Commentary

Secrets Of A Data Scientist: Five Strategies For Optimizing RTB Campaigns

by Chris Sukornyk , February 5, 2013

Campaign optimization is crucial for any RTB-driven campaign. However, with so many variables to consider, what are the most important factors that determine a successful campaign? To get to the bottom of the question, I interviewed Clement Chung, who holds a Ph.D. in machine learning and is the lead data scientist at Chango (full disclosure: I'm CEO of Chango). Below I outline his top-five optimization tips for RTB-based campaigns.

1) Say it three times: "The algorithm won't save me." Ironically, a Ph.D.-holder in machine learning candidly pointed out that too much emphasis has been placed on algorithms. Algorithms are obviously important, but they have their limitations. To be effective, algorithms need sufficient data -- and since each campaign can perform differently, it takes time for them to be effective. It's not uncommon for a campaign to have spent 30%-40% of its budget before algorithm-based optimization has a huge impact. Therefore, an initial setup of a campaign has a large impact on results -- and the setup needs to be influenced by human intuition. For example, choosing the right segments, or the right site-list, has a tremendous impact on the campaign.

2) Focus on granular measurement. Granularity is one of the golden rules of RTB media buying. You cannot optimize what you cannot see. The more granular a campaign can be the more performance a campaign manager and the algorithm will achieve. Try to avoid a situation where large numbers of people belong to one segment. Different platforms have different levels of granularity, and you must ensure that you can optimize at very discrete or micro-segment levels.

3) Frequency caps & data freshness - Frequency capping is a common tool that still holds a lot of value. While the algorithm should auto-tune your frequency, ensure that your starting value is appropriate for the medium. For example, frequency caps on FBX should be almost double any other exchange. The "freshness" of your data is also extremely important, particularly if you are targeting search events. Ensure that you can optimize by limiting the freshness of the data. For example, the data may tell you that retargeting anyone who has visited a site 30 days ago is a waste, since those indivduals are never going to purchase, so why waste media dollars on them?

4) Isolate your tactics - It's OK to experiment, but you need to isolate your tactics. Ensure that your daily budget is not being consumed by some experimental segments or keywords. This works differently in different systems, but the point is that no single segment or tactic spends an overwhelming portion of a campaign budget.

5) Focus on creative. The so-called "last mile" of digital campaigns is easily overlooked. Creative needs a clear call to action. Creative used for retargeting campaigns is usually very different from creative that prospects for new customers. Pure retargeting campaigns often can coast with less focus on creative because of brand association. On the other hand, prospecting creative needs to clearly explain the problem being solved and cannot rely on brand association.

6 (bonus secret) Know your goals. Ensuring that you start a campaign knowing what actual success looks like is crucial. Many clients believe they can optimize campaigns for CTR, CPA and ROAS simultaneously. What view through attribution are you going to use? Have you considered the impact of other digital marketing campaigns running simultaneously?

In many respects, the rules of real-time marketing are still being written. Nonetheless, gains will be made by heeding the advice of a data scientist deeply involved in the everyday optimization of hundreds of campaigns.

rtb

Next story loading