Today’s column is written by Madhura Sengupta, director of ad product technology at Edmunds.com.
Audience extension – the practice of expanding a digital advertising campaign beyond a publisher’s owned and operated website into the realm of the wider Internet – is a common tactic to drive incremental revenue.
To capture more marketing dollars, publishers buy media on behalf of their advertisers that continues to target users after they have left the site. For example, prospective car buyers on a publisher’s car research site are valuable beyond just their sessions on site. The publisher can continue to target these car buyers and other in-market consumers as they browse The New York Times, YouTube, Facebook and more off-site sources. In this way, publishers pivot into the world of digital agencies.
In order to deliver highly targeted audiences at scale, publishers need to think carefully about how to achieve the most efficient and optimized media buying. While many publishers will pay a premium and opt for managed services from an agency, the most effective way to maximize margins is to cut agencies out and bring the capability in-house through a demand-side platform (DSP) – a programmatic buying platform that uses real-time bidding. Once a DSP is selected, campaign implementations must be operationalized with an ad operations team.
Popular choices for DSPs include DataXu, Rocket Fuel, Google DoubleClick Bid Manager (DBM), The Trade Desk, Facebook, Turn, AppNexus and MediaMath, among others. The selection process can certainly be overwhelming for publishers.
To conduct a competitive analysis, the process should be broken into several evaluation criteria: accurately forecasting offsite inventory; executing buys across multiple devices; optimizing spend based on performance metrics; and using third-party and lookalike modeling in order to increase audience reach.
This process would likely be similar for an advertiser or agency since both are also trying to optimize spend and efficiency. Unfortunately, some DSPs don't offer to publisher clients the same features that are available to agencies (likely to alleviate agencies' fears that DSP vendors would work with publishers in a way that would cut agencies out of the loop).
Powerful Audience Management And Forecasting Analytics
The ability to efficiently purchase media is only useful if a publisher can accurately forecast and sell audience extension packages. A DSP should contain forecasting filters for creative format, size, geography, etc., in order to know how many and which people a publisher can reach.
In addition, a bid-landscape curve can allow publishers to understand the competitive environment for particular audiences. For example, if a publisher changes its max bid from $5 CPM to $7 CPM, it may be able to reach twice the audience.
After forecasting, publishers need to ensure that they can actually buy media across multiple media sources (video, native, display), platforms and devices. In addition, DSPs should enable buying in both the open market and private marketplace in order to maximize exposure.
Note that there are a few platforms that have created walled gardens with custom capabilities. For example, YouTube inventory is only available through Google’s DBM. Facebook’s mobile inventory and cross-device targeting is only available via Facebook Direct. If publishers want access to specialized features, they may need to employ more than one DSP.
Additionally, it is important to research whether the platform provides different levels of access to publishers vs. agencies. For example, DBM restricts certain features from publishers, such as YouTube inventory, keyword contextual targeting and cross-device targeting, and reserves them exclusively for agencies.
Bidding Strategies And Optimization Levers
When executing various types of buys, it is important to evaluate whether the DSP platform is capable of cost per mille (CPM), cost per click (CPC) or both types of bidding.
CPM-based bidding is usually more popular for branding and impression-based campaigns. Alternatively, both Facebook and Google AdWords contain CPC-based bidding options, which is inherently less risky for a publisher that is conducting a direct-response or performance-based campaign. It is important to note that optimizing for clicks may be narrow-minded since “ad clickers” are not necessarily converters, and people who view the ad may convert at a later time. For this reason, it is important for publishers to place conversion pixels on their advertiser’s landing pages and optimize for downstream conversions and KPIs.
Other optimization levers should also be studied. For example, The Trade Desk promotes the concept of “bid factors,” or being able to multiply a base bid by a certain percentage in order to increase win rates. Let’s say the base bid was a $5 CPM. A publisher may want to bid twice the base bid ($10 CPM) for more valuable above-the-fold inventory, which is inherently more “viewable,” while only bidding half the base bid ($2.5 CPM) for below-the-fold inventory.
In addition, many platforms have their own auto-optimization and tuning capabilities, so these are worth testing. Note that auto-optimization often requires a minimum budget and audience size in order to employ machine-learning tactics and be effective.
Access to granular reporting is critical in order to know how to optimize and whether the optimizations are working. Publishers should be able to view performance metrics such as CPC or cost per action (CPA) by various dimensions, including format, creative, fold (placement), time of day, browser, site and day of the week. Reporting and analyzing the data is critical to understanding where a publisher can procure the best-performing inventory in the most cost effective way, in terms of which factors contribute to the lowest CPC or CPA.
For example, a publisher may notice that its CPA is always higher than average on Internet Explorer vs. Chrome – thus, it may be worthwhile to completely remove Chrome from the campaign. It is especially useful if all reporting is available to download via API into a data warehouse. Once there, it can be used for more advanced visualizations, insights and development of bidding strategies.
Lookalike Modeling And Third-Party Data
Lookalike modeling is an important component of further expanding reach beyond simply retargeting users who have been to a publisher’s site already. This tactic is based off of analyzing patterns of a smaller “seed” audience and using a set of common attributes to create a larger “lookalike model,” either from an existing audience or from the wider Internet using third-party data.
For example, a publisher can track which users take certain actions on their sites, such as shoppers who place certain items in their shopping carts. The publisher can build a lookalike model with a larger audience based on this characteristic.
Most DSPs enable publishers to build lookalike models that leverage larger third-party data sets. On The Trade Desk’s DSP, for example, publishers can set up lookalike modeling and pay to layer third-party data – such as in-market auto intenders from DataLogix – on top of their own audience, either as a percentage of media spend or a flat rate.
While this is the norm among most DSPs, there are some platforms like Facebook that do not charge additional fees to build lookalike models since their inventory is their own.
Making A Selection
These are just a few considerations in a publisher’s DSP selection. There are a wide variety of variables involved, and it is useful before making a final decision to test a few platforms for key factors, such as ease of use, transparency and customer service.
Although the DSP selection process can be lengthy, the time investment upfront certainly pays off in the form of more sophisticated sales products, increased efficiency and, best of all, more revenue.