3 Tips for Evaluating Data

This is part 2 of a 2 part series in leveraging data for Programmatic Direct.

Trust, but Verify

Don’t assume that all data is created equal in terms of quality and accuracy. Here are some ways to evaluate data quality:

• Ask about how and where the data is collected.
• If it’s probabilistic or modeled, ask about the inputs and the confidence levels.
• Request and review the random record samples.
• Inquire about the freshness of the data: when are the users removed from the “in-market” segments? 30 days vs. 90 days makes a big difference.

Test & Measure

Many buyers lump together data from various providers in order to increase scale. While the campaign setup becomes easy and efficient, learning about performance by segment becomes difficult. This means you can see overall performance, but you might not learn how each segment or data provider performs.

The remedy?

1. Take your time to split your campaign into multiple line items and target each to data from a specific provider.
2. Now you can measure the performance of each data source.
3. Don’t forget to apply statistical methods to verify significance.
4. Mind the attribution. Attribution of conversions to ad impressions and the data used in ad campaigns is a very complex subject. Often multiple campaigns target the same users, and the ad impression that was served last gets the “last touch” attribution credit. This is one of the reasons why retargeting campaigns appear to outperform any other targeting. A better way to measure effectiveness of data would be to designate a control group that is not targeted by the ad campaign.

Don’t Overlook Modeled Data

Modeled data is not bad data – as long as the provider models an outcome instead of an attribute. Most marketers drive a specific outcome: purchase, registration, a political donation, and so forth. A good model will use the desired outcome as a seed and will help you find other users who match the seed users’ characteristics across multiple dimensions. Even branding campaigns should reach users who are similar to your best customers.

Sophisticated modelers combine hundreds of profile signals to create complex custom multi-faceted models, which can successfully predict users’ propensity to convert. It’s simply more sophisticated than a person running an Excel model. Modeled data may get a bad rap, but machine learning can be very effective at determining ideal targets.

*Disclaimer: The above are suggested best practices to be used with discretion. Online behavioral advertising should follow all applicable Digital Advertising Alliance self-regulatory principles, including all opt-out requirements.

As a best practice, ask your data providers whether their data collection practices are compliant with all applicable data privacy laws and regulations.