What Is A Demand Side Platform or DSP?

Last week, I wrote an article introducing and describing data management platforms (DMPs). While DMPs continue to merit important discussion in the changing world of programmatic ad buying, it is important to recognize that this discussion cannot be completed without talking about DSPs. In fact, this discussion is a key prelude for future posts which will cover the evolution of real time bidding platforms, DMP-DSP architecture, and a whole host of issues that I am currently working on & fascinated by. DSPs represent the activation side of the programmatic ad buying process. At the end of the day, when it comes to advertising and analytics, activation is what I am most concerned with. This is because data and segments are, in my mind, only useful if they can be used to impact an organization's ability to think about, plan for, and optimize the different channels, categories, or processes that exist within their business. To that end, when one thinks about the DMP-DSP relationship, without a solid link between the two entities, a DMP ends up being simply a persistent repository of information that provides only surface insight.

From a high level, describing a DSP is simple. A DSP is a piece of software that automates & streamlines the purchase of advertising & media. DSPs help streamline the process of buying advertisements online by interfacing with any number of ad exchanges that exist. That said, in order to understand more deeply how a DSP works, one has to simply step back and reflect on the fundamental process that surrounds the purchase of an advertisement through a programmatic channel. Essentially, within any particular advertising exchange, you will see publishers take advertisements and expose those advertisements to DSPs for purchase. These DSPs determine the price that will be paid for the advertisement by the buyer, with the final price being determined during a real-time auction.

It is important to recognize that this process is remarkably similar to the historical media buying process that many in the traditional offline media arena are already familiar with. In this process, advertisements are bought and sold by sales staff from human buyers who contact sales associates by phone, fax, or other methods. Without question, it is not hard to see how such a human oriented process could be prone to inefficiency, slow, and difficult to scale. These drawbacks are the very reasons why programmatic has been so attractive to many in the digital advertising space. In the programmatic environment, the buyer is the DSP who bids for the advertisement against other competitors within the exchange (rather than through a phone line). This removes the manual aspects of the bidding process, creating a purely programmatic process that fascinates me, particularly based on my previous & current work in computational finance (as well as interest in high frequency trading technology & architecture). Moreover, thinking about time, the historical process, which contained a human agent, would be regularly measured in hours and days. This is contrasted by the programmatic process, which can be executed at scale using various technology pieces, with completion being measured in milliseconds or faster. This is just one powerful example of how technology continues to revolutionize the marketing and advertising space; a revolution, that will no doubt continue in the years ahead.

Now that you have a good understanding of what a DSP does, the next point relates to the optimization of the DSP's decision making ability. As mentioned originally, the problem of determining the bid price for an advertisement is a "simple" optimization problem that is covered in the undergraduate & graduate classes that I teach. Determining the thresholds that should be paid, as well as the type of advertisements to be purchased, should arguable reflect fundamental information about about the specific customers that an organization wants to meet. These assumptions can be improved by segmentation. In fact, this is where the DMP can offer tremendous value, since, as a consolidated database (which can ingest & aggregate data from many different sources, including third party), the DMP provides a single source of information that can allow the DSP to make better decisions. In fact, using skills rooted in statistics, one can build models and systems that can enable the advertising buying process to reach more of the right customers in different environments, while also potentially reducing the short and long term costs for an organization in a substantive way. These potential benefits make investments in data science expertise truly worthwhile, and also provide a way for marketers and even human resource teams to also measure the effectiveness of their data science / digital media staff who help mediate this process.

Today, there are a number of companies doing some very exciting things in the DSP space. There are also some very interesting innovations going, one of which is talked about below. Currently, the DSPs that I am concentrating on most frequently include X+1 (Rocketfuel), MediaMath, DoubleClick (Google), Adobe Media Optimizer, and a few others. I spend most of my time thinking about how to build robust software architectures that optimize these tools and connect the data within them. I also spend significant time trying to use statistical methods to improve the quality of segmentation being done before data flows into the DSP. One area that has caught my attention in the past few weeks relates to the evolution of channel type specific DSP's, including DSP's that exist for mobile networks. I have been digging into the work being done by Smatto, Human Demand who was recently acquired by IgnitionOne, and others, particularly as more and more customers or donors interact with various organizations using mobile devices.

Without a doubt, the DSP world is one that is tremendously interesting. It is constantly changing, complex, and requires marketers to continuously learn in order to ensure that the decisions they make about any DSP put their organization in the best position for long term success. If you are currently in this space and ever want to chat with me about DSP's in general, DMP-DSP marketing technology architecture, or marketing strategy relating to all processes touched by this larger group of technologies, please don't hesitate to reach out! As always, I am very happy to share what I know, and am even more excited to learn from those who are also active in this ever evolving space!