Content apps and points of failure

How the nature of content defines the long-term failure point in content-centric apps

The supposed great misery of our century is the lack of time — John Fowles

It’s no news that we are living in a world of content glut. There is more content than any of us can realistically consume in the time we have. Many a great businesses have been built around enabling humans to create and consume more and more video, audio and written content effectively. However, given that human creativity has no limits and consumer consumption preferences are fluid, it’ll continue to be hard to anticipate successes and failures of content-centric apps.

My goal here is to develop a framework for thinking about content-centric apps such that their outcomes become easier to reason. To start, let us first define content, specifically a “unit of content”. A unit of content is something that one can appreciate standalone i.e. without necessarily needing to consume any other content along with it. Some examples are an article on Medium, a video on YouTube, a song on Spotify, a season of a series on Netflix (an episode could be a unit only if the episodes in the series are largely independent or each other).

The content-centric app stack

Stack for content-centric apps

Any content-centric app works across one or more of the three layers — sourcing, curation and presentation.

Sourcing layer is the part of the app that enables the app to source more content by enabling new content creation or through purchase of existing content. For example, for Medium, its sourcing layer is its freely available blogging tool which I used to write this post.

Presentation layer is the part of the app that exposes the content to consumers in a suitable form. For example, for YouTube, its presentation layer is its website, mobile and TV apps with all their nuances such as the theater mode.

Curation layer is the part of the app that functions as the brain, taking the large volume of content gathered by the sourcing layer and deciding what content to push to the presentation layer at what moment for which consumer. For example, for Spotify, its curation layer is its ML models that power Discover Weekly and playlists recommendations with consumers playing their part in creating and sharing playlists.

Not all apps play effectively across all the layers of the stack at all times. For example, Anchor, in its current form, is primarily focused on the sourcing layer as a tool for podcasters to create podcasts easily. That said, apps can always expand to other layers of the stack.

While content-centric apps might want to be cutting edge across all the layers of the stack, it’s usually one layer of the stack that turns out to be the most challenging depending on the nature of the content and the business context.

Divergence in the nature of content

Apps mapped on agony vs. volume as a proxy of their unit of content they mostly have e.g. Newsletters for Substack.

It’s obvious, but the experience of consuming text is not the same as that of consuming a video, and the experience of consuming a short video is not the same as that of consuming a long video. What works for one type of content might not work for the other. One way to understand the nature of content is to look at it as a 2×2 of Agony vs. Volume.

Agony quantifies the extent of suffering one might experience on consuming a unit of content of average quality. Agony is lower for a unit of content that is entertainment focused and to consume which only a small amount of valuable time is required. Defined this way, most YouTube videos are low agony because they are short and one typically consumes them when there is nothing better to do. If you consume a few YouTube videos that weren’t truly great, you aren’t really upset. It also helps that YouTube is free.

By contrast, it is more agonizing to watch an average series on Netflix instead of spending that time watching one that all your friends are talking about. Though it is entertainment, it is supposed to be very high quality entertainment that is worth paying for, and out of which we want to extract maximum value. You also need to invest a reasonable amount of time in watching at least a few episodes to determine if a series would be worth continuing to watch or not.

Given this lens, long-form written content is the highest on the agony scale. Though it takes little time to consume, it takes a lot more focus than mindlessly watching YouTube videos. It isn’t entertainment and is instead something that we engage-in to learn. We typically need some discipline to set aside some valuable time for consuming such content, so it’s important that it better be worth it.

Volume quantifies how much content of a particular type is (or could be made) available on the web. Content type here is defined brutally in terms of length of the video or the article without making judgment about its quality. Volume of available content is determined mostly by how easy or hard it is to create a particular type of content of a passable quality.

Layers as points of failure

As mentioned earlier, it’s usually one layer of the stack that turns out to be the most challenging depending on the nature of the content and the business context. That layer is likely the point of failure in the long term. Eventually, a failure in any layer percolates through the stack i.e. sourcing layer will fail if curation or presentation layers aren’t set up to maximize the rewards for the content creators.

“Sourcing layer” failure

The risk of sourcing layer being the failure in the long-term is the highest for content that causes high agony but can only have limited volume, the most obvious example being a Netflix series. This puts a lot of pressure on the sourcing layer as we saw reflected in Netflix’s recent earnings. While this doesn’t mean that the curation layer is healthy, there is definitely an element of choice paralysis that is increasingly obvious whenever one turns on Netflix, it does mean that failure in the long term will likely come from the sourcing layer.

“Curation layer” failure

The risk of curation layer being the failure in the long-term is the most extreme for content that causes high agony and has high volume, the most obvious example being long-form written content. The bar for curation is just much higher. The failure of curation layer is usually a business model failure rather than an execution failure. Business model dictates what kind of mechanisms are put in place for curation to ensure that the best content creators get rewarded in a sustainable manner.

The failure of this layer is the most evident in the case of Medium whose excellence in sourcing stands in stark contrast to its suboptimal curation. Medium has failed in the curation layer because of two reasons: One, its compensation model ensures that that the only way for writers to make a living on Medium is to play the volume game i.e. publish a lot of articles so that cumulatively they can get eyeballs needed to translate the $5 per month subscription per member into something tangible for the writer. Two, it has failed to develop any higher level trust signals for the quality of content. Claps on articles are pretty much the only signal as opposed to say publications whose editors are incentivized to carefully curate content. Publications on Medium never really considered themselves to have an editorial responsibility beyond a certain basic level; they are all trying to just aggregate as much content as they could.

Medium now has the responsibility on itself to be the centralized curator which it possibly can’t do a great job at given the nature of the content (high agony and high volume). Compare this to Substack that considers the unit of content to be the newsletter as opposed to an article, and facilitates a direct connection between the newsletter creator and the consumer. The curation, therefore, happens at the human curator level which is likely to be very high quality, and there is all the incentive for it to be high quality because the creator of the newsletter is earning a stable subscription income for it.

“Presentation layer” failure

The risk of presentation layer being the failure in the long-term is very little. While presentation layers need to adapt to changing customer behaviors, successes or failures in this layer are mostly short-lived. The longer term advantage lies in the curation and/or sourcing layer depending on the nature of content. The apps most susceptible to these short-lived failures of presentation layer are either legacy businesses that haven’t yet caught up with large shifts in the new forms of presentations or businesses dealing with content forms that are in the state of flux.

An example of the former would be The Economist which, until recently, had a terrible mobile app that limited engagement inspite of excellent content. An example of the latter would be short-form videos which have been going through a shift to vertical video (and stories) with the increasing ubiquity of internet connectivity on mobile phones. Even long-form written content is going through some presentation layer flux with newsletters in email inbox gaining traction as the preferred method of consumption.


Creation and consumption of content is central to human creativity, and to us understanding and enjoying the world around us. Given this deep relation to human identity, content-centric apps are very powerful and also very hard to get right. I hope that looking at these businesses through the lens of the content stack and the nature of content is a helpful step in developing some intuition around these businesses. Feedback welcome.

Twitter @setparth

Physical assets & leverage windows

Physical assets & leverage windows

Is Tech’s new found love for physical assets justified? Yes, but only in certain scenarios.

Ownership of physical assets has historically been a dirty word in tech. The Valley has been spoilt by years of asset-light growth from companies such as Google, Facebook, and recently, Uber. The valuations of these companies have amazed even the most optimistic observers. Most of these business arguably have “network effects” that create winner-takes-all markets.

However, as customer usage patterns evolve and the opportunities to create large asset-light businesses diminish, there is a newfound interest around full-stack businesses. While full-stack business is a broader term that encompasses a range of operationally intensive businesses, my goal here is to talk about a subset of tech businesses that hold physical assets for the long term (a year or more) as a core part of their business + have a direct consumer touchpoint.

This classification includes businesses such as Amazon, WeWork, Sonder, b8ta, etc., and excludes businesses such as Opendoor which hold physical assets for the short term as part of a transaction. These businesses have built (or are building) demand on top of physical assets that sit at the core of their day to day. While a lot of businesses have physical assets, some of these businesses are special in that the physical assets are a significant point of leverage, resulting in high multiples.

When physical assets provide leverage

The best physical assets for tech companies are the ones that don’t exist but are really needed by consumers, and on the top of which it’s possible to build a demand driven network effects business.

I believe that as long as the 3 conditions below are met, holding physical assets for the long term provides a huge point of leverage.

Condition 1 — Demand for a new type of asset

There is a huge consumer demand for the kind of physical assets the business is creating, and the physical assets of that type either don’t exist or at least don’t exist at the desired scale

Condition 2 — Unlocks network effects

Significant network effects can be unlocked on the back of these physical assets that would provide the business defensibility

Condition 3 — Long expiry date

A shift in industry that could turn these assets into liabilities is far far away into the future (not at least in the next 5 years)

Deep-dive into some types of physical assets

Let’s look at a few examples to develop an understanding of the framework above. It’s important to note that physical assets that provided leverage yesterday might not be leverage points if investing in them today.

e-commerce warehouses

“Not a point of leverage any longer”

When they were set up, Amazon warehouses offered as high a leverage as one could imagine. With the promise of fast shipping, they triggered the demand that brought more suppliers to Amazon at cheaper prices. However, if a business was to set up warehouses today, Conditions 1 and 2 would both not be met. That is the reason Walmart is getting cold feet on future investments in its warehouse network. Today, the bar for creating network effects (Condition 2) on the back of warehouses is much higher; one has to promise one-day or faster shipping. What’s worse, a network of sophisticated third-party warehouses has already emerged, making warehouse aggregation a more realistic strategy (i.e. Condition 1 is not met).

Cloud kitchens

“A point of leverage today”

Food delivery is a hard business, with a constant need to provide consumers with the specific food item/cuisine they might want, faster. Cloud kitchens have emerged as a physical asset that meet all the 3 conditions. They trigger network effects (Condition 2) by generating demand within a food delivery ecosystem (say Uber Eats) that makes more such kitchens feasible, enabling the food delivery company to provide cheaper and tastier food, faster.

While there are already some third-party cloud kitchens (largest being Rebel Foods), there are very likely gaps in food types and density (location) that justifies companies such as Deliveroo holding cloud kitchens themselves. Over time, a more elaborate network of third-party cloud kitchens will undoubtedly emerge but there isn’t time to wait for that in the bloody food delivery wars; cloud kitchens would play their part in crowning the winner.

Co-working spaces

“Not a point of leverage today unless strategically differentiated”

The need for co-working spaces came about as it became easier to start companies and gig economy took off. Co-working spaces met all the 3 conditions at the time WeWork pioneered the concept. The network effects mentioned in Condition 2 were around the relationships and partnerships one could build by being in a WeWork as opposed to being a different space. Going forward, WeWork’s focus on membership benefits through the creation of a software stack would further extend the network effects beyond local network effects.

If one was to start a co-working business today, meeting Conditions 1 and 2 would require the co-working spaces to cater to a very different taste/community or use case. That is exactly what The Wing has done by promising deeper relationships and partnerships among women. Many other undifferentiated co-working spaces are unlikely to have any leverage, they will just be average businesses.

Service apartments

“A point of leverage today”

Airbnb exposed consumers to the wonderful world of alternate accommodation, and now consumers have an increasing preference for such accommodation over hotels. While there are many Airbnb accommodations, there is a dearth of reliable stays for professionals. That is where the new crop of startups (Sonder, Lyric, Domio), holding an inventory of service apartments, comes in. At this moment, they satisfy all the 3 conditions. Condition 2 is satisfied in the same way Condition 2 is satisfied for Hilton — more members means more certainty that one would find a Hilton on their next stay and the more the value of the loyalty membership.

Building new hospitality brands is the stated goal for all these startups. A few years later, building a business around holding these kind of assets would become much harder.

Retail stores

“Not a point of leverage today except in very few cases”

Most retail stores, though in-vogue, aren’t the point of leverage. They don’t trigger network effects (Condition 2). This includes stores for Warby Parkers and Caspers of the world. These stores are definitely good for business, but don’t drive network effects. However, the retail stores of the kind b8ta is building meet all the 3 conditions. They meet Condition 1 because online discovery for cutting-edge products is hard and there isn’t any way to try these products out without buying them. They drive network effects (Condition 2) similar to how Walmart stores drove them in their time — more stores meant better (and more) suppliers at better prices.


After a couple of decades of returns from asset-light businesses, it is time to appreciate how the physical and digital together can drive huge business outcomes. Physical assets held within leverage windows, and with a network effects based demand generation model on top, can be THE fuel for successful businesses of the future.

Service & Marketplace-model fit

Service & Marketplace-model fit

Services are ambiguous, and creating an online services marketplace requires finding a fit b/w the service and the marketplace model

The landscape of marketplaces for services is littered with failures. There are many a stories of promising startups that failed to create an enduring business. Services, by nature, are ambiguous and their lack of standardization makes it hard to create liquid online marketplaces. Plumbing, for example, isn’t a fixed priced SKU you can buy on Amazon and be 100% sure that it would solve the pipe leak in your house.

Services span a whole range complexity, from low complexity services such as food delivery to high complexity services such as home remodeling. There is no “one size fits all” marketplace model that works for all services. However, an understanding of the patterns of what marketplace models are best suited for what kind of services would make it easier to spot opportunities and to avoid failures.

My attempt here is to articulate these patterns by defining the 4 categories with a strong service & marketplace-model fit, and a couple that are misfits. Below is a chart that lays out services on complexity vs. frequency of service, and marks the 4 categories where successful marketplaces have been (and and can be) built.

Categories of marketplaces with a Service & Marketplace-model fit

Categories of Fits

Fit Category 1 — Lead generators for medium/high complexity (medium/low frequency) services

Given the ambiguity inherent in service marketplaces, lead generators have been the most obvious and the earliest way to bring services online. As the name suggests, lead generators are not involved in the transaction and make money simply on surfacing (or establishing connection) between the consumer and the lead.

The prime example of a marketplace in this category is Thumbtack, which is a lead generator for services across all complexities and across all frequencies. However, I suspect a big part of Thumbtack’s revenue comes from medium and high complexity services such as event photography compared to low complexity services such as home cleaning.

Lead generators could take a share of the market in any complexity level but typically, the value of the lead (and the trust the marketplace creates behind that lead) increases in proportion to the complexity of the service. There is no reason for one to use lead generators for procuring low complexity services when there might be specialized marketplaces, offering a much better consumer experience (Fit Category 2) for procuring those services.

A good example of a successful lead generation model in a specific domain (everything home related), but cutting across medium and high complexity services, is Houzz. It builds trust for the leads by showing detailed photos of their past projects, Houzz badges and awards, affiliations, reviews, etc. Arguably, Houzz has taken away some of Thumbtack’s market in home services but Thumbtack’s staying power comes from the trust and brand awareness it has engendered because of its breadth of services. Zola is another good example in the wedding domain.

Fit Category 2 —On-demand transaction marketplace for low complexity (high frequency) services

This category encompasses specialized marketplaces focused on low complexity (high frequency) services such as food delivery, ride hailing and grocery delivery. Prominent examples are DoorDash, Lyft, Uber Eats, Instacart, etc. These marketplaces go beyond lead generation, and into transactions, and they owe their liquidity to the demand that is generated (and maintained) on the promise of a much better consumer experience for buying services compared to lead generation marketplaces. In that sense, they deepen the market and create more liquidity than there would have been otherwise.

However, given the low complexity (and the commoditized nature) of the services being provided by supply side, trust mechanisms in these marketplaces are less important, and marketplaces in this category have a hard time preventing supply-side multi-tenanting with direct and indirect competitors. Supply-side keeps on switching across marketplaces and jobs (same person could do grocery delivery today with Instacart and ride hailing tomorrow with Uber), and this leads to very tough unit economics. Andrew Chen has a good post about it. Such marketplaces will continue to exist, though there will likely be consolidation that combines multiple low complexity and high frequency services into one marketplace.

Fit Category 3 — Managed transaction marketplace for medium complexity (medium frequency) services

This category is relatively underdeveloped, with Puls and Setter being good examples in home services, and Soothe and Glamsquad being good examples in wellness. The services offered by these marketplaces sit in the sweet spot of being complex enough that supply-side is not constantly churning (and multi-tenanting) and not complex enough that the buying experience can’t be standardized into a few clicks. This enables these marketplaces to offer a consumer experience 10x of what the lead generators could provide, thereby capturing sustainable demand.

These transaction marketplaces have a high degree of “managed” component. Puls, for example, vets its technicians and provides a 90-day guarantee on its services. Its buying experience is also very much tailored to each service with as much clarity on pricing as reasonably possible. Setter, on the other hand, introduces a trusted home manager into the mix to instantly help find the right provider.

More specialized marketplaces, on the lines of Glamsquad and Soothe, might emerge in this category as more service buying continues to move online, and more segments of services (e.g. dog training — just a random guess) become large enough to justify specialized marketplaces.

Fit Category 4 — Managed transaction marketplace for high complexity (low frequency) services

This category is also underdeveloped, with Upwork being the best example of being successful in the desk-work space. Note that some of Upwork’s services might better fit Category 3 above. The marketplaces in this category are what James Currier called Market Networks. The focus for the marketplaces here is to bring as much of the offline negotiation involved in a complex transaction online, into an interface managed by the marketplace. The power of these marketplaces comes from a few aspects (not all might be present in one marketplace): a) vetting of suppliers b) SaaS tools to work with these suppliers (before and after one has hired them) and c) guarantee when things go wrong.

BuildZoom is an evolving example of such a marketplace for home projects in the consumer space. Managed by Q is an evolving example for office projects in the B2B space for some of its services. It should be noted that the consumer experience marketplaces in this category need to create is very different from the experience that marketplaces in the Category 3 above create. Therefore, it’s very hard for a single transaction marketplace to effectively span across the Categories 3 and 4.

Note: As consumer expectations are evolving towards expecting lower and lower friction, marketplaces in Category 3 and 4 would pose a threat to lead generators (Category 1). That, in my view, is the reason Thumbtack went through all the pain to change its matching approach to be more instant and effective.

The Misfits (or Loose fits)

While it’s interesting to understand the categories of fits, to appreciate the nuances of service marketplaces, it’s also important to recognize the misfits. Below are 2:

Misfit Category A— Lead generators for low complexity (medium/low frequency) services

Prominent example of a marketplace in this category is TaskRabbit. With its positioning as “get affordable help”, TaskRabbit cornered itself into mostly being an option only for low complexity (medium frequency) services that one wouldn’t be able to find elsewhere at a better price e.g. home cleaning, or low complexity (low frequency) services that one wouldn’t be able to find elsewhere at all e.g. standing in a queue for a ticket. To be clear, TaskRabbit is not a failure in absolute sense, but it turned out to be much more of a niche player than it seemed at one point of time because it did not do what was needed to build trust in buying medium complexity services.

Misfit Category B— On-Demand transaction marketplace for low complexity (medium frequency) services

Prominent example of a marketplace in this category is Handy. Its mainstay is home cleaning, which not only is medium frequency, but also suffers from the same supply-side churn we had talked about in Fit Category 2. The medium frequency and low complexity aspects of home cleaning makes it really hard to retain supply, and unfortunately for Handy, there also isn’t any silver bullet to providing consumers a 10x experience when buying cleaning from Handy as opposed to buying it from a cheaper lead generator such as TaskRabbit. This is unlike other low complexity services such as food delivery or ride hailing, where the high frequency and on-demand nature of services at least creates an opportunity for the transaction marketplaces to offer a 10x consumer experience compared to what a lead generator could provide.


Obviously, service marketplaces are hard, and I have been wrecking my brain to understand what leads to the varied outcomes. Hope the categorization of fits and misfits above resonates with you, and it provides a framework for evaluating opportunities and avoiding pitfalls. Please share any feedback.