition on Facebook, enabled by a an agreement between the two companies, fueled Zynga’s rise.
Instagram, “the last photo sharing service”, took advantage of a behavioral discontinuity, the explosive growth of mobile phone snapshots driven by high quality cameras in smart phones, to threaten Facebook’s photo sharing dominance.
Last movers can be a challenged by incumbents when the last movers have no discontinuity to bring to bear, face the network effects of incumbent transactional market places, are challenged by low margin competition, or face significant IP battles.
Priceline, “the last online travel agent” was founded in 1997. Bill Gurley wrote a great post outlining how Priceline won the lion’s share of the online travel market by reducing their rake. The larger the market place, the greater the monopoly, the more it can reduce rake to starve competitors. Craigslist’s strategy of largely ignoring monetization has prevented thousands of competitive market places from competing with it.
Amazon’s ability to operate at basically zero profit prevents ecommerce threats from rising. Compare their revenues to their profit. It’s hard for any ecommerce company to raise the capital required to build a competing business when your competition is happy with zero net margin.
Akamai exerts huge pressure on the CDN market by litigating competitors out of business. Here’s a quote from a Gartner report: “How to tell when a CDN has arrived: Akamai sues them for patent infringement.”
While it’s true that Microsoft and Google eventually came to dominate their industries, it wasn’t necessarily because they were last. Of course there were operating system companies and search engines that followed (Jollicloud and Cuill) for example. Instead, Microsoft and Google’s market power, their ability to keep their advantage in the market over time by anticipating discontinuities has enabled them to remain top dog — at least for now.
Takeaways for Startups
When looking to take advantage of the last mover advantage, ensure that you can leverage a clear discontinuity in the market.
Taking advantage of this sentiment, Expensify employs a very deliberate marketing tack: “Expense reports that don’t suck.” Talk to anyone who uses antiquated expense report systems and they are bound to sigh and complain, frustrated by the experience but resigned to the fact they can’t do much about it. Expensify provides those people with a better alternative and, most importantly, empowers them to change the way they work.
This pattern is true for Heroku and developers inside of large companies. It’s often much faster to build and deploy a project externally on Heroku than to coordinate with IT to deploy on internal resources. And those developers are much happier for it.
Dropbox and Box and Yammer, all of these companies are allies of the end user. To build internal support support, CoIT companies offer much better products to users, build momentum within an organization and eventually enable users to convince or demand that IT change vendors or policies.
As critical as building the right product for end users, CoIT companies ought to provide the right tools for the IT organizations who are often challenged by end users creating security challenges, demanding better tools, and on the whole pushing the IT team beyond comfort.
Each one of the companies presenting today will have a problem statement that focuses on the antagonism between employees and IT. Frustration is the indicator of opportunity.
Reading through the tech press since the Facebook IPO, you might get the impression venture capitalists are still reeling from that apocalyptic offering, believe no further successes can be had in the consumer web, and so are fleeing the consumer web in droves to pursue enterprise investments.
That’s because in the past year or so most major tech publications have swung from focusing on consumer products to enterprise companies. GigaOm made this transition first, now TechCrunch and PandoDaily are following suit.
But the problem with sounding the alarm for the ebbing consumer investment is that just not true.
Consumer investments historically have garnered massively disproportionate press coverage compared to their volume because consumer products are easier for readers to understand than enterprise technology. Consumer products evoke emotional reactions, drive page views, and build business. It’s hard to jump for joy or trigger lots of retweets over data center virtualization innovations unless that’s really your cup of tea.
To put this into perspective, what do you think is fraction of venture investment dollars last year were in consumer companies?
How about over the last 17 years?
Here’s a chart of NVCA data for that period. Click on it to enlarge it.
So 84% of venture dollars invested over the past 17 years have been invested in enterprise companies. It’s clear that enterprise investments are the bread and butter of the venture business. And that trend isn’t changing.
This massive swing toward enterprise investing never existed. It’s a fallacious perception. Enterprise investing has always been the norm and will continue to be for quite some time.
I started working in ad tech in 2005 and during the past eight years, the ad tech ecosystem has progressively become more sophisticated, competitive and oligopolistic. It’s hard to innovate in ad tech. But if you’re looking to start a company in the sector, you’ll need to amass proprietary data or develop a market place with unassailable liquidity to vie successfully in the market.
A Mental Model for the Ad Ecosystem
The structure of the ad ecosystem, greatly simplified, looks like the image above. On the left, the advertiser supplies dollars that flow to the right. The DSP, demand side platform, uses algorithms to inform an advertiser’s media purchases; i.e., which websites and mobile apps will perform best? The advertiser and DSP purchase media on the ad exchange which is an electronic market place where advertisers can buy media algorithmically and in real time, called RTB for real-time bidding. On the other side of the exchange, the publisher uses supply-side platforms to find the best paying advertisers to buy their ad inventory.
There are hundreds of SSPs and DSPs, thousands of advertisers, millions of publishers but only a handful of exchanges: Google’s DoubleClick, Facebook’s FBX, Yahoo’s RightMedia, MoPub, Adap.tv and a few others. These exchanges, like most market places, exert huge network effects because advertisers are attracted to the exchanges with the most inventory selection/liquidity. The exchanges see every transaction and have unparalleled visibility and data access into their respective ecosystems.
Data, data, everywhere
If there’s one defining characteristic of online advertising, it’s data. Advertisers buy data and license algorithms to find better inventory. Publishers sell their data and license other algorithms to find better advertisers.
In order to compete in an ecosystem of data, a startup has to bring one of three advantages to market: better algorithms to use on the same data as everyone else, better data than anyone else or a market place with the largest volume of ad inventory in a segment.
Better algorithms is the fastest way of getting into market as a startup. Similar to starting a new quant hedge fund, you develop a novel trading strategy that works and sell it to customers. But competition in ad-tech is just like the financial markets - as soon as others see your strategy working, they are likely to copy it. Over time, the marginal advantages of better algos erode. Unless a startup continues to invest heavily in algorithm improvement, it will forever be in a cat-and-mouse game.
Better data: Where algorithms can be conquered, proprietary data is unassailable. With access to richer ad performance data, more detailed user data, more granular conversion funnels, your startup has created a significant barrier to entry. Better data means better results. And if you’re the only game in town, then you’ll attract big advertising budgets.
Getting access to better data is very challenging. It means finding and partnering with publishers and/or advertisers on an exclusive basis for some period of time. And then leveraging that data to build a successful DSP/SSP/ad network.
Market places in ad tech, as in the rest of the tech industry, are beautiful things. They are natural monopolies, capital efficient and are strategically valuable. Building a new ad tech market place is the most challenging way of entering the market because of the strategic role these products play in the ecosystem.
The most successful startup market places (RightMedia, Adap.tv, BlueKai, MoPub) each took advantage of a discontinuity in the market place (inventory glut, video ads, rich user data, mobile ads) to develop a foothold in the market faster than the incumbents. Over a few years, each of these companies built liquidity into their market places and now are the leaders in their segments.
The Recipes for Success
As the ad tech ecosystem has bloomed, competition has increased dramatically. To best position your new ad tech startup for success and develop a long term advantage, you need to develop leverage by developing proprietary data sources or by creating a market place based on some technology discontinuity. In other words, bring something to market that no one else has and that is difficult to copy.
On the day of Tableau’s IPO, a company known for innovating in data visualization, I thought I would share the most impressive HCI concept I’ve seen in a long time.
In my view, Bret Victor is on the forefront of human computer interaction design. In the first two or three minutes of this video at Stanford, he demonstrates his home-built software that combines data analysis with visualization. It’s magnificent and really hard to describe because it’s so novel.
Victor uses words to create and manipulate the drawing and it all seems so natural and fluid. The interaction with the software evokes a conversation between two designers over a piece of paper - which is exactly the kind of interaction that makes software seem human and natural. I hope we see an explosion of these types of tools in the future.
When I was a little boy, I watched a cooking show on Sundays called “Jacques Pépin.” Over the course of 30 minutes, Jacques would orchestrate a symphony of raw ingredients into a dish that I yearned to smell and taste. That chicken paillard looked sumptuous.
Over the past few weeks, I’ve been coding in Rails 3 quite a bit, building a collection of tools to be more effective as a VC. And oddly enough, I think back to Jacques Pepin quite often.
The Jacques Pepin of the Rails world is Ryan Bates, creator of RailsCasts. Watching Ryan’s shows I’ve learned how to build AJAX search, deploy an identity management system and code beautiful charts in ten minutes each.
There’s something intoxicating, exciting, and liberating about watching someone convince you of your capacity to perform a seemingly challenging feat, whether cooking or coding, by showing you how easy it is to do. It’s a testament to the power of video in teaching.
Buttressing Bates' tutorials, StackOverflow has become my essential tech support community. Paste an error message into Google and the first link is bound to contain a fellow coder’s struggles and eventual solution to the problem I’m now facing.
The combination of tutorial expert video and support community is immensely powerful because it shares the expertise of a master with many while scaling the support and help needed by a mass audience. I believe this form of teaching will become ubiquitous for this reason. It’s already happening in coding with Railscasts and StackOverflow, university education with 2U, gaming with Machinima and many others.
Jacques Pepin never did tell me how I botched that homemade mayonnaise. But there’s someone on the web who has watched the same episode, made the same mistake and found a solution who might.