TLDR: This article concerns market and product strategy, particularly in fast-moving technology spaces. It provides some simple frameworks to think about, plan, and communicate strategic positioning at a range of different hierarchies both for life science companies and AI companies. While markets are different, the underlying tenets of strategy remain similar. The overarching question is how to position companies and products to win.

The main sections of this article are:

1. Introduction

2. Products and Platforms

3. Capital Costs and Benefits

4. The Five Forces

5. Market Navigation

For more like this, visit: jasonsteiner.xyz

Share

Introduction

Yesterday I built Perplexity.ai.

Okay, this is a bit of an exaggeration, but I did build a prototype of the underlying experience — that of an "answer engine" in lieu of the 10 blue links of a search engine. It was an expansion of the experiment I had done previously building a RAG system on the BioRxiv database1 that was augmented to include functions for web search APIs, scraping, summarization, and a simple citation section. Below is an example of the output with in-line references— and it even mirrors some of the features of what Perplexity.ai would produce (albeit in a less polished form) — and it's accurate.

Figure: A homemade “answer engine” along the lines of Perplexity.ai

Now, admittedly a prototype is not a product and the engineering behind the delivery of a product experience like Perplexity offers is impressive, but this is demonstrative of the points I made in the previous article about the extraordinary enablement of AI tools in supercharging the ability of people to create.

It is also demonstrative of how ubiquitous this will be — the accessibility being equally powerful and consequently challenging to differentiate.

In many respects, even highfliers like Perplexity fall into this category, though their competitive advantage is more solidly grounded in financial resources, talent, and momentum than anything technically differentiating (at least as it appears from the outside). Deep pockets and influential backers can provide strong defensibility in a market where scale and speed matter most which have been defining characteristics of the tech industry for the last two decades.

Some in the AI industry have mused about the single founder unicorn company — one individual who can run an enterprise valued at $1B. Given the raw infrastructure resources to reach the scale necessary to achieve a billion-dollar valuation, this seems challenging, but perhaps not too far off. Indeed, at least in the LLM foundational model space, 9-figure valuations are being ceded to single-digit headcount companies. However, all companies engaging in infrastructure, foundational technology development, or the application layer of the future of AI encounter the same set of strategic questions. While the pace, scale, and scope of what might be built is accelerating, I would argue that the basic principles of product strategy remain the same.

This article will outline a set of frameworks that provide a way to contextualize product and company strategy in the age of AI. They are generally not new ideas, but every market and situation does have a unique application. The purpose of strategic frameworks is somewhat to aid planning, but they are most useful for communication of structured thinking. For managers and executives who have strong experiential or technological intuitions about how the world may unfold, such frameworks are useful for bringing others along.

I will keep things as simple as possible, but no simpler. The two primary frameworks that I'll discuss are those of the perennial Product/Platform conundrum and the 5 Forces framework. At the end, we should come around to a simple conclusion:

I believe this quote is from part of Sequoia Capital's investment framework but cannot find the reference. Regardless, it's a litmus test:

• Unique solutions enable pricing power which appears in attractive gross margins

• Compelling solutions have market pull which shows up in attractive operating margins.

Combined with large markets, this makes for attractive businesses. The question of strategy is a question of how to find and build these characteristics.

Products and Platforms

Bill Gates has been credited with saying:

A platform is when the economic value of everybody that uses it exceeds the value of the company that creates it.

Certainly, in the world of operating systems, this holds true. The aggregate value of people using computers and their operating systems for any purpose vastly outweighs the value of the platforms themselves. This concept applies to both hardware and software. Platforms are enabling — and successful ones enable tremendous value for end users. Illumina sequencers, Google, CRISPR, GPT4, iPhone - all are examples of platforms, some physical, some software, and some technological.

A feature that they share is that they have broad utility that stretches significantly beyond their immediate instantiations and customers can build additional products and services on top of them or with them. There is always room for disagreement between the product/platform distinction, and the best products are also often the best platforms. It's fair to say the ambiguity is warranted, but a simple rubric to classify them is at the intersection of addressable market size and accompanying end value capture.

Simply put, a platform has a wide addressable market size with a fractionally smaller end value capture and a product generally has a narrower market size and a typically larger end value capture. This definition is based partially on the idea that platforms become integral parts of the production or composition of end market products, but along the lines of Bill Gates’ statement, must make customers significantly more value than they extract.

This distinction between products and platforms is purely economic — the market and product strategy for each and what drives them will be discussed in the next section.

To take a practical example from the techbio world we can compare and contrast the strategies of Gingko and Solugen — both synthetic biology companies that have elected to take different strategic trajectories in the product/platform dimension. Briefly, both companies have a foundation in organism engineering. This field is generally concerned with how to genetically engineer single-cell organisms to direct them, for example, toward the production of specific chemical or biological molecules. The applications of such genetically modified organisms are rather broad and cover everything from industrial chemicals and enzymes to pharmaceuticals, agriculture, etc.

The strategic distinction of the two companies is stark. Gingko aimed to take a "platform" approach, advertising that a very large market was available across all of the potential industries and applications for their platform and that they would be able to capture a significant fraction of downstream value from these large markets. The valuation of the company in its 2021 SPAC was based on nearly $10B in potential downstream value capture on market sizes that they pitched in the $T’s2. This seemed like the best of all worlds — the market for a platform is wide and their value capture would be deep. However, the contingency is that customers would both need or want to use their platform and said customers would actually develop products that would be valuable. Both of these were, and remain, questionable.

The contrasting strategy is that of Solugen — which has elected to take a product-first approach and use its platform as the internal engine to create product advantage. For example, among their first products was hydrogen peroxide — a commodity chemical with a global market size of ~$3.2B3. Not nearly as flashy as Gingko, but almost certainly more tangible. In this strategy, their addressable market segment for any specific product is likely smaller but more established and their potential value capture of the economics is potentially higher — e.g., they own a larger fraction of the production and commercialization. In an ideal world, an internal technology platform becomes sufficiently developed that it can legitimately support multiple products so the value of the end product markets sum for the total addressable market surface of the platform while maintaining a high-end value capture. Solugen has expanded to address multiple different market segments in this way4.

This is the driving idea behind techbio platforms. Similarly, if an internal platform in this case becomes sufficiently robust in its support of product end markets, it has the potential to be turned inside out like AWS, where Amazon turned their compute back-end infrastructure from an internal product enabling cost center to an external revenue generating platform service. It is worth mentioning here that this product approach does not always work if the value capture is not tangible — such was the case with Zymergen when their end product strategy for selling polymers into the industrial chemicals market fell through resulting in an implosion of the company's valuation shortly after their IPO5.

The market size and value capture trade-off are clearly seen in the current strategic shifts at Gingko. Per their CEO's comments, their desire to capture more end product value stunted their ability to reach their projected market segments (e.g., customers balked at their financial and IP terms), and their current strategy to dial back value capture terms is aimed at expanding addressable customer markets6. This trade-off is at the center of the product/platform paradigm.

Capital Costs and Benefits

Pursuing a platform strategy is generally a capital-intensive endeavor because you must build infrastructure that provides significantly more value than any one customer can generate on their own across a sufficiently large surface area of applications. A visualization of the economic difference between the product and platform approach is below.

Figure: A more detailed description is in the text. Platforms are often capital intensive and rely on broad application use cases and generally lower end product value capture. They can be considered wide and shallow. Product-centric approaches have narrower markets, but higher end value capture potential. If product approaches are translatable across internal resources, a platform capability can arise over time (see next image) as occurred with AWS.

Some of the basic principles behind this figure are as follows:

• Pursuing a platform strategy requires building and offering capabilities that span a sufficiently wide application space (or are sufficiently generalizable) to be useful for a large customer base.

• In a generalizable application space, a platform has to be superior to what any individual customer can do for their specific use case - this is the “customer baseline threshold”.

• For any specific end use case there will likely need to be investment on top of a platform’s capability to develop that use case. This is the commercially viable performance threshold. In a developed platform, the gap between the platform’s performance capability and viable product thresholds is low which enables the development of many products in a low-barrier fashion.

• In general, for any given customer, the value ascribed to a platform will be the differential between their alternative capabilities/substitutes and what the platform provides — this is the capturable value for the platform provider. The ability of a platform provider to capture upside is dependent on this gap — the larger the differential between a customer baseline and platform performance, the higher the likelihood of upside value capture.

• The economics of platforms are based on the ratio of platform investment — e.g., performance over application space, and platform return, e.g., value capture potential from customers multiplied by the scope of applicable customers.

• In a product context, the economics are markedly different and simpler because appealing to minimum customer capability thresholds or generalizable application space is not a requirement.

• Investment in building platform capabilities can grow incrementally with internal programs and individual product success in a more sustainable economic fashion. An example of this is how Amazon was built.

  

  Figure: In this case, platform infrastructure was built primarily to support Books as a product. The capabilities were expanded (e.g., transferred learned) to other large library product lines like music and eventually to support a much larger range of products. The increasing generalization of the platform infrastructure ultimately led to its ability to be directly commercialized as AWS.

  In the Product —> Platform approach, the utility of the platform grows incrementally with successive product expansion until it can be sufficiently broad to become commercialized as an independent platform. This is a more sustainable way of building a platform model.

AI & Bio

The framework described above is relatively simple, however, the market specifics will determine the application. For example, how a company distinguishes or creates its performance differential to a customer baseline is often rooted in capital investment and technology development. In the LLM foundational model space, and in the blitzscaling strategic paradigm, this differential is almost exclusively capital and talent delimited - and these become table stakes for an aspiring platform company. A foundation model LLM company almost necessarily requires $Bs to compete. In hardware, this differential is often driven by IP and technology differentiation. In biotech R&D the capability differentials and consequent value capture potential is often rooted in proprietary data and IP.

Monopoly Mergers

In monopolistic settings, this comparison of products to platforms is often irrelevant. For particularly and uniquely enabling products/platforms, there is much less need to make a trade-off between value capture and addressable market. This is the case, for example, with Apple's app store, for a long time, it was the case with Illumina sequencers, and can often be the case with patented foundational technologies. Developing both a dominant product and enabling that product to be a development platform is economic nirvana.

The Value Economics of Different Industries

The determination of value capture to market size is highly industry dependent. In much of the life sciences, large markets are represented by therapeutic blockbusters which are inherently both few and far between and explicitly product oriented. Increasingly, however, value in the life sciences is shifting toward platform infrastructure. For example, manufacturing technologies for cell and gene therapies and AI-driven discovery and development platforms will consume an increasing share of the industry’s economics.

In the technology world, including AI, large markets are often represented by universally utilizable capabilities like compute and storage. While it may be argued that companies, such as the large cloud providers extract significant value from their customers, the value is not generally directly tied to the value of products that are built on top of such platforms so arguably the percent value capture can be arbitrarily low while the addressable market scope is very large.

This simple economic calculus is just one part of the strategic puzzle, the market has a say as we'll discuss below.

The Five Forces

The Five Forces were introduced by Michael Porter in 1979. They seem simple, but most business concepts are. Their utility is in providing a framework for communication. Some have argued that such frameworks are outdated — based on a pre-tech world where markets moved more slowly. Other frameworks have arisen such as blitzscaling, but most of these are flavors of the Five Forces. To keep things simple, they state that the economic viability of a business is driven by the following:

1. Buyer Power — how much say do your customers have in your economics?

2. Supplier Power — how much say do your suppliers have in your economics?

3. Threat of Substitutes — how easy is it to swap out your product for another?

4. Threat of New Entrants — how easy is it for others to enter your market?

5. Competition — this is actually a derivative of the first four and assesses the level of competition in the current market

   

Figure Ref7 Porter’s Five Forces Framework for market analysis

A few specific examples to provide context:

1. Buyer Power — very high in biotech with respect to pharma partnerships, very low in consumer API wrapper products like those from Perplexity.ai. Generally, enterprise customers are high and consumers low.

2. Supplier Power — very high in specific markets, particularly (pseudo) monopolies. During the last 15 years, supplier power for NGS has been very high for Illumina, current supplier power for NVIDIA is through the roof. Supplier power for foundational LLMs is being somewhat attenuated by an oligopoly and being challenged by open-source models, but frontier models still have tremendous supplier power. In keeping with the Perplexity theme, a core component of supplier power, at least in the near term, is access to a quality search engine index such as those built by Google and Microsoft. The specific dynamic is the reason that Perplexity is positioning itself as an alternative to Google — which may pose supplier risk for Google Search API access if Bing were not available. Given the highly disparate market positions of Microsoft and Google in search, it is very much to Microsoft's advantage to provide indexing to companies like Perplexity, at least until Perplexity gains sufficient resources to develop its own index, or Microsoft develops a competing "answer product". This is a line that many AI startups may need to walk if they wish to remain independent in a market of tremendous big-tech supplier power.

3. Threat of Substitutes — the two primary ways to address substitution threats are via customer lock-in or significant product differentiation. Cloud platform companies have achieved this through substantial data egress fees that make it very expensive to move away from their platform. AI and techbio companies aim to achieve this through proprietary technologies or customer integrations. A key dimension to mitigate substitutes is the "Time to Value” — the time for a customer to receive value from your product. In R&D, this time can be considerable and can depend on a wide range of factors that are outside of a company's control. This leaves substantial room for the threat of new substitutes. If the time to value is short (and presuming value is high), substitute products are much less of a threat. It becomes imperative for companies to deeply understand how customer value their products and on what time scale.

4. Threat of New Entrants — this is often called the "defensible moat". In the AI space, for application companies, this threat can be rather high — given the enabling tools that are being developed. Companies that seek to strengthen their defensibility to new entrants often do so with capital, talent, brand/users, technology, or proprietary resources. One of the driving forces behind capital concentration in the tech world is precisely to make it difficult for new entrants to compete in the market.

5. Competition — competition is a derivative of the above four forces and can be considered as a separate category or in the context of the above.

Update: As of July 25 (3 days after this post), OpenAI8 announced their own SearchGPT product which presumably would compete with Perplexity. Perplexity has noted that they use a variety of backend LLMs and are not tied to a single provider which reduces their supplier dependency for foundational LLMs, but this trend will likely continue.

May the Force Be With You

The five forces described above are useful for thinking about product and company strategy for a typical company, however, the real drivers occur when they start to overlap — when suppliers and buyers become substitutes and competitors.

There is some natural tendency toward this — powerful suppliers almost always move into the application space and start competing with their customers. Sam Altman said that it's useless for startups to try to compete with OpenAI on applications in model building but it's their job to do so anyways. The recent class action lawsuit against Illumina is related to their foray into the application space with GRAIL and a litany of anti-trust issues that arose in Europe and other jurisdictions9. For some time, in the molecular diagnostics industry, investors shied away from companies built on Illumina technology not just because of the aggressive supplier leverage, but also because of the potential for Illumina to actively compete with its customers. In the non-invasive prenatal testing (NIPT) space, Illumina acquired a company called Verinata and applied pricing pressure on the top end of the market and supplier pressure on the bottom, effectively squeezing companies that were building on top of their technology.

Similar issues are common in the technology industry. A significant portion of venture capital dollars in Silicon Valley end up going to cloud providers. This trend is being exacerbated in the AI space, with such cloud providers investing directly into equity rounds of high-growth AI startups. While it looks like a substantial endorsement, the underlying dynamic is that a significant portion of that investment will flow directly back to them in the form of service payments. Additionally, by virtue of their investment signal, said startups will likely be able to gather significant outside investor support that boosts the valuations of their equity holdings. This is a soft form of recycling investments and can result in significant upside optionality for the large platform providers. The other option, of course, is to simply directly acqui-hire their competition (in lieu of acquiring the company) as Microsoft did with Inflection and Amazon did with Adept1011. It’s been floated that this strategy of acquiring people vs companies is a way to avoid potential anti-trust issues — which was a hypothesis of why both Microsoft and Apple stepped in and then away from board seats at OpenAI12.

Another market dynamic may be affecting the life science industry where pharma companies can ride public perception to significant investment returns. For example, if a pharma company has a market capitalization of $50B and promotes a partnership with a leading AI company with a $100M upfront payment, if their stock increases by 5% on the positive perception, it would yield a $2.5B valuation return on a $100M investment.

These trends may appear to bode well for high profile startups that can ride the wave of public perception to increase their valuations for both talent and capital concentration. This in itself is one of the most defensible strategies startups can pursue in technology spaces that are fast-moving where value has yet to be clearly defined.

Market Navigation

The market dynamics above are all factors that affect how companies survive and thrive in complex markets, the pace of technological development notwithstanding. There are two key points to take away from this article:

1. Economics drive company success long term. This should go without saying, but it is so very often not sufficiently attended to. The frameworks described above drive economics. The dominant variables are often forces external to any one company. Often, markets are all that matter.

2. Simple frameworks are often the most useful. It behooves executives and entrepreneurs to employ them effectively, for communication, clarity, and execution. Despite rapid progress in technology, fundamentals still remain. I previously wrote a more extensive article about predicting the future13 using simple frameworks.

Closing Notes

I was recently at an event focused on the theme of “3 Founder Unicorns” — the trend mentioned above about AI companies scaling to billion-dollar valuations. This particular event was focused on AI Agents. Toward the end, a question was posed about the biggest challenge that companies were encountering, and almost unanimously it was getting users to understand and adopt products. It was not technology, talent, or capital. Taking a moment to ask the basic questions of what is driving the dynamics in a market and where any enterprise sits in that ecosystem is both an intellectually interesting and practically useful exercise.

Share

For more like this, visit: jasonsteiner.xyz

References:

1

Building Automated Researchers

Jason Steiner

·

June 21, 2024

Preamble and TLDR: This Substack is about providing intuition in a semi-technical manner for readers at the intersection of biology and technology about what is becoming possible. Developments in the capabilities of AI systems and, more importantly, what they are enabling are critical to understand. I have always been a proponent of knowledge managemen…

Read full story

13

The Unfamiliar World - Predictions for the Future

Jason Steiner

·

January 16, 2023

Heuristics for Predicting the Future There is a heuristic that indicates that the best prediction of the future durability of anything is equal to its historical duration. For example, the best prediction for how long Chopin will be popular in the future (at least among a certain group) is precisely equal to how long Chopin has already existed. Certain …

Read full story