WSJ: Customer-Service Algorithm Boosts Revenues
Afiniti in the News

WSJ: Customer-Service Algorithm Boosts Revenues

Customer-Service Algorithm Boosts Revenues

By John Murawski, The Wall Street Journal

A matchmaking algorithm trained to predict personal rapport between call-center agents and customers is making inroads at companies and boosting their revenues.

G6 Hospitality LLC, the parent company of Motel 6, is one of the latest businesses to adopt the artificial-intelligence technology from Afiniti Inc., using it at a customer-service center in Puerto Rico and with contractor agents who work across North America.

When customers call, the algorithm uses their phone numbers to access data that may include census information, credit scores and details pulled from providers such as credit-reporting firm Experian PLC, data clearinghouse Acxiom Corp. and marketing technology firm Allant Group LLC. The system typically ingests between 100 and 1,000 characteristics about callers: where they live; what publications they subscribe to; how often they travel; how often they upgrade hotel rooms; if they pay bills on time; if they pay by check, direct debit or credit card; how many phone lines they have; and if they are calling by cellphone or landline.

Within 50 milliseconds, the algorithm matches the caller to the available agent who has had the most sales and bookings with similarly profiled callers.

G6 has been using the technology since November and has seen a 4% revenue increase at its call center. The company is aware of the metric because it runs the system for 45 minutes and shuts it off for 15 minutes every hour, generating a continuous A/B test to compare outcomes, said G6’s chief information officer Jessie Burgess.

Afiniti said the algorithm typically boosts call-center revenue by 4% to 6%.

To date, more than 150 call centers operated by over 30 companies have deployed Washington-based Afiniti’s system. Clients include Sky PLC, Virgin Media, AT&T Inc. and Caesars Entertainment Corp.

However, pulling and integrating the data to make the system work can be a challenge.

It took Afiniti more than four months to test and implement the system for G6, requiring software upgrades to the hotel company’s 2000s-era automatic call distributor so that callers and agents could be paired within a split second.

Most deployments take three months or less. But Afiniti CEO Zia Chishti said accessing call-center data to train the algorithm can be challenging because clients’ databases are unique. In some cases it has taken a year to get a system up and running, he said: “It’s a fairly hefty lift.”

G6 saved time by limiting the amount of historical data that had to be pulled and converted for training the algorithms. Instead, the company used new customer data from incoming calls for training the system, Mr. Burgess said.

There was no loss in accuracy, Afiniti said, but G6 needed three more weeks to collect enough data to test the algorithm. However, pulling and converting the historical data would have taken two months, according to Afiniti.

While it may seem a mystery why people click with customer-service reps, the algorithm can pick up the likelihood of good chemistry, said Chris Farmer, Afiniti’s chief marketing officer.

Laws prohibit companies from classifying customers by race or ethnicity, though ZIP Codes can sometimes serve as a proxy for someone’s race or income.

Afiniti said it ensures it follows data-privacy laws and takes compliance seriously.

Not all customer data is relevant for making a good match. Early on, Afiniti tested Twitter feeds in the algorithm but decided against scouring social media because that information didn’t increase predictive accuracy, Mr. Farmer said.

The AI call-routing system doesn’t cause longer hold times or other delays, and doesn’t require retraining or new scripts for the customer-service reps, Mr. Burgess said.

Customer centers are still largely dependent on humans. However, more operations are being automated. Natural language processing, machine learning and other types of AI are expected to fully automate 40% of customer interactions in 2023, up from 25% last year, according to an April report from Gartner Inc.

Originally published in The Wall Street Journal:

Keep Reading

See all News
'How patterns can power better treatment' - Zia's piece in The New Statesman

'How patterns can power better treatment' - Zia's piece in The New Statesman

Effective AI requires diversity and teamwork, says VP Caroline O’Brien

Effective AI requires diversity and teamwork, says VP Caroline O’Brien

Julie Bishop joins Afiniti's Advisory Board

Julie Bishop joins Afiniti's Advisory Board