Personalisation: rule-based vs machine learning
It makes sense to address your customers personally on your website, in e-mail campaigns, and advertising, nobody doubts that. In practice, two personalisation methods are used: rule-based and machine learning. When do you use which method?
The usefulness of personalisation can be seen and demonstrated in many different ways. Segment, for example, researched on its effect on the customer. It showed customers who have a personalised shopping experience are more likely to purchase. They will also tell about their experience on social media, and to family and friends. Econsultancy, on the other hand, researched which channels you should personalise to realize the most conversion. Although most marketers are engaged in personalising e-mail, it appears that personalization in search engine marketing has the most effect. Another study by Hubspot shows that personalizing the call-to-action, regardless of the channel, converts 202 percent better than standard versions.
We can continue with many examples like this for a while. And no marketer will contradict that personalisation is meaningful, but how do you get started? There are two options: rule-based and machine-learning personalisation.
Most marketing automation applications are rule-based, it is determined by if-then scenarios. For example: when a website visitor is close to Utrecht, then a nearby event will be shown.
If you organize content and campaigns based on rules only, you have to think of all scenarios and make many assumptions. Which is a lot of work. A simple shop or travel organization has more than two thousand variables that play a role in personalisation. If you have to come up with all the correlations, you will be busy for a while. Therefore, rule-based personalisation is not suitable for product- and content recommendations, nor refined segmentation.
The main advantage of rule-based personalization is that you can tailor your message to certain segments, if applicable. Additionally, you have insights and influence on the recommendations. The disadvantage is that the rules have to be set up manually and kept up-to-date. Next to that you have to make choices about which segments you are going to personalise; which is not practical at all. Moreover, often when setting up rule-based personalisation, assumptions are made about correlations which turn out to be different in practice.
In machine-learning personalisation you ‘feed’ algorithms with data, which they use to search for patterns. What message the customer or prospect will see is based on this. This goes far beyond rule-based personalisation. The algorithm has more refined segments than a marketing team can come up with; even down to the level of one-to-one communication. For example, that women from fifty to sixty who live in Groningen, have a relatively low income and visit the site on Monday, can be persuaded with scarcity. The more specific you are, the better the results.
The biggest advantage of personalisation with machine learning is that you can serve every visitor one-on-one with virtual content. Additionally, you can quickly personalize the website / app / e-mail campaigns and it is automatically done. Compared to rule-based personalisation, only little manual work is needed. But, to get the most out of it, you have to set up a good strategy. Choose a powerful platform, and take time to train algorithms.
Combination is the strongest
Marketing automation is generally still rule-based today, but actually machine learning should become leading. Most thinking is done by the algorithm, but you can overrule it if that’s logical. Overruling does not make sense when you train an algorithm for predictable things such as delivery time or someone’s birthday. Then you can better set the a rule for sending a congratulatory mail on that specific day.
Lets take automotive as an example, you can set as a rule that summer tires can’t be offered in the next four months. This can be done with machine learning as well, but it takes a long time before the correlation between the time of the year and the tire type is made. Then it is better to train the algorithm to look for less predictable patterns. For instance, in which weather forecast that type of customer changes their tires, and if the income of that customer or the fact that he is a business driver influences this.
For example, within fashion, it could be that someone who buys a white t-shirt twice, goes to black in the third transaction. Machine learning recognizes this correlation automatically and also knows exactly what the characteristics of people are who buy two times white and then black. When you had to think of that scenario with your marketing team, it would take way too much time. Moreover, you should regularly check whether black is still the fashion color, so your content continues to fit with market dynamics. Time is precious, so why not let technology do the work for you?