"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 Rachit Srivastava, Data Scientist at PlaceIQ.
Thanksgiving is around the corner, meaning great food, family and, of course, Black Friday.
Mobile ad targeting and strategy have advanced significantly this year, and mobile marketers are reaping the rewards of proven, solid algorithms and enjoying consistent uplift in success metrics. They should be feeling pretty comfortable with being able to target audiences efficiently on this crazy shopping weekend, right? Why would their algorithms fail them now?
Black Friday is like no other time of the year, and it may require marketers to step out of their comfort zones. This is no time for autopilot. If you want to pass the ultimate test, preparation is key.
On Black Friday, consumers are more inclined than ever to shop, and advertisers want to take extra advantage of this. This means that overall there will be more competition for targeted ad impressions served that weekend than on other days of the year. Advertisers will funnel an increasing number of ads towards these limited impressions in order to cash in on the intent of consumers and channel them toward stores.
From a technology perspective, this requires some serious infrastructure, and it can test the capacity of companies’ systems to levels never experienced before – even up to the breaking point. The situation can be likened to how it was for Nasdaq systems the day Facebook went public or for the Obamacare website the first day it opened.
It is essential that the infrastructure can support the volume of and competition over impressions associated with Black Friday. It is also essential that algorithms and strategies are scalable from an efficiency perspective. In other words they must be able to support the volume at the required speeds. Otherwise, you run the risk of completely missing Black Friday due to a technical glitch.
If your infrastructure is covered – great! But don’t relax with a glass of eggnog yet.
Major Challenge 2: Confidence in the Algorithms
Consider the following situation. You find a coin and flip it 10 times. It lands on heads 10 times. Are you convinced that this is not a fair coin? Probably not, because, although 10 heads in a row with a fair coin is a low-probability event, it is nowhere near impossible. Now let’s assume you observed the coin to flip heads 900 out of 1000 times. In this scenario, you are more likely to be convinced that coin is not a fair coin. This example is very similar to how marketers and analysts observe and test targeting algorithms in the mobile ad domain.
Using simple tests, such as A/B testing, companies can try to measure how well they are targeting audiences. One may select two groups of people and target one of those groups using an algorithm, while randomly serving impressions in the other. Both groups must be served the same number of impressions and contain the same media.
If the group with the targeting algorithm has a higher success rate than the randomly served group after one test, it may be a fluke. But if the targeting algorithm has consistently created an uplift in the success metric over the course of several tests throughout the year, one can be pretty comfortable that it works. The subtle yet extremely important factor here is that if an experiment has been shown to work in similar conditions consistently, it is considered successful.
Why would anyone have reason to doubt that these algorithms will work on Black Friday?
The answer is that Black Friday weekend is very different from the rest of the year. Consumers have more intent than ever to purchase products, and they are also more likely to be shopping with friends and family as a social activity compared to other days of the year. In fact, many families make shopping part of their holiday tradition, right along with the turkey. The “conditions” in which the experiment has been shown to work are not the same as those on Black Friday.
So how do you increase your confidence in targeting during the biggest shopping weekend of the year? The savior here is your technical infrastructure. The number of impressions that must be served during the day and the concept behind the coin toss example described above can be used to your advantage.
After serving a larger-than-normal number of impressions during the day, if an uplift in a success metric is observed, one can become confident that the uplift is because of the algorithm or strategy and not some sort of sampling bias.
This large number of impressions also allows one to test out different strategies and adapt them during the day. To do this efficiently, marketers must ensure they have adaptable algorithms as well as a variety of algorithms. This is not the time to rely on a small set of algorithms – proven or unproven – because things can go downhill fast if during the initial part of the day there is no positive uplift in your success metric. This situation can be likened to statistical arbitrage based trading strategies. In this market, adaptability is key, and the ability to detect changing market conditions could make or break your Black Friday.
The Ultimate Test
Black Friday is a challenging yet very interesting weekend for mobile ad targeting. It really can be considered the ultimate test for a platform from a scalability and accuracy perspective. Provided systems are robust and scalable, it is also a great day for companies to make their mobile ad targeting algorithms stronger and more robust. It really is the performance in this “ultimate test weekend” that can differentiate the great ad-tech companies from the rest.
Add some leftover turkey, and Black Friday can become one of your favorite days of the year.