Programmatic Explained: How DSPs Work


Programmatic exchanges, second price auction, supply side — the terminology of mobile ads has changed in recent years. In the process of making advertising smarter, the industry has seemingly also made it more confusing.

The culprit? Automation. Advertising is increasingly handled by algorithms. But not everything can be automated. Advertisers and publishers still need to understand the basics to maximize the value they get out of the market.

Following up on our popular Crash Course in Programmatic, the Chartboost team will now share with you more insight on demand-side platforms from Betty Wan, Head of Business Development & Partnerships at Liftoff, a leading performance-based marketing and retargeting platform for mobile apps.

How a DSP is different from an ad network

A DSP — short for Demand Side Platform — buys ad inventory on behalf of advertisers based on pre-set criteria. To put it more simply, DSPs match their clients (advertisers) with the right users on the right publishers. A DSP has no ad inventory of its own, instead functioning like an air traffic controller directing ads based on previously-collected data and predictions of future behavior.

Why would anyone use a DSP?

In the dark ages of digital, humans did the work of DSPs, writing up ad proposals and insertion orders and sending them to publishers. Since the process was manual, advertisers back then had to wait up to a month before optimizing. With a DSP’s algorithmic approach, optimization is possible in almost real-time, negotiating and serving ads faster than any human can. “Machine learning is a​n active​ model that’s optimizing at every impression so it’s doing far more bookkeeping than a human could do,” says Wan. DSPs can also attribute performance metrics and tweak spend in milliseconds.

Machine learning, a technique in artificial intelligence, is increasingly used to give DSPs a 30,000-foot view into user behaviour. Liftoff has access to user behaviour from over two billion devices — essentially, all the active devices in the world — and watches for patterns that it then uses for targeting. “When our machine learning locates a user across the exchanges who fits the profile of a high-value individual, Liftoff then bids on the opportunity to serve that person an ad based on the data-predicted likelihood that the user will click, install, and ultimately become a customer,” explains Wan.

How can publishers make their game more attractive to DSPs?

Liftoff’s first piece of advice for publishers is to focus on the quality of their game, particularly on long-term engagement. In other words, advertisers and good game developers are generally aligned in their goals. Wan also recommends that developers reach out to their supply-side partner (a company such as Chartboost) for feedback on how to further improve ad placements for players. In cases where a developer is using a waterfall, Wan recommends allowing open competition from bidders rather than forcing potential ad sources into preset positions in the waterfall. Simply put, DSPs believe that among reputable companies, the highest bid should win.

Data sharing is, of course, also important to DSPs. “In a world where programmatic is getting bigger and bigger, it’s important for all ecosystem partners — including publishers, SSPs and DSPs — to use OpenRTB standards so that we have a unified and consistent way of transmitting, receiving, and responding with data,” says Wan. Developers who have additional unique data to pass on, like session score or GDPR-approved demographic data, can also benefit.

What are DSPs doing to combat fraud?

Fraud is a persistent problem in mobile. Yet the industry still struggles to mount a unified defense, partly because fraudsters are also taking part in the tech race with some advanced fraud techniques powered by machine learning.

Enabled by the same powerful software used to knit together millions of data points into patterns of user behavior, DSPs do their part by catching discrepancies. “​Our user database contains the mobile usage history for almost every active mobile device in circulation in all key mobile markets. This allows us to accurately model activity that’s normal and to detect behaviour that’s abnormal,” says Wan.

A DSP can also respond and shut down suspicious channels faster than human marketers. Wan outlined a few of the types of fraud that DSPs like Liftoff watch for:

  • Low-intent publishers, ​such as apps ​with abnormally poor downstream conversion rates or click-intent metrics
  • “Hyperactive” users, or users that report an unrealistic amount of advertising activity, like clicking on hundreds of ads per day
  • Non-human traffic originating from data center​s
  • Anonymous traffic – Liftoff identifies all users by their device IDs to hold all sources of traffic accountable

Of course, whether the topic is CPM bidding, fraud, or data sharing, DSPs are just one part of the fabric of advertising. Later in this blog series, we’ll take a look at the publisher end, where the supply-side platform (SSP) is helping to organize and auction off inventory.