When Capital One rolled out its automated fraud-check texting system a few years ago, the bank was surprised to find how many possible answers its customers could find to a simple yes-or-no question.
While around 85 percent of people responded to the alerts about potentially suspicious credit card activity with a “confirm” or “deny” as directed, the rest of the replies ranged from fat-fingered typos to a “yep, that was me” or even intimate bits of travelogue—“yes, that’s the purse I bought in Philadelphia last weekend while visiting my sister”—says Ken Dodelin, the bank’s vice president of conversational artificial intelligence product development.
