Episode 1: The Problem with Rule-based Segmentation
If you’re already using segmentation, you’ve probably had great success with it. And for good reason – compared to writing an email to everyone on your mailing list, writing only to users who show engagement on the site makes a huge difference. It improves your conversion rates dramatically, by 100% or more.
But a rule-based system is still a very simple instrument. By defining personas and building the rules around them, we pretend these user types are more than what they actually are: guesses. No flesh-and-blood person fits our persona – an individual’s behaviour is vastly more nuanced than a persona makes it seem. This means you’re missing targeting opportunities.
Machine learning is the only way to encode enough of the potential user behaviours to start treating each user as an individual, based on his unique set of actions.
The problem with rules
Rules have hard boundaries. Either a rule applies to you or it doesn’t, and there’s no grey zone in between. Machine learning gives you soft boundaries so you can work with probabilities – when it comes to selecting individual users for a marketing campaign, you can do so with unmatched precision. Effectively, you’ll be able to pick and choose a conversion rate and select only those users who fit it.
The subjectivity problem
Rules developed by people tend to be biased because they’re based on biases and opinions, even if they’re expert opinions. Machine learning eliminates subjectivity by looking at every piece of data objectively. It finds the patterns a subjective human analysis would miss.
The problem with evolution
Rule-based segmentation is easy to explain and understand. Evolving from rule-based segmentation to machine learning can be difficult because it isn’t easy to visualise the process. Some people call machine learning a black box because it’s hard to communicate what’s going on inside. There are just too many detailed patterns being stored. But that’s also precisely the advantage of machine learning.
What if I already have rule-based segmentation?
Machine learning is segmentation plus. Instead of segmenting based on an artificial user type or simple rules, you base it on the data directly. Instead of having a few rules, you have thousands of rules. This powerful combination will give you another huge boost in conversion, like the one you experienced when you went from no targeting to rule-based segmentation.
If you’re interested in getting that boost back, we can help. Contact us for details.