The Myths, Magic and Mastery of Programmatic Buying with DSPs
Khara Hutchinson is Mediaplex’s Media Campaign Operations Supervisor and an advisor to the MOJO Demand Side Platform (DSP) product team. Behind the MOJO spoke with Khara about real-time bid strategies on exchanges, misconceptions, and the client launch of MOJO DSP and Mediaplex Media Services.
At a high level, what is a demand side platform (DSP)?
A DSP, at its heart, is a kind of ad server focused on engaging with real-time inventory sources. It has a predictive engine and the ability to automate bidding based on its algorithm’s ability to understand data and business criteria. The reason for using a DSP is to manage the sheer volume of both opportunities to bid and data required to execute a bid. These both require a high level of automation.
DSP should also have in place mechanisms for brand safety, targeting by geo, white and black lists, and segmentation. They should also support incorporating first-party and third-party data to drive the algorithmic engine.
In short, the most important aspect of a DSP is to use data to accurately predict the value of a user. And specifically, how do you value impressions and users on non-guaranteed inventory.
How do DSPs facilitate real-time bidding (RTB) on ad exchanges?
Ad exchanges have independent user interfaces you can go through. But if you’re trying to manage multiple exchange buys, a DSP not only enables efficiencies across multiple inventory sources but is the brains behind the operations. DSPs provide a way to programmatically buy in large numbers across multiple properties and enables efficiencies by aggregating workflow.
One benefit to a DSP is it supports a dynamic CPM on the exchange. This means that the winning bid pays the second-highest bid amount. The advantage to this is your winning price reflects the marketplace versus a set price by a publisher.
Isn’t exchange inventory just remnant publisher spots?
Remnant really just means unsold inventory, which can average 30% or more of a publisher’s total stock. I can’t think of a single premium publisher that does not send inventory to one or more exchanges.
Many publishers require inventory to be masked to avoid cannibalizing their guaranteed buys. But those high-CPM publishers do have media available on the exchanges for a fraction of the price. Again, this doesn’t mean that exchanges replace guaranteed buys at all. Inventory from portals and premium publishers fluctuates according to their guaranteed sell through and is most difficult to find when most advertisers need it most – in Q4.
Exchanges are about reach, efficiency, and the ability to value each individual impression based on whatever criteria your DSP is using for this.
It’s not the Wild, Wild West. There is good inventory. Your DSP should be able to capture the mid and long tail. But it’s also not about cookie bombing. Your execution must be data-driven and your DSP of choice must embrace that. No brand’s campaign is going to act like anyone else’s so the decisions you make must be backed up by data.
The DSP should also provide site transparency. If you’re relying only on publisher self-declared information about a site, it can come back to haunt you. Your DSP should provide safety ratings and utilize ad verification technologies.
How do you get started?
It all begins with a data strategy. If at all possible, look at data, trends and audiences that convert. Identify segments that work. Start to develop the data model to identify look-a-likes. For example, if you know users are unlikely to convert after three days, block those users from seeing any more impressions. That data may not be available in campaign, but might be known anecdotally or through offline analysis. In short, everything you know about existing users and buying cycles is fuel to the DSP to make the right decisions.
What’s Mediaplex’s Role in RTB?
Mediaplex is blazing some new trails in the world of programmatic buying. We are building out our DSP offering, which is essentially an addition to MOJO Adserver, utilizing many of the same backend optimization on creative and conversion tracking technologies. In short, you can run both traditional buying and programmatic buying on a single platform UI. We are also integrating with Adsafe to determine when to bid or not to bid.
From a planning and targeting perspective, we developed a comprehensive predictive modeling engine to value each impression based on individual attributes versus audience segmentation or site content. We’ll support lookalike modeling as well as contextual targeting based on site content. All the data modeling will be driven from information stored in the MOJO DMP, which is integrated into the platform.
Additionally, we’ll offer the use of dynamic creative to aid in creative optimization for retargeting purposes.
We are also adding Mediaplex Media Services, a full-service offering where brands let us curate their RTB campaigns. We work with clients to assess their goals, develop a data strategy, scope out pricing and audience requirements and execute the campaign. With analytics and attribution included, we will provide all the reporting as well.
It takes about a week to get a campaign running and you need just enough investment for the DSP optimizers to kick in.
Can a brand drive their own campaigns?
Like any process, programmatic buying can be learned. However, maximizing its use doesn’t happen overnight. I’ve been doing this for over five years and I’m still learning. The tools have greater capability, the data is richer and DSPs are more user-friendly. Three years ago, DSPs were focused on site optimization. Today there is capacity for more data points and more precise algorithms.
In terms of staffing, you need to have people who understand advertising and analytics, understand auctions and the DSP’s technology, and be able to spot trends.
What are the big mind-shifts that brands need to be ready for?
You need to have a clear idea or expectation of what will work. Be honest about your goals so that the algorithm can serve that goal. In traditional media buying, if might have a $10 ROI goal, but you tell your agency $20 to try and artificially maximize it. An algorithm doesn’t understand that kind of subjective finagling. It does what you tell it.
Another point is that you can’t expect performance on par with a direct buy. The overall volume may be lower and the effective CPC might be different. Just as if you would never compare search with display, you would not compare programmatic buying to results from traditional display.
And finally, we are dealing with algorithms. That means in order for the DSP to learn how to optimize itself, it must achieve a point of statistical relevancy. It has to have a couple hundred clicks or actions to confidently start predicting impression value.
Are there particular vertical industries or business models provide specific challenges in the RTB world?
In the beginning, industries with long sales cycles struggled with exchange buying. But over the last few years, I’ve seen a shift in expectations that have revealed exchanges to be an integral part of their media plans. Moving from being “conversion-only” to “brand-introducing” has solidified DSP line items on media plans for many of these clients.
Advertisers with small audience targets (for example, reaching foreign language speakers in a particular city or big ticket luxury items that expect conversions) struggle on exchanges. Small audiences are difficult to scale in terms of volume, and can create havoc on algorithmic bid engines.
Big ticket items, even those purchased online, require a lot of consideration by consumers. You’ll find that these online conversions, search generally wins the last touch game. You’re not likely to get many high consideration purchases from a single click on a banner. Clients like these who don’t value touch points prior to last touch generally don’t have a good experience with exchange buying.
Pharmaceutical clients have data collection and usage regulatory requirements that make it almost impossible to retarget. However, general audience targeting (women, age groups, mothers, etc.) is allowed and if expectations from an exchange buy are set correctly, it can still be valuable to pharma advertisers. Another caveat to mention is creative restrictions. Pharma clients are required to list all the legal jargon, such as side effects, in their creative. This usually requires a rich media format, which presents its own issues on the exchanges.
What types of creative are supported in the exchanges?
You can run all formats of creative on exchanges. One factor is that in most cases, the media must “stay within the bounds” so overlays, expandables and other interrupting creative still has to go through traditional media buying.
Over the last few years, there has been an emergence of format-specific exchanges. That was quickly followed by acquisitions such as ValueClick purchasing Greystripe and Google snatching up Admob in the mobile arena. The same goes for video. The acquisitions were followed by integrations within existing platforms, at least on the front end, so that mobile and video inventory was made available. The amount of mobile and video inventory available currently is drastically smaller than traditional desktop banner inventory. Video comes at a premium on the exchanges (think $8-12 CPM) and it can be difficult to track performance. Mobile is pretty inexpensive currently but in its infancy in regards to tracking performance. Both mobile and video are expected to grow exponentially and the exchanges will benefit from this.
How can a planner feel confident that expectations will be met using programmatic buying?
Programmatic buying should be done in conjunction with other marketing including search and direct display buys. Like display, exchange buys can move users into the conversion funnel and drive them towards an eventual purchase. The key is performing the right amount of due diligence. Brands have to ask themselves: “What is my goal for this buy? What do I expect from the money spent on the exchanges?” This means going beyond measuring clicks and conversions.
An exchange can add a ton of impressions to a campaign making it very effective for branding, but can RTB affect direct response campaigns?
Better DSPs can differentiate between a user’s worth in terms of your goal and bid accordingly and use a multitude of factors to determine the value for each impression. That very aspect makes the exchanges an integral part of a DR campaign to get a valued impression. When I first started in this business, the goal was improving click through rate. Today it can be any performance metric such as a purchase of some kind at a set effective CPA value. Not all campaigns are created equal and success varies depending on advertiser, rate, seasonality, etc.
Why do some suggest that when buying programmatically, a planner should not utilize audience targeting?
Adding targeting to your exchange buys limits reach, interfering with the need to gather enough data to start making sound, confident predictions in bidding. Also, every user should be bid on and the value determined by the DSP based on the factors its bidding engine takes into consideration. This is a good way to test a DSP’s capabilities – how smart is “the Brain” behind the platform.
I advocate running a line item that has no audience (outside geo) targeting alongside line items that leverage learning about which audiences perform best for the campaign. Remember: Performance is not always a click or a conversion. It could be getting new users to engage with the brand via click, adding new users to the remarketing pool with no click, etc.
How can reach and frequency be measured when there is no targeting set in place?
Reach and frequency are determined by cookies, not targeting.
For clients programmatically buying for the first time, who have no data, what should their expectations be and what is their ramp up time?
Even if you don’t have set data in the exchange, you likely have data from other types of campaigns. This is part of fueling the data model and strategy to hit your targets in the general area. Once that initial campaign is running, you need to run enough impressions and collect enough data points on clicks and conversions to let the algorithms of the DSP kick in. We often know in about a week as to what works and what doesn’t.
If you need a site list to get pre-approval, DSPs are out the door as an option, aren’t they?
That’s a common misconception. DSPs are 100% still an option as they can target a site list, even if that DSP has never seen that domain before. Too small of a site list will net very little delivery and you’ll miss out on the premium inventory that is masked. A better option is to discuss concerns, types of sites that are not to be served on, and pick vendors that have strong brand safety controls in place.
If targeting is possible, what types of layering can be done?
This depends on the DSP but inclusion and exclusion of geo, sites, browsers, device types and content categories are almost always available. Attributes like behavioral segments (soccer moms), demo (female), age (25-34) are available but more often at a premium via third party data providers.
True or false: the more restrictions you put on the DSP buy, the fewer actual options you’ll end up with and therefore you’ll have a reduced ability to actualize the spend?
Absolutely! Less is more at the launch of campaigns. Trust your platform and your campaign managers (in house and with your vendors) to optimize your campaigns for best possible results.
How do you do flighting? How do you ensure placements off-hours?
Flighting is normally done to “stretch out” your ad spend over time. In the exchange buying is done in near real time and therefore it doesn’t have the same concepts of your financial or time requirements. In the amount of time that a web page loads for you, that impression has been put up for auction to the highest bidder. Each bidder has evaluated what’s known about you and raised a hand with the amount it thinks you’re worth.
Flighting and day parting is handled by the DSP deciding what to bid for a single impression, how much it can spend overall in a given time period, and when to pass on a bid opportunity.
Mediaplex, San Francisco
Mediaplex, San Francisco
Mediaplex, San Francisco
Mediaplex, San Francisco
Mediaplex, San Francisco
- April 2013 (1)
- September 2012 (1)
- July 2012 (2)
- June 2012 (1)
- May 2012 (9)
- April 2012 (15)
- March 2012 (15)
- February 2012 (10)
- January 2012 (12)
- December 2011 (14)
- November 2011 (13)
- October 2011 (9)
- September 2011 (11)
- August 2011 (10)
- July 2011 (7)
- June 2011 (7)
- May 2011 (9)
- April 2011 (9)
- March 2011 (7)
- February 2011 (6)
- January 2011 (5)
- December 2010 (5)
- November 2010 (5)
- October 2010 (7)
- September 2010 (8)
- August 2010 (5)
- July 2010 (6)
- June 2010 (10)
- May 2010 (6)
- April 2010 (5)
- March 2010 (6)