Organizing Chaos

Today Dan Farber paraphrased some comments by Munjal Shah about the Riya.com business model. The comments refer to extracting value from user-generated content and rewarding the 'long tail.' Those of you following Riya know they are in the process of launching a site to automatically tag photos with faces and text. These metadata make it easier to organize and share photos. They also make it easier to search for photos that have value to you.

The principle behind this is as old as the commercial Web -- all of nine years. Brooks Fisher (now a VP at Intuit) was the first person to invent and implement the concept of keyword advertising back at Infoseek in 1996. (We were doing all of 2M page views/day!) That was when it became clear that as the audience grows, the value of the ad inventory expands exponentially. Targeting equates to value. Targeting specificity increases as volume increases, lifting the value of the entire inventory. It is a virtuous cycle. (This is also part of why search has a winner-take-all dynamic, but that's for another time.) Adding the structure unlocks the value.

This is the power behind the Riya business (and another soon-to-launch business in which we have invested). Adding structure unlocks the vaue. It attracts users by improving the experience. More users generate and attract more content. Content expansion increases the value of targeting. Value is extracted by making the content more searchable, and ultimately, reusable. Scalability and consistency are key to unlocking the value. Scale and consistency limitations are also why I think manual tagging regimes have a limited role in the Web platform.

If this sounds like a machine-learning bias, guilty. If it sounds like the core of how search engines monetize, that's no accident. Search engines are the most mature. I prefer to think of them as targeting engines.

The generic rule - add structure and mine for money by organizing the chaos. We'll see it played out many, many times over the next several years. Because it works.

Comments

[Gaurav]

I am not convinced that automatic tagging of photos will enable better targeting of advertising. The things that can be tagged automatically are too specific (people's names, or maybe places). Beyond that manual tagging has to come into play (like eveybody attending Les Blog might be using the same tag on their flickr uploads). And even with manual tags, from an advertiser's perspective, photo search doesn't seem too compelling - except maybe to travel businesses. When I search for LesBlogs on Flickr, I get ads for cigarettes! :)

But then I am not on Riya alpha and perhaps they really have figured it out all well.

[Steve Krause]

Great blog, Peter.

A historical note: In 1996-1997, the targeting/value relationship was dampened because ad inventory was growing faster than many sites' ability to sell it, and there was no Overture/Adwords-style marketplace to clear the low-end inventory. This led to a situation where, instead of delivering more clicks via targeting, it was easier to simply oversupply impressions, which got the extra clicks via brute force.

[Steve Krause]

(By "oversupply impressions" I mean "deliver more impressions than the customer paid for" to make the CPM paid seem justifiable via the clicks generated.)

[Peter Rip]

Absolutely right, Steve. The real problem was CPM instead of CPC. The early "media model" of CPM was a flawed paradigm that underexploited the web's ability to deliver an action instead of mere "impressions."