In digital advertising, real-time bidding makes on-the-fly pricing decisions about a handful of factors determined by the platform, like geographical region or time of day.
Today, digital ad platform AppNexus is announcing it has modified its platform to accept custom algorithms from advertisers — so that pricing decisions could be driven by anything the advertiser wanted.
The AppNexus Programmable Bidder is in its alpha phase and is available only to selected clients at the moment. Chief data scientist Catherine Williams told me that, at this point, the company has run outside algorithms successfully for one unnamed client.
In a real-time bidding platform, digital advertisers use automated tools that propose prices for ads under certain conditions — such-and-such webpage, this kind of user, that region, which time of day. Publishers have automated tools to accept the best prices and then load the ad.
Most of this happens in the blink of an eye, between when you click for a webpage or an app and when the ad loads.
Williams said that, previously, AppNexus used “only eight or nine variables” that could trigger a real-time change in an advertiser’s bid. For instance, bids might automatically go up or down based on the availability of ad space in a specific geographical region or how recently an ad has been shown to a group of users.
If the advertiser really wanted millennials on weekday evenings, they would set up a fixed campaign. But if more millennials showed up just as price-bidding was happening, and the platform didn’t support bid changes for that factor, too bad. This means the advertiser might not be able to immediately bid higher for the users it really wanted.
Of course, advertisers or their agencies could always build their own real-time ad bidder to handle more or different bidding variables, but that involves a considerable amount of time, expense, and technical complexity.
“Suppose you’re a marketer with a whole bunch of data about your users,” Williams said, “but you haven’t been able to tie it to meaningful advertising.” That user data could represent micro-segments of users who, say, often click on hotel ads on summer weekends when the ads are accompanied by discount coupons. The previous AppNexus platform didn’t allow real-time pricing to respond to many kinds of user variables.
But now the hotel advertiser can supply its own algorithm that is designed to pay more, immediately, when those targeted users show up.
The rule of thumb for digital advertisers, Williams pointed out, is “you want to bid your expected revenue.” In other words, pay more when you see the opportunity so that the ad shown is yours.
To accomplish bring-your-own, AppNexus has built its own programming language and parser. Advertisers create a logical model of their own decision trees — when this shows up, do that — translate it into AppNexus-ese, and upload it through the platform’s API.
CEO Brian O’Kelley said in the announcement that he’d like to see advertisers testing their own algorithms against AppNexus’ in what could become a robust ecosystem of different, fine-grained ad-targeting chunks of code. The company also raised the possibility that startups might specialize in creating rentable algorithms.
No other digital ad platform is undertaking this kind of roll-your-own coding, Williams said. She suggested that, if it catches on, advertising could become more user-relevant, as it would be honed by the combination of many variables for which the right advertiser has paid more.