Who doesn’t love the smell of fresh new IP addresses? Well, anyone charged with warming IP addresses for...
We have refactored our engagement scores, and to say “We” I mean “Robots” have refactored our engagement scores.
Early Goals
When we first started the company with a commitment to providing the very best data possible. As the tools have continued to improve allowing us to get closer to the goal. The most recent iteration of that goal introduces leverage modern machine learning technology.
Data Pool
We have been accruing data for the last ten years assembling the largest complaint data pool int eh world. Our approach has been to identify key performance indicators and build algorithms keying off of those indicators. This battle-tested approach has yielded phenomenal results for customers ranging from fourtune 50 brands to small startups. And yet, we continue to ask the question around efficacy and performance.
Machine Learning
Our data pool contains five raw metrics (send, open, click, bounce, and unsubscribe). These metrics provide both data for a machine learning training model and also data method by which we can test the efficacy of a given model.
The Approach
Machine Learning is all about data. You tell the machine learning model what your goal is and feed it data. Our goal was to increase the predictability of a click. The model then tests the data for successful and unsuccessful outcomes. We took the existing successful algorithm that has served us very well and called that the baseline. An algorithm with a positive result meant that it preformed better than the current algorithm, and a negative score meant that it preformed worse. The combination of the dataset and algorithm could now take recency and frequency into consideration, identifying key micro-patterns that are then leverageable by marketers for better performance.
The Results
This is where the magic happens, we were able to see a 20%+ improvement over the existing model. 20%
The Downside
It used to be really easy to explain the difference between a 4-star and 5-star email addresses. A 5-star email address had engaged with multiple brands over the last 14 days. Easy to explain. Now, a 5-star email address has a 1-star higher likelihood of engaging than a 4-star. What does that mean? It means that 5-stars are better than 4-stars. It means that you will get more activity out of a 5-star than you will out of a 4-star. It means that the more that you email a 5-star email address, the greater the chance that you will pick up more engagement over time. Conversely the lower you are on the star scale, the less likely you will get activity from a given email address.
The Upside
Your email campaigns will perform better.
More emails reactivated – The machine learning-based algorithm focuses on the predictability of engagement which translates to even greater success.
Faster IP warming – You will have a more accurate warming schedule that takes less time to complete while warming IP addresses.
The Respondibility Curve – This is a term that we made up. It means that these emails may respond, they have shown a history of responding for other brands. The curve being that the more that you email these address, the more likely they are to respond because that is what they have done historically for other brands.
Conclusion
We believe that the advent of these scores ushers in a new era of performance for our customers and we are excited to enable the next generation of performance. Customers will be given a choice and the old algorithm will be available so that they can continue to benefit from the scores we have been providing marketers for the last 15 years. The new scores will also be available and their increased efficacy is going to pay for itself with marketers immediately.
