The propensity to buy prediction allows you to evaluate how likely a customer is to purchase products with specific characteristics, such as brand, category, color, and more. You must admit that this is powerful knowledge, as an effectively applied propensity to buy calculation can give you key insights into how to distribute and diversify marketing communications, avoiding one-size-fits-all marketing, in which time and resources are blindly allocated to the entire customer base.
Leveraging predictions allows companies to be much more efficient in their marketing, leading, on the one hand, to increased sales by reaching customers with a high probability of buying a given product and, on the other hand, to lower costs by avoiding sending campaigns to customers with extremely low propensity to make the desired conversion.
If you collect data and have a product catalog, you can start your journey with the propensity model. From a general point of view, using a propensity score involves 4 steps:
There are many use cases where propensity predictions can be applied, to name a few:
However, to take full advantage of such use cases, you must use propensity in the most effective way. To do this, you need to choose precisely the audience you want to reach, define the content you will provide, and consider the timing of propensity calculations.
Let's take a closer look at the 5 elements that will lead you to success.
A segment is a keyword here - in most cases, there is no need to calculate propensity for the entire customer base. The audience choice should be related to the channel in which you want to reach customers and the use case you want to prepare. If you want to send SMS, consider customers who have agreed to receive SMS communications. If you want to promote a campaign on a mobile app choose users who have a mobile app. Find users who have interacted with your brand in the last 30 days to reach only active customers. On the other hand, don't narrow your segment too much by adding too many conditions. In case it is impossible to calculate a particular user's propensity, for example, due to lack of interaction, the user will not be included in the calculation result.
The result of a propensity calculation is an event that indicates the propensity of how likely someone is to make an anticipated purchase. Since propensity to buy is used to find a group of users who share a common interest in making a transaction, it is worth using percentiles, which provide the most efficient way to find such a group. If you want to find the 20% of customers with the highest probability of making a transaction, choose users above the 80th percentile. On the other hand, if you want to exclude the 20% of customers with the lowest probability of purchase, select everyone above the 20th percentile. Percentiles allow you to control the size of the group you will reach with your campaign. Let's say you want to make a prediction for a segment that consists of 100,000 customers. Whether you want to create communications for 10k, 15k, or 20k customers - you can use percentiles to find them.
Calculating the prediction a few days before sending the campaign misses the point of using its results since propensity is more accurate the closer to the start of the campaign it is calculated. If you send a newsletter campaign every Thursday at 6 pm, train the model on the same day. If you have a campaign scheduled weekly, you can recalculate the prediction every 7 days and have fresh results when sending the message.
However, if you want to calculate propensity precisely on time, it would be even better to rethink your campaign management. Maybe you should reverse your approach and not fit the prediction into the communication calendar but make the campaign launch dependent on when the prediction is calculated. For example, you can schedule a propensity model recalculation every 7 days. The result of such propensity calculation will be the corresponding event that appears on the customer's profile and can be used in automation. In this way, you can run the automation for each user who got a propensity event and e.g., is in the top 20 percentile of users willing to make a transaction and then send a campaign.
To have the best proof of how the propensity to buy can help you take your work to a new level, you should prepare an AB test to check your hypothesis. Suppose you want to reduce the cost of an SMS campaign. Make one campaign to customers with a high probability of making a transaction and another to customers with a low probability. In that case, you'll test whether sending such a campaign to individuals with a low propensity makes sense. If you have customers with a high probability, send half of them a coupon code and the other half no code, and see what difference the coupon makes. Remember that these two cases test different things - the first verifies the model result, and the second verifies the content. Depending on the implemented use case, you always need to define KPIs to verify the results of each AB test variant.
The propensity to buy any product, a specific category, or brand - all of these can be easily enhanced with recommendations. If you have users with a high probability of making a transaction - send them personalized recommendations that are tailored to their preferences based on their history. If you find users interested in a particular brand - send them personalized recommendations of that particular brand. If you want to promote a specific product and have found customers most likely to buy it, expand the communication with cross-sell recommendations that may interest them. This is an ideal combination as you won't have to guess which users to reach but also which products they should get - AI will do the work for you.
Predictions now form an integral part of modern marketing operations. They eliminate guesswork and help marketers make decisions based on real insight derived from comprehensive data analysis. Synerise's prediction model is a codeless tool that will streamline your work and performance. You can make accurate predictions of future customer actions and behavior, leading to more informed business decisions and implementing strategies that are most likely to succeed.