Predicting Ad Performance With The Touch Of A Button
David Milton Wednesday, April 3, 2019
Share this page with a friend
As a marketer, your main focus is making sales. That’s is the ultimate goal, sure. Keeping track of those sales can be a daunting task.
The most asked question asked of me is: “What software do you use to track sales?” Everyone is looking for that one ‘Magic Bullet’ app designed to do everything with the touch of a single button.
Well, I found it. As I wrote in a previous edition of ‘Funnel Magazine,’ Bayesian Statistics is an approach that allows you to predict the future performance of an ad or funnel, with a degree of certainty.
The difference between the ads is expressed as ‘Lift.’ You can have Positive Lift or Negative Lift. When dealing with money or costs, I like to see Negative Lift (lower prices, costs). When looking at impressions or sales, I make decisions based upon Positive Lift.
Given a specific data set, we can predict the outcome of ‘Set A’ outperforming ‘Set B’ with a degree of certainty, expressed as a percentage, and referred to as Lift.
If ‘Set A’:
If ‘Set B’:
After a 4-day test, we crunch these numbers into a Bayesian calculator and discover that ‘Set A’ has a 72.4% chance of collecting more Clicks than ‘Set B,’ over time.
Additionally, ‘Set A’ is given a 60.1% chance of making more Sales than ‘Set B.’ Therefore, ‘Set A’ is more engaging (Clicks) and will result in more sales.
Data ‘Set A’ has 72.4% more Positive Lift (Clicks) and 60.1% more Positive Lift (Sales) as compared to ‘Set B.’ Averaging these metrics yields an overall Positive Lift of 66.25% versus ‘Set B.’ Averaging is optional, but it can help solidify your decision.
When using Lift, there is always a clear-cut winner. Once Lift becomes negative on one ad, the comparison ad shows more positive Lift. The Lift is transferred to and from each ad, making it easy for humans to make confident decisions.
Lift can also be expressed as ‘Trend.’ I’ll explain the concept of Trending in an upcoming article. Easy stuff. Now, back to the naysayers….
Most statisticians would say there isn’t enough data, or the test wasn’t run long enough to collect accurate data. They would be using the Frequentist approach to statistics, and their assertion would be incorrect.
The Frequentist approach reveals what happened, not what is likely to happen. There are only a couple of traditional Madison Avenue ad agencies, and only one (to my knowledge...me) digital marketer using the Bayesian model.
Bayes provides marketers with a huge advantage, their very own crystal ball. You will get your products to market, faster. You will save money on testing and you will have an edge over other marketers, even Tony Robbins’ of the world.
As marketers, we don’t want to know what happened, we want to know what is statistically most likely to happen.
Next...Using Trending to predict the future.
Bayesian calculator: https://abtestguide.com/bayesian/
More Articles By David Milton