“Data-Driven Thinking" is written by members of the media community and contains fresh ideas on the digital revolution in media.
Today’s column is written by Erik Weiss, chief revenue officer at Datacratic.
One of the biggest benefits of digital advertising is the ability to use performance metrics. Attribution may be disputable but some campaign conversion types are not, which has led to a boon in performance-based businesses.
Traditional brand advertisers have also entered the fray, requiring measurable digital objectives. Consequently, competition to demonstrate lift is intense. It’s not uncommon for advertisers to put multiple networks, publishers and demand-side platforms on a campaign to see which will best perform and receive the lion’s share of the budget.
But past returns do not guarantee future performance. The winners need to repeatedly prove themselves for future dollars. To stay competitive, media buyers deploy a myriad of optimization techniques, but I believe there is no one size that fits all. Each step requires its own unique blend of tools and know-how, a careful balance that is a combination of art and science.
Companies may manually divide campaign line items into hundreds of sublines to target each possible tactic for lift, and then rearrange spending accordingly. Others hire teams of data scientists to apply the latest machine learning capabilities. Still others seek out and apply unique data sources to their campaigns. Survival requires doing whatever necessary to eke out an advantage.
Vendors may offer full platforms and services where optimization is either managed or built in. Some are known for proprietary approaches and technology that seem to perform competitively. Yet performance can be fickle. Optimization models, inventory, time of day, geo, behavior, first-party data, third-party data and other targeted attributes may show positive gains and then suddenly cease to maintain their position. The best data scientists may struggle with mediocre data and limited inventory. And then the data and inventory that works the best can change quickly.
To perform competitively, marketers need to make continuous investments in inventory sources, data, campaign probing, technology and a well-trained campaign team. Even with these approaches, finding the right trade-off between learning and performance is difficult to automate and there are various components to test and optimize. For example, targeting the right audience through advanced modelling requires integration with a data-management platform or internal data store.
When managing one's own technology stack, the costs of the infrastructure need to be efficient enough to justify the investment. Server costs, for instance, cannot take up a meaningful percentage of the media budget.
Even with the most advanced technology that can remove many manual tasks associated with optimization, care from an experienced campaign manager, trained in data science-based methods, is needed. Conversions used to build models must be far-reaching and current to find the users and impressions with the highest probability of conversion. Marketers need robust data to properly predict the likely economic value of an impression. Continuous probing for new types of converters across inventory types and behavioral signals keep the optimization fresh.
There is no silver bullet, be it format, data, inventory source, algorithm or manual approach, to achieve sustainable superior results. There is a blend of science and know-how, reach, setup, maintenance, testing, tools, analysis and re-examination.
There are various models and methods for each step of the way to optimize across audiences, creative types and formats, second-price auctions, guaranteed inventory and closed marketplaces. There needs to be an organizational commitment across business and technical departments.
Optimization is a journey, and each step requires creativity, discipline and leadership to be successful.