Algorithms are a commodity

by greg on September 22, 2009

Andrew Chen’s posted a worthwhile rant about ad network targeting claims.

I believe optimization algorithms are a commodity. They’re not trivial – I’m certainly not claiming I could write one – but you hire a good quant and you’ve got one. Hardest part is hiring the quant. There may be some minor differences between algorithms, but real competitive advantages come from better data.

To illustrate, check out DataXu’s TechCrunch 50 presentation. Smart guys using real-time bidding plus an algorithm fueled by test buys to cherry-pick the impressions from ad exchanges that are most likely to convert for their CPA offers. But many companies will try to do the exact same thing. If major differentiation between them occurs, it won’t be due to the algorithms, it’ll be due to their inputs.

{ 2 trackbacks }

Yardley.ca >> Algorithms are a commodity « ecpm blog
September 22, 2009 at 7:05 am
Rubicon Project Gets $9 Million; Microsoft On Holy Grail; Permuto Launches Display Ad Platform
September 23, 2009 at 4:16 am

{ 6 comments… read them below or add one }

Brian O'Kelley September 22, 2009 at 8:37 am

Good point, as always.

b

Mike Baker September 23, 2009 at 5:43 am

Greg, your points are valid – as far as they go. Static model / algorithm creation for ad targeting is well trod ground. And data and proper inputs are an important piece of any predictive solution. The hard part (and what we are trying to do at DataXu) is dynamically selecting the right data features, implementing the most effective set of algorithms on the fly, and automatically “course correcting” to web / ad traffic turbulence. As a side note, in my view the new commodity in the interactive ad space is likely to be third party cookie-tethered consumer data, which anyone can buy and use from 10+ credible sources and growing ….

Tim Ogilvie September 23, 2009 at 2:10 pm

Very true – Chris Dixon made this point as well (though more generally than in regards to ad optimization)

http://www.cdixon.org/?p=340

Key question becomes where & how to find opportunities for unique data….

Darren Herman September 23, 2009 at 5:13 pm

Greg, dead on as always.

Rob Leathern September 24, 2009 at 12:10 am

A good point. But does this just mean that scale wins = since more data?

greg September 24, 2009 at 8:16 am

Rob — no, I’d say it’s the usefulness and exclusivity of the data that’s key. Scale is less important in an exchange environment since you no longer have to buy massive amounts of inventory to find the small proportion of impressions you’re able to effectively monetize.

Leave a Comment

Previous post:

Next post: