I recall the first time I heard the term "DSP." It was in a meeting with the Right Media team at Yahoo in New York in 2009, and the team said to me, "We now have a name for what you guys are!" and that term was DSP -- demand-side platform. I knew my company, XA.net, was not an ad network or an ad agency, and so our new moniker emerged, which seemed to describe something different from those well-understood models.
One of the key distinctions (to me at least) from a network seemed to be that the DSP would align its interests on the side of the buyer, and not the seller. In the intervening time, however, many companies have jumped on this term to call themselves a DSP, perhaps sometimes to overcome some negative connotations associated with being an ad network, to take advantage of newly focused attention in the space, or, in many cases, because they think it truly best describes what they do.
A demand-side platform is a technology/intelligence layer on top of display inventory and display-buying systems that makes media buying more accessible, easier, or more efficient. DSPs evolved as a response to a need by advertisers and agencies for more streamlined management of display media buying specifically within ad exchanges like the Right Media Exchange (RMX). These exchange systems were new and often difficult to use, and needed simplified interfaces. Luckily (in the case of RMX) APIs allowed campaigns to be created directly in their systems and reporting to be gathered without a human being having to log into those systems and use their complex user interfaces. DSPs and ad networks also started working with emerging data companies like BlueKai and eXelate, which wanted to offer data to media buyers without having to sell them media (unlike the early behavioral ad networks like Tacoda), and DSPs started to show they could help with that.
Then in 2009 the concept of real-time bidding (RTB), or "callout bidding," emerged, where instead of buyers designing a campaign in someone else's system, they would instead get a feed of information about specific impressions upon which they could then bid. With DSPs already focused on adding technology on top of media buying, many of them started to experiment with RTB and are now in the process of building systems around this still-nascent way of buying media. It eliminates headaches around managing lots of different audience pools for various exchanges and systems (instead, the real-time bidder gets the chance to match up users to their own database and then decide whether to buy the impression), but creates headaches of its own with new technology needs that many legacy technology platforms have seemed slow to adapt to.
Two business models and hybrids emerged thereof for DSPs:
- Selling a platform to aggregated media buyers/sellers (like agencies or networks themselves) as a software license model, on a CPM or some combo thereof
- Buying media on behalf of agencies, networks, or advertisers directly in an optimized way
Naturally as the needs of the market have evolved, DSPs have evolved too, and this has certainly created additional confusion. For example:
- A focus by many DSPs on enabling agencies to buy on ad exchanges or RTB has become augmented by some, allowing additional buying across publishers directly, streamlining buying workflow, and making some DSPs look a lot more like technology-powered agencies.
- Some have shifted from rabidly working only with agencies (so as not to threaten them) to include ad networks as customers, then to include advertisers they deem to have no preexisting agency relationships.
- Some DSPs have chosen to focus on performance-oriented advertisers (and perhaps even allowing CPA-based buying), while others instead relied on enabling agencies with technology for them to get additional cheap exchange impressions for brand campaigns -- many of which ad networks had previously supplied to them.
- "Agency trading desks" arose and have largely utilized DSP technology to power their media buying, instead of creating their own technology. Odd situations persist where some agencies are mandated to use these trading desks for buys but are still buying directly from DSPs and exchanges themselves.
Some DSP executives have argued online over the last six months about what the true definition of a DSP is, and here are (paraphrased) some of their points (with my take):
- DSPs must have fully self-service interfaces. It's also been said that a DSP should expose any feature or tool that an ad exchange or data provider provides. Having a robust and functional user interface seems hard to argue with, though the degree of functionality is another issue. If the interface is a superset of all features available in all media buying systems, it will be impossible for the DSP client to use sensibly without help. Simplification should be one virtue of these platforms.
- DSPs are fully transparent, starting with pricing and fees, and should not mark up data or media costs without their clients knowing. The DSP must remain neutral and have no allegiances to any publishers or vendors. These are both important aspects, yet they are still evolving, as site-level transparency is sometimes limited by exchange contracts or publisher relationships. Fees and pricing appear to be the most important differentiator many DSPs claim versus networks.
- DSPs must be able to provide global frequency capping across all inventory sources. This is definitely of value, but unfortunately the concept is currently mostly an aspiration, unless the DSP also has control of or visibility into all (including non-RTB) media buys, which is unlikely in the near term for most large clients.
- DSPs should provide unified optimization, analytics, reporting, and conversion attribution. All but the last part are important to the idea of a DSP, but universal attribution is a more difficult issue that also will need collaboration amongst display, search marketers, and others to help solve. Alternately, the DSP should be required to control all media buying for an entity.
- DSPs enable their clients to leverage various forms of data (both theirs and third-party). This is truer than ever, and the dissociation of data from media means the client can get best-of-breed vendors -- not an inferior mix of media and data. The other point here is not only to help the client use data, but to use it in an internally and historically consistent manner (i.e., learn from past mistakes and past audience performance).
- A DSP should calculate the value of each impression in real time individually, relative to the various characteristics of the impression. This is a misconception about real-time bidding. Any system at the scale of tens of thousands of bids or impressions per second will require some level of shortcut to group like individuals together. Even still the granularity will be unparalleled in any other medium (and even moreso compared to other online channels), but impressions are not individually valuable enough, or scarce enough, to be individually valued at runtime.
- DSP equals RTB. This is not at all true. The benefits to streamlined, centralized workflow and audience/data and buy management are huge. And with RTB still accounting for less than 5 percent of all display media buying, clients can still get a lot of value out of streamlining and having more transparency into non-RTB exchange buys, ad network inventory, and direct publisher deals. RTB is becoming a sine qua non for DSPs and will increase in importance. A DSP should own and operate its own core technology around campaign management, real-time bidding, optimization, and core reporting functions, but by no means should it be defined today solely by its real-time bidding functionality, since that part of the market is still nascent.
In summary, the very fragmentation and market diversity that led to the creation of DSPs are the forces that make it hard to pin down precisely what the DSP category is (if anything) and what a DSP should be. Broadly speaking, DSPs appear to be companies using more technology than those in this space ever have, providing more transparency into pricing and inventory than ad networks have done, and hopefully delivering great results for advertisers. Large ad networks with proprietary inventory relationships, data, and optimization are still able to provide a lot of value to clients that are comfortable with the level of transparency they can provide on inventory and pricing (and besides, there are now lots of new vendors providing a peek into some of these aspects, whether the network or publisher likes it or not) and assess if they perform or not. It thus behooves agencies and advertisers to ask potential DSP partners what exactly they do, how transparent they are about the components thereof, how good they are at making it work, and (perhaps) assessing how long it would take to build that functionality yourself versus use, partner, or buy that technology.