The heart and soul of marketing is targeting strategy, and when it comes to digital advertising, that strategy is increasingly data-driven and algorithm-based. Compared with traditional media, digital advertising has the advantages of targeting/personalization, precise performance metrics, and much greater flexibility in terms of reach and cost. These differences also spark two dramatically different views of advertising strategies like two dueling gunfighters in the Wild West.
In the east corner is the shotgun approach: let’s take advantage of the Web’s abundance of low-cost ad impressions and canvas the Web until we land on enough interested users to cover the cost. This is also known as cookie bombing.
In the west corner is the modern day sharpshooter: if the average response rate is around one basis point, only highly-targeted, well-placed ad messages can get the job done effectively. It’s a targeted approach with a well-researched plan (by strategically analyzing data to effectively drive performance), and no collateral damage.
The collateral damage of flooding the Web with cheap ads includes a negative consumer experience as well as impacting the publisher’s ability to maintain quality content. The past success of digital advertising has demonstrated convincingly that it is not quantity, but quality that matters. For online advertising, data-driven audience targeting is that line between highly relevant ads and spam. In fact, audience data and data management has been one of the fastest growing areas in the last year. For example, if you compare the latest version of the now infamous Terence Kawaja logo slide with an early version, you can easily see, by the number of logos, that “Data Suppliers” and “DMPs and Data Aggregators” have more than doubled.
Predictive audience models (a.k.a. algorithms) are like heat-seeking missiles. The financial industry has long used predictive models to predict the behavior of the top 1 percent highly valuable customers or seek out the bottom 1 percent most risky behaviors that could lead to big losses. It is typically done by carefully feeding all the user profiles and user transactional data into a set of highly complex, nonlinear machine learning models. In digital advertising, the goal is to predict which users are most likely to respond to which ads. Therefore, the “heat” in the heat-seeking missiles is the particular pattern in user profiles and user transactions that are most indicative of a positive response to the ads. The model building process is a mathematically sophisticated deduction process based on large quantities of data.
Learning from all the historical patterns and trends, predictive models can then evaluate every combination of user behaviors, ads, content, and context. It then enables advertisers to get to the right user, at the right time, in the right place, and with the right message. This formula is as good as gold – in 2009, Netflix paid a million dollars for its movie recommendation model and many considered the price dirt cheap. What did the company pay a million dollars for? A better model. The industry has come a long way from mediocre technology that gets the job done to building complex optimization algorithms that operate in milliseconds, turning data into predictive targeting and bid pricing. This level of complexity creates a more sophisticated approach to media buying. With today’s computing power, brands can leverage the full scale of digital media data, statistically analyzing different variables such as time, location, page contents, recency, frequency, and velocity of user interactions. This impacts both performance and the understanding of a consumer’s decision process.
In RTB, bids/costs become random variables that dynamically reflect the demand and supply of trading. Thus, an algorithm should do three things in real-time: intelligently explore inventories for each campaign along behavioral, contextual, and social dimensions; predict the precise values of each ad impression for each campaign; bid aggressively based on the dynamics across each auction market/website/inventory type, etc. As a result, advertisers will see performance lift across more targeted campaigns.
For example, leveraging the power of RTB optimization, a top telecom provider successfully launched and rebranded its high-speed ISP product line. Its RTB portion out-performed the non-RTB audience targeting by 44 percent, and out-performed run-of-network by 562 percent. The data behind this showed two things: audience targeting out-performs no audience targeting by being much more relevant, and RTB outperforms non-RTB by broadening the reach and reducing the cost of media.
Audience targeting matters even more for top-tier advertisers who also value brand image and longer-range influence to users. For top agencies as well as advertisers, focusing on audience targeting is an incredibly powerful way to differentiate from peers. While in today’s online advertising there are still loopholes that allow the cookie bombing approach to work on small-scale tests, you can only rely on true technology and deep science to build full-scale, sustainable success. As the saying goes, it’s the algorithm, stupid.