ai

Inside an AI-Powered Savings Algorithm

A penny saved via algorithm is a penny earned
article cover

Giphy

3 min read

Keep up with the innovative tech transforming business

Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.

A penny saved via algorithm is a penny earned. Now try 500 billion pennies—i.e., $5 billion. That’s the amount AI-assisted savings app Digit has socked away for its users.

  • How it works: Using supervised machine learning, Digit’s model analyzes a user’s spending patterns, then aims to tuck away unnoticeable amounts.

Hayden chatted with CEO Ethan Bloch for an inside look at Digit’s algorithm.

Mixed signals

Signing up for the app allows it to analyze months’ to years’ worth of your transactional data—and to make judgment calls on your behalf. Like many personal finance apps, Digit shares data and analytics with some third parties and can use aggregated personal info to recommend products or display content.

Four main signals are behind every savings decision:

  • Your checking account balance (and whether that’s high or low for you)
  • When you’ll likely be paid next (and how much)
  • Which bills may be coming due next (and what they’ll cost)
  • How you’ve spent over the past week (and how that compares to your norm)

Since Digit’s model uses supervised ML, it’s in constant competition with itself to improve. Each time a new version of the model wins out, an engineer decides whether to greenlight it.

Getting schooled

It took the Digit team between six and nine months to develop V1 of the algorithm. The process was especially slow-going because—like any K-12 kid—the model only had five days a week to learn. (Money transfers + weekends go together like peanut butter and mayonnaise.)

  • Another hang-up: For every user, Digit’s algorithm makes a savings decision once daily—meaning it initially took the team a full day to gauge the effects of any changes.

The fix: In the early days of algorithmic development, engineers built a basic savings simulator that allows for up to 15 simulations per day, providing a boost to learning speed.

It’s complicated

As Digit adds more customization—and, soon, investing and retirement features—it’s had to tweak the algorithm in ways it “wasn’t designed for in the beginning,” says Bloch. In effect, he means AI’s greatest nemesis: common sense.

Scenario: Let’s say it’s safe to move $40 out of your account today, but you’ve got three goals: building up an emergency fund, paying down credit card debt, and saving for a vacation. How should your $40 be divided?

  • Even a human financial advisor might need to consult with you here. For ML, which lacks context and nuance, it’s a much bigger hurdle.

“We still have a lot of work to do, I think, to make this more intelligent,” says Bloch. “We’re just at the beginning.”

Keep up with the innovative tech transforming business

Tech Brew keeps business leaders up-to-date on the latest innovations, automation advances, policy shifts, and more, so they can make informed decisions about tech.

T
B