Search Specialization Is Inevitable: A Darwinian View of Graph Search

post_image240113b

Like everyone — well, at least everyone in the tech world — I’ve been reading about Facebook’s new search initiative, which has been covered with the same breathless intensity as the discovery of a new vaccine or new planet.

As you might expect, much of the coverage has been one-dimensional, focusing on the product features of Graph Search (a rather inelegant name) and the consequences for everyone from Yelp to Google.

I have a different perspective to share, one that comes from looking at Facebook’s announcement from a market structure and market evolution point of view. In other words, I want to step back from a narrow, tech-centric view of this launch, and put it in a larger analytical context. After all, search — as we speak about it — is a market, just as coffee and department stores and automobiles and thousands of other categories are markets. 

One phenomenon of markets is that they oscillate between specialization and non-differentiation. This is unlike Darwin and evolutionary biology, where all living things are on a one-way march to specialization.

An example of market vacillation between specialization and non-differentiation can be seen in specialty stores and department stores. Until the late 19th century, every retail store was a “mom and pop” store selling one type of commodity. Then department stores arrived, offering a dazzlingly wide array of merchandise all in one place. To survive, small stores had to evolve and offer something different, beyond just convenience,  such as better service and more personally curated merchandise.  

The bounce between department stores and small retail stores has continued to this day. Experts have proclaimed the inexorable success of the all-inclusive department store and the demise of the boutique, only to be found later bewailing the end of the department store era.

To bring this excursion back to Facebook, the history of search has largely been a march to non-differentiation (with some notable exceptions). After all, the whole point of Google’s PageRank innovation was to bring an algorithmic solution to finding the most relevant result in any category, a technology which on its face appears to obviate the need for any specialization since its approach enabled it to be, paradoxically, broad and narrow. The triumph of the horizontal.

Over the years, there have been attempts at vertical search, many in the mid-to-late 2000s. The argument behind this layer of specialization is that a product focusing relentlessly and innovatively on a specific category can do a better job than a product providing information on all categories.

Health was a vertical that saw a lot of action during this time because of the size of the market and the perceived complexity. There was (and is, I guess) Healthline, which raised an additional $14M as recently as 2010. Another example is the now-dead iMedix, which tried to combine health with a social layer. (Sound familiar?)

One form of vertical search which has succeeded, although you probably don’t look at it that way, is the vast and unwieldy category of product search. Here, Google has met its match with Amazon. As a story in Forbes reported last September:

  • Forrester Research found that a third of online users started their product searches on Amazon compared to 13 percent who started their search from a traditional search site; and 
  • comScore found that product searches on Amazon have grown 73 percent over the last year while shopping searches on Google have been flat.

Importantly, we’ve also seen the specialization of search within the Facebook ecosystem. Bing — which is also Conduit’s search partner — powers what is essentially a vertical search that matches a search term against “likes” of your Facebook friends, as well as profile information. Graph Search is a basic evolution of Bing’s vertical search on Facebook (although it remains unclear whether Graph Search is actually powered by Bing or not). And for a while now, Bing search results have been populated with Facebook results, which is another step in the evolutionary process of undifferentiated search becoming specialized.

Of course, Conduit’s Community Toolbar innovation — which broke the search functionality out of the constraints of the search home page and made it instantly distributable across the web — can be viewed, in this framework, as a move toward search specialization because it allowed long-tail publishers to embed search into their community offering as part of their user experience. And stay tuned, Conduit’s flagship toolbar will soon be evolving into even more specialized search tools.

When you put all this together, it’s clear that we’re now at an inflection point. The arrival of both crowd-sourced content and the social revolution have hastened the evolution from the non-differentiated search market structure — that started with Yahoo! and Alta Vista and then gave way to Google — to specialization. TripAdvisor and Yelp are, essentially, vertical search engines for travel and restaurants. They provide other functions, but at their core that’s what they do. 

Facebook’s Graph Search is specialization that blends elements of vertical and horizontal search. The vertical component is the people, places, photos, and interests that Zuckerberg described in his launch pitch. The horizontal dimension is the Facebook social graph: your first-degree connections and beyond. When we look at Graph Search through this optic, it becomes clear that the inevitable market dynamic of specialization vs. non-differentiation has fully arrived in the search category. 

So it’s not a matter of whether Facebook has created a Yelp-killer or not, but how these parallel ecosystems will coexist and evolve. I’m convinced that there’s room in the search world for well-executed social-vertical searches, but it’s not as simple as saying that the recommendations and choices of your friends on Facebook are always and unquestionably superior to a general Google search. 

Let’s take the Mexican restaurant in Palo Alto example that Zuckerberg used. As he explained it, using the taxonomy of “people, photos, places, and interests,” Graph Search would map these intersections to identify Mexican restaurants that his friends had visited. It doesn’t take much analysis to point out the flaws in this system. My friends may have lousy taste in Mexican food, or perhaps they only had a cocktail at the restaurant, or maybe they were dragged there against their will, or they may have been bribed into a Facebook “like” by the seductive treasure of a free margarita. It’d rather do a Google search using Eater, Chowhound, and other sources, and perform my own DIY triangulation.

Some of the other examples provided also feel like a solution in search of a problem. For example, Graph Search can identify the best-liked photo of Mr. and Mrs. Zuckerberg to decide which one to use on a Christmas card. This seems like a giant stretch — and it’s not even representative, since your most visually perceptive friends may not be prone to the “like” ritual or they may have missed the post entirely. Given all the possibilities of Graph Search, it says something that they couldn’t concoct a more compellingly useful example of their own feature.

But it will get better; I have no doubt of that. There is a vast and exponentially growing amount of useful data that resides in your social network, which can be mined and re-assembled to answer specific questions. As Graph Search’s specialization evolves, consumers will, in parallel, evolve new search behaviors that match the right engine to the right task. We’ve seen this in messaging — where specialization has created email, texting, and further refinements like Snapchat and its vanishing photos.

At the same time, following the path of evolution, when products of specialization no longer become useful, they remain as vestigial organs, as in the appendix and the modem. Graph Search will gain maturity and become a useful way to harvest information from Facebook; new search platforms that we haven’t anticipated will emerge; and Google itself will morph based on the “evolutionary pressure” of the environment. All this is fun to watch because it happens in weeks and months, not in the millennia of evolution.