“What happened? Did Amex computers get dumber? Amex people feel less adventurous?”
Both are certainly possible. I can understand financial institutions exercising a (possibly over-the-top) amount of caution when trying to identify behavior, based on consumers’ wariness of anything from a bank that feels creepily-too-familiar.
For a counterexample that’s just as underwhelming, the number of false positives generated by Mint is truly staggering, particularly given the data sample size they must have at hand.
“You spent $3,000 on fees last month!” gives you a minor coronary the first time, but by the second and third time, you’ve realized it’s just an algorithm’s inability to accurately assess context given incomplete (but rich) information.
I agree with your initial premise — in both cases, I think this is an example of companies trying to justify their value. This quest is going to create lots of churn in the AI market for a long time to come.