If you’ve heard about Machine Learning, you are probably thinking about Artificial Intelligence. You are not wrong. It is a subset that sounds cooler than the old A.I. term.
In a nutshell, Machine Learning means “learn from data and predict an outcome.”
The high-level process is simple:
1) Define an outcome: “I want to get the highest LTV costumers.”
2) Provide a training dataset: a CSV containing the text copy of all the ads that drove the highest LTV costumers.
3) Create a test dataset: a list of potential texts you want to test in your new ads.
4) Run ML: using a classification algorithm, for example.
5) Evaluate results: ML will tell you which ad text copy is most likely to perform better. Once you’ve tried them out, you can use the winners as a training dataset.
The steps I described above belong to a supervised-learning/classification technique. There are more techniques you can test.
Why should you go into the trouble of using ML? There are three main reasons:
1) Media analysts might do a good job evaluating which of 2 ads will perform better (based on what they know so far.) However, when you are evaluating 2,000 ads, the job isn’t fun anymore, and a machine can help.
2) We, as a human, are lazy (“do I really need to test 100 combinations?”) and biased (“let’s try with cats photos, everyone loves cats.”) Machine Learning can give us an angle we haven’t though yet and guide us to create high-performing ads.
3) It is cost-effective to test only those combinations (whether text, images, videos) that are most likely to work. Otherwise, we might be spending money and time on ads that could be good, but not great (i.e., we are not learning.)