top of page

Getting Started with AI

Starting Right with AI

You might be eager to start using AI to optimize your business. Even if you have some idea of what AI can do, you might not know what’s necessary in order to use AI in a way that provides real value to your company. Getting started with AI seems hard, but breaking it into four steps makes it much easier.


  • Decide what to predict.

  • Get historical data in order.

  • Turn predictions to action.

  • Enhance your actions.


To illustrate these steps, let’s look at customer attrition at the bank American Savers Cooperative (ASC). Attrition prediction and customer retention is one of the most common AI use cases for any business. So let’s look at how ASC uses AI business tools to keep their customers.


Decide What to Predict

The first step in using AI effectively is to figure out just how to tell it what you really want to achieve. You do this by clearly defining, in measurable terms, what you are trying to predict.


Consider the case of customer attrition at ASC—what does it actually mean to lose a customer? Let’s say a customer holds both a checking and savings account and they close the savings account. Did they attrit? What if a customer has only a single account with the bank but they just moved 90% of the assets out of it. Is this attrition?


Before ASC can use AI to help predict and reduce their customer attrition, they must first come up with a concrete definition of what attrition means for their business. They should be able to look at their data and clearly answer whether the customer did or did not attrit.


Get Historical Data in Order

The second step in using AI effectively is getting your historical data in order. As the saying goes, “The best predictor of future behaviour is past behaviour” and this also holds true for artificial intelligence. 


In ASC’s case, they already have a lot of historical data about attrition. They also have a concrete definition of attrition and can answer confidently whether a customer did or did not attrit. That information is exactly the kind of data AI needs to train itself. In fact, one truism of AI is that if you can’t report on it, you can’t predict it.


Note

Some companies may need to consolidate data spread across many disparate systems, requiring integrations to get all the data into one place. It can be a real challenge, but it’s absolutely necessary.


Many businesses are analytics-driven, and they already run many in-house reports to measure the health of their business. This leads to another truism of AI: if you do report on it, you often want to predict it. In fact, the outcomes measured in these reports are a great starting point for AI optimisations. The outcomes have already been decided; the data is already there.


Turn Predictions Into Actions

The last step is to turn prediction into action. In the case of ASC bank, their AI is predicting if a customer will attrit. For that reason, the results will come back in the form of a probability. For example, one customer is predicted to have a 15% chance of attrition whereas another customer has a 30% chance.


ASC could use this number in a number of different ways. They could put it right on the contact record. Better yet, they could give their retention team a list of customers, prioritized by likelihood to attrit. Maybe it’s time to offer a special promotion to all customers with over a 25% chance of attritting.


Enhance Your Actions

Now that ASC wants to send a special promotion to all their customers, they need help creating an email. With the help of AI, specifically gen AI, they ask Einstein to generate a special promotions email. Einstein populates an email and they’re happy with the response. They change a few words to make it sound more personalized and send off the email to all their customers. Not only did they make progress with their attrition predictions, but they showed their appreciation toward their customers.


To make the most out of AI you must have a concrete definition of what outcomes you want to optimize, historical data to train on, and a plan of action for how to use predictions. 

Comments


bottom of page