Optimization has always been a core component of online marketing. The ability to see what’s happening and adjust pieces of a plan to make it work more efficiently and effectively is quite thrilling. That promise is, in many cases, what drew a lot of us to this industry.
Until recently, when we spoke of optimizing a campaign, we were referring to direct response metrics: response rate, conversion rate, cost per acquisition, and so on. Now we can broaden that discussion to include brand metrics, including unaided awareness, purchase intent, brand favorability, and message association. I base this on some work we’ve done at our agency and a new Advertising.com offering.
I’ve discussed how our shop has been able to effectively replicate existing publisher-side segment-based targeting tools such as Yahoo’s Fusion Marketing and New York Times Digital’s wide-angle targeting. We achieved this in large part by applying Poindexter Systems’s Progressive Optimization Engine (POE) as an automated optimization tool.
What’s unique is rather than optimizing based on direct response metrics, we optimized based on the likelihood a user belonged to one of our target segments. We did this by tapping a database of user profiles during the ad-serving process. If a user displayed anonymous data points indicating she’s likely to be in our target segment, we served an ad. If not, we passed the impression back to the publisher.
What we’ve accomplished is significant. Yet not all advertisers use segment-based targeting. We wanted to leverage previous lessons and apply them to more straightforward, brand-oriented campaigns. The solution was fairly simple. Rather than tapping a data profile, why not tap an ongoing research initiative, such as a Dynamic Logic AdIndex survey?
The basic premise is to field qualitative research concurrently with the ad campaign. From the results, we develop profiles of users who indicate a higher likelihood of being affected.
For example, you create profiles based on users who report larger increases in the specified brand metrics. Profiles consist of anonymous data such as time of day, day of week, ISP, site they were reached on, bandwidth, geography, PRIZM code, language, and other criteria. Once that’s established, all that’s required is a smart serving solution capable of identifying a potential user’s anonymous characteristics, comparing them to the “high-value segment” your research provided and optimizing based on how well the two profiles match up. Theoretically, this should result in your ads being served to people who are more likely to be affected by them.
As with most things, there are some limitations to this approach. Anonymous data is not always available for all users. Geography, which can be a very important factor, is not accessible for AOL users. There’s about an 85 percent accuracy rate identifying those users whose location can be identified. Another limitation is the sample size of the research that feeds the model. Typical AdIndex surveys have cell sizes of about 300-400. Because the approach taps the AdIndex results through the life of the research, sometimes the model optimizes based on samples of 100 or fewer respondents. Obviously, this grows as the campaign matures. But even at 300-400, it’s the biggest limitation.
Not long ago, a team from Advertising.com’s Baltimore headquarters came out to present some new offerings. Much to my surprise, they explained to our team exactly what we’d been developing to optimize based on brand metrics. I couldn’t believe my ears. They basically explained the methodology we had just recently developed and applied in-house.
Advertising.com’s approach, like ours, has pros and cons. The only real con is AdLearn is limited to the Advertising.com network. I’m not suggesting Advertising.com has a bad network — it doesn’t. I simply see it as a limiting factor.
Its major advantage over our model is the number of survey respondents who make up the test cells. It struck a deal with Dynamic Logic, allowing it to recruit respondents for the duration of the campaign (it suggests a three-month minimum for optimal performance). This makes sample sizes many times greater than what an advertiser would see with a standard AdIndex survey. The increased sample size makes optimization that much more accurate and effective.
Advertising.com’s other advantage is its ad-serving technology. It has a tried-and-true optimization system that’s won awards over the years and has always performed for us. This is not to detract from Poindexter — we’re firm believers in its product, too. But Advertising.com’s AdLearn is a very effective optimization solution.
Developments like this one are important. We’ve come a long way recently in gaining acceptance as a branding medium. Melding traditional optimization with qualitative research furthers that quest. This is something you all can take part in. Speak to your clients about it. You’ll look smart. And feel free to drop me a line if you need help getting it rolling.
Mark is off this week. Today’s column ran earlier on ClickZ.
27-year-old Swede Felix Kjellberg, who goes by the name PewDiePie on YouTube, has found himself at the center of a firestorm.
The explosive growth of video in 2016 makes 2017 an important year for video content and as more publishers are tempted to use it, it’s useful to consider the best strategies to maximise its effectiveness.
Apple has announced that with the next update to iOS 10, they will limit the number of times an app owner can pester a user for a rating.
2017 will be a watershed moment for video, as consumption moves from the TV to other devices.