At Crosshatch we’re fascinated by tech cycles and how and when they create opportunity.
On repeat for us has been Dani Grant and Nick Grossman’s “Myth of the Infrastructure Phase,” a historical study of how “apps” like the light bulb or the airplane inspire infrastructure like the electrical grid or the airport –– not the other way around.
A big part of why this is even a topic of conversation is that everyone now knows that "platforms" are often the largest value opportunities (true for Facebook, Amazon/AWS, Twilio, etc.) — so there is naturally a rush to build a major platform that captures value.
This AI cycle is no different – many are endeavoring (we are!) to find a seat delivering infrastructure for this new age.
But a teased infrastructure phase can be a trap: breakout apps come first and inspire infrastructure before infrastructure has a right to rise. And the evolving capabilities of AI introduce new challenges to sizing where in the cycle we are.
For instance, AI apps with problems like limited context and reasoning of 2022 that limited AI memory and logical capabilities seemed to imply the righteous rise of infrastructure. But models themselves have improved now with over 100x the memory and reportedly near-human capabilities in some benchmarks. Problems you might solve for customers with infrastructure today could be resolved by better models tomorrow.
We saw this in February when Klarna reported resolving nearly two thirds of customer queries with their own application of OpenAI model calls. Despite the availability of many startups shipping customer service SaaS solutions to do this on behalf of Klarna, Klarna built this themselves.
What we see in the sequence of events of major platform shifts is that first there is a breakout app, and then that breakout app inspires a phase where we build infrastructure that makes it easier to build similar apps, and infrastructure that allows the broad consumer adoption of those apps.
So now for us, Crosshatch is building a new identity layer – personalization infrastructure for transferring state information across applications – to make AI that remember us across distinct interactions available on any surface with any model in a way end-users control.
And so we wonder: Are we falling prey to this natural rush to build infrastructure before a breakout app unlocks the right to build supporting infrastructure? Or does the history of personalization and networking somehow resist this pattern – maybe we’re bound by some other framework?
In this blog, we attempt to map waves of apps => infra => apps => infra … to the history of personalization and networking.
What we find is that the history of personalization and networking also follow these waves. And if we’re reading everything right, personalization is just about to enter an infrastructure phase.
A history of personalization and networking
We set out to study the history of personalization. We include networking in our study because personalization is an act of aligning objects to an individual’s tastes or requirements. Since candidate objects to be personalized are administered (and sometimes owned or rightfully governed) by third party entities, the act of personalization on the internet requires some networked communication, and is why we include networks in our study.
We start early with the industrial expansion of the railways. “Apps” Montgomery Ward (1872) and Sears (1886) shipped catalogs of items with hundreds of pages. Most consumers had only shopped at local general stores within traveling distance. A catalog with thousands of items offered a first glimpse into personalization, expanding consumer choice to find items among many that were most suitable to them.
This led to the rise of logistical and distribution infrastructure. To meet the demand generated by these catalogs and ensure the delivery of goods, robust logistical systems were developed like UPS (1907) and USPS’ Parcel Post service (1913). Market research infrastructure like Nielsen (1923) to help businesses better understand their customers.
But as the distribution of goods grew, brands “apps” like Proctor and Gamble, Unilever (1929) and Colgate Palmolive (1928) looked to differentiate themselves and their individual product lines. P&G created a brand management system (1931), where each brand manager controlled all aspects of their brand – like running a small business. Each brand manager studied their product’s audience, and aligned their brand’s positioning with the studied needs of the customer. These apps pursued new messaging in the forms of sponsored radio and soap operas. Apps began to talk to their audiences directly instead of intermediated by a catalog.
Then there was the wave of broadcast infrastructure with NBC (1926) and the FCC (1933) that set the stage for brand-driven broadcast personalization. Brands like Coca Cola and P&G refined their brand identities and product offerings on broadcast radio (1927) and later TV for deep connection rather than broad reach. These apps personalized through messaging that aimed to set the zeitgeist, transforming simple products into symbols of lifestyles and aspirations. Infrastructure like the RCA color TV (1954) enabled Mad Men to help apps like Marlboro to espouse aspired lifestyles (“Marlboro Man” 1954) leading customers to aspire or relate to brand personas. As demand for personalized messaging rose, infrastructure like Zip Codes (1963) and OCR deployed at the USPS (1965) allowed post offices to read and sort up to 42,000 addresses per hour.
Infrastructure like Epsilon (1969) and Acxiom (1969) enabled brands to further scale their mailings, enabling mailings personalized to each user.
With the internet came apps like messaging (1970) and email (1972) that sparked an infrastructure wave including Ethernet (1973), TCP/IP (1973), ISPs (1974) and Oracle’s Version 2 (1979). These advancements paved the way for the emergence of web portals like Prodigy (1990) and AOL (1991), which in turn fueled the development of search engines and web browsers in the early 1990s.
Then there is the next wave of apps, which are services like American Express Membership Rewards (1991) and sites like Ebay (1995) and Amazon (1996).
There was a problem though.
Every visit to a site was like the first, with no automatic way to record that a visitor had dropped by before. Any commercial transaction would have to be handled from start to finish in one visit, and visitors would have to work their way through the same clicks again and again. It was like visiting a store where the shopkeeper had amnesia.
This lead to new infrastructure: the cookie (1994) was born, with a patent filed in 1995 and deployment in Internet Explorer in 1995.
This small piece of data stored on users' computers became the critical tool for enabling websites to remember returning visitors, their preferences, and the contents of their shopping carts, from their site and even others.
Then came the era of enhanced e-commerce and personalized digital experiences. The cookie enabled sites like Amazon to recommend products based on previous visits, and eBay to enhance user experiences by remembering login details and viewing preferences.
Web analytics infrastructure company Omniture (1995) rose to help websites better understand their audiences and traffic. Next came apps like Expedia (1996), Realtor.com (1995), Zappos (1999), Hotwire (2000), TripAdvisor (2000), Trulia (2005) and Zillow (2006).
Then there was the wave of infrastructure like AWS (2006), the iPhone (2007) and Shopify (2006) and the cycle continued into apps like Amazon and eBay for iOS (2008), Spotify (2008) Instagram (2010) and Pinterest (2010) and then infrastructure Segment (2011), BigQuery (2010), Epsilon Agility (identity) (2012), Acxiom Identity Solutions (2012?), iBotta (2011), Fetch Rewards (2013), Databricks (2013) and TensorFlow (2015).
This lead to AI applications like ChatGPT, Pi, and Claude with millions of users. And privacy-focused browser updates like Safari (2021) Private Relay and Chrome (2023) Privacy Sandbox.
The history of personalization follows other tech cycles, with an interplay of apps to infra to apps to infra and so on. AI has hastened the pace of innovation, with multiple huge leaps happening in a single year.
The next infrastructure phase in personalization
Like applications before them – Montgomery Ward to Amex to Ebay to Pinterest to Zillow – all inspiring an infrastructure phase that democratize and scale the innovations and experiences that enabled breakout apps.
AI applications like ChatGPT (2022) and Pi (2023) gave us a taste of what a Personalized AI could be. Now every board is asking leaders how they can include AI in their business. Internal enterprise applications are under way but consumer applications have lagged.
We were kids back then, but consumer AI today feels to us like the internet must have felt in 1994. AI across domains lack state. Each new AI we engage with – whether on an app, a swing-set, a clip-on, or a browser – doesn’t know what sessions we just had with the last AI, as governed by someone else. Our sessions with one are never known by the next. Like the shopping sites of 1994, each time we land on a new AI powered surface we have to work our way through the same context again and again. Introduce ourselves again and again. Everything must happen in a single service.
The breakout app in 2022 was ChatGPT.
Netscape shipped the cookie.
Crosshatch is shipping the wallet.
