Machine Learning is Reducing Churn and Preventing Fraud in Recurring Billing
Have you ever loaded a picture onto your Facebook page to realize that people in that picture were automatically recognized? That is thanks to DeepFace, a facial recognition system that touts an accuracy of 97.25 percent.
DeepFace is one example of machine learning, a subset of artificial intelligence in which computers are ‘trained’ to compile data to predict specific results. It is widely used by other high-profile companies such as Google, Uber and Netflix.
In the world of recurring payments, machine learning is mainly used to predict customer behavior, reduce churn and fight fraud.
At one time, artificial intelligence and machine learning something firmly entrenched in science fiction movies—and often, to be feared. However, according to Pega, a software developer for digital process automation and customer relationship management, as many as 84 percent of the population has already used an AI device, while 34 percent of respondents were aware that they had interaction with the technology.
We encounter machine learning every day. Turn on Pandora and play your favorite station. A new song comes on that catches your attention, enough so that you give the song a ‘thumbs up’ on Pandora.
Instantly, that piece of data—this song that you like—is saved and fed into Pandora’s Music Genome, which consists of 400 different attributes, not just the artist, but the lyrics, the music composition and specific rhythm patterns. The end result is that your station just changed ever so slightly, and will deliver more music similar to the song you liked.
While some companies don’t actively discuss how machine learning may be helping them, others are actively promoting it as a tool in their arsenal. For example, Amazon Web Services (AWS), Amazon’s cloud platform, explains that they have been “investing deeply” in artificial intelligence for more than two decades. With machine learning, the company uses the information to drive sales and analyze paths in their fulfillment centers.
Similar practices can be used with subscription payment systems, so that companies can better predict outcome from various pieces of data information.
Steps in developing a machine learning system for recurring billing
- The machine learning system has to be trained. The training portion involves loading subscription data from users. The more data that can be inserted in this step, the better the predicted outcomes.
- Constraints are developed. This may include how much a customer is billed, the billing cycle, etc.
- A ‘dummy’ user profile is created. This profile is based on any number of metadata. One piece of metadata would be the customer profile, including age, gender and other demographics. Another is the customer’s behavioral profile and history.A customer’s payment information can also be fed into the algorithm, such as the usage of a third-party billing system, a debit card or a credit card.
- Specific user information After the metadata is established and the generalized user profile established, it’s time to put the system to use by loading specific customer data.
- Predictions are created. These predictions are based on a number of different scenarios, such as charge successes and failures. It may predict that on specific days, a charge is more likely to go through, because it coincides with a paycheck deposit or other commonly scheduled deposit.
The outcome can optimize billing practices or make suggestions for future billing strategies, which, in turn can help reduce churn.