Machine Learning offers a water bill discount to wealthy Portlander

Tim Boyle couldn’t quite believe his eyes. The CEO of Columbia Sportswear pays a lot of bills, both personally and for his company. It’s unusual for a supplier to offer him a big discount on something he pays full price for.

But there it was: a letter from the Portland Water Bureau. “Dear Timothy Boyle,” the Oct. 14 letter read: “Your account has been randomly selected to participate in the Smart Discount program.” Discount percentage: 40%.

Every year, thousands of Portland Water Bureau customers fall behind on their payments. Last year, KPTV reported that clients owed the agency $28 million in late payments.

Boyle wasn’t one of the customers who fell behind. Last year he earned more than $1.6 million in compensation, and his Columbia shares are currently worth more than $1.7 billion. He also uses a lot of water: The last time Ww did his ‘Water Hogs’ feature, in 2021 Boyle was rated as the city’s ninth largest residential water user, using more than 770,000 gallons, which we said was “enough for 124 Portlanders to take an eight-minute shower every day to take’ a year.”

In short, Boyle does not seem to be the ideal candidate for a 40% discount. So what gives?

The answer: AI.

THE PROBLEM: The Water Bureau recognizes that many customers are unable to pay their bills in full. So it has a $10 million customer assistance program, similar to programs that electric and gas companies offer that subsidize customers who can’t pay. The problem is, as the Water Bureau explained to the city council earlier this year, that the program, which is means-tested (i.e. only offered to cash-strapped customers), regularly falls short of its target of serving 10,000 customers annually. to help. The agency has been doing this since its founding in 1995. In fact, the agency has never given discounts to more than 8,500 customers in a year.

THE SOLUTION: In February, the City Council approved a $350,000 contract with a vendor to use machine learning — a subset of artificial intelligence — to more effectively distribute rebates to needy customers. The contractor, a company called SERVUS, is running two 90-day pilot programs to test whether the algorithm can accurately identify customers’ ability to pay. Boyle is one of the customers who selected the algorithm for the initial test, which covers a 90-day billing period. It doesn’t mean he’ll be selected for the second round, or for the long-term discounts offered to customers next year.

The Water Bureau hopes that during the trial periods, SERVUS can identify customers who would qualify for a discount but have not sought it in the past. “With a randomized control trial, we’re reaching a broader, more diverse group of customers, capturing people in different financial situations – those who typically don’t apply for jobs or feel comfortable sharing income information,” said Quisha Light, director of client services at the Water Bureau. “This unbiased, randomized approach gives us a true picture of how the program could work for everyone, not just those who already feel like they need help.”

BOYLE’S TAKE: Columbia’s water-guzzling director says any scenario in which a customer is offered a discount on his resources is an indication that the Water Bureau needs better management. “It’s ironic that I’m a water pig and the agency is now offering me water,” Boyle says. “I don’t want the discount; they should give it to someone who needs it.”