|Intro Brief Campaigns Words Bids Ads Site Prospects|
What am I bid for these 150,000 words? Let’s say you had 5 or 6 words you were bidding on – and you could see how they were ‘working’ for you. It wouldn’t be too difficult for you to figure out you could raise the bids a little on the ones that were doing particularly well for you – and to lower them on those that weren’t.
Now imagine you were doing this on 10,000 keywords in parallel. Or 150,000. Or even 750,000 as we have to do on one of our clients. And the problem becomes just a wee bit more complex.
Now add to it the fact that there is no easy way to predict how an will do at any position - indeed we have some belief that many ads do better at the bottom of a page (say, position 10 on a page with 10 ads) than at position 6 or 7. So the “yield curve” as our resident mathematicians put it, is not linear.
We do a few things – we try to predict the “clickability” of a keyword in advance, using some interesting insights on how search engine result pages work. And we use pretty advanced mathematics to figure out the optimum bidding combination of bid prices across all keywords in parallel, to get the best Return On Investment.
This involves knapsack algorithms, singular value decomposition, eigenvalue computation and other terms that haven’t come close to the advertising and marketing business in a long time. If any of this excites you, and you’re a potential client, please go here. If, as it’s more likely, you’re an optimization person or just mad about mathematics and marketing, do apply for a job here.