Saturday, September 14, 2013

    Staring into the Future: Programmatic Buying and Selling on Mobile

    Recently, I've had the fortune to work on a new-ish technology that is making its way into mobile.  That technology is known as Real Time Bidding, or RTB for short.  It is the programmatic buying and selling of ad inventory and ads and it is a truly amazing technology.

    RTB has existed for several years on the web world, but only in the past year or so has it made its way into mobile.  Programmatic buying has existed for longer in places like stock exchanges where the majority of buying and selling is done programmatically.

    So how, exactly, does RTB work?  At its simplest level, the entire process starts with the opportunity for an impression.  Let's use our friend Sarah to illustrate the process.

    Sarah has just dominated a stage in her favorite competitive shooter on her phone, "Kill 'em All".  She is between levels and as is common with games it's an opportune time to show Sarah an ad while she waits for her next opponent to enter her killing field.  This is where our process starts, with the opportunity for an ad impression:  there is supply of inventory to be filled.

    The game she's playing is hooked up to an RTB exchange and, likely via an embedded SDK, sends a signal to the exchange to let them know that they need an ad.  It's likely that there has been some limited information sent along about Sarah, or that the exchange knows a bit about Sarah from previous transactions.  At the very least, it is known that Sarah is an avid gamer from the games she plays and that she's likely between the ages of 18 to 25.  This information is packaged into what is known as a Bid Request and is sent off into an auction to be bid on.  Part of that Bid Request is the info about Sarah that's been collected.  It will also include info about her device and the type of ad that she is ready to be shown (in this case, it's a full screen interstitial).

    The Bid Request goes out to various DSPs (Demand Side Platforms).  These DSPs see the request and immediately begin processing it.  In lightning quick time, they are prepared to serve an ad that is targeted at Sarah.  This ad goes out in a Bid Response and most times will contain the ad to be shown along with the price they're willing to pay for the impression.  This price is effectively a bid on the inventory that Sarah's ad opportunity represents.  In general, whoever bids the highest will win the right to show the ad.

    The exchange at this point will pick the winning ad, generally based on bid price.  The exchange will then send the ad on to the Kill 'em All game, which will then show the ad to Sarah.  As soon as the ad renders on the screen, what is known as a pixel will fire off.  This pixel notifies the winning DSP that they've successfully shown an ad and that they won the auction.

    Here is the thing that I think amazes me the most about this process.  Everything that I outlined above happens in milliseconds, typically less than 250 milliseconds (that's a quarter of a second).  We have gone from the need for an ad to showing the ad in less time than it takes to move your mouse and click something.

    The other thing that amazes me is that all of the decision making has been done programmatically.  There are signals within the request that the DSPs can use to determine how badly they want to show the ad.  Using machine learning algorithms and bidding strategies, the DSP can effectively maximize their impressions to those that will earn them the most money.

    On the side of our app, Kill 'em All, they take a cut of the impression bid and share the rest with the exchange.  The DSP pays out as soon as the impression is shown.  If, for whatever reason, and ad is delivered down to Sarah's device, but not shown, no money changes hands.


    Anonymous said...

    Здравствуйте! Я студентка 4 курса факультета психологии.

    Для написания дипломной работы ищу людей, проживающих гражданским браком , и молодых людей и девушек, в возрасте от 20 до 30 лет.

    Хочу предложить несколько методик для заполнения, гарантирую описание результатов каждому. Если вас это заинтересовало, жду писем на адрес: Спасибо
    [url=][color=#E4F4FE] [/color][/url]