Subscribe Now!
There is no shortage of data providers pitching database marketers on how great their data source is.
Do good database marketers throw away all their existing data sources to use the new and (allegedly) improved data? No. They run tests. They run simulation models to see if the new data source produces different results indicating who to market and how to market them, and they test the potential new data source to see if it produces superior response and conversion rates to the existing data source(s).
Too bad customer service doesn’t do this.
Or at least, it doesn’t seem like they do this. From what I can tell reading the banking press and blogosphere, customer service departments in banking seem ready and willing to shut down all existing service approaches and rely on AI (artificial intelligence).
But will AI-driven customer service truly provide superior customer service? And how would we even define what “superior” service is?
Is anybody testing this?
Let’s Run a Test…
To test the hypothesis that AI delivers superior customer service, we have to define use cases, or situations, that can be evaluated. Honestly, I don’t know what those use cases should be. They can’t be too simplistic, like simply telling someone their account balance. And they can’t be so complex that the situation only arises once in a billion interactions.
I was thinking about this recently while I was at the airport. I was traveling from Des Moines to Chicago, ticketed on a 4:00 pm flight. My boss was also going to Chicago, and unbeknownst to me before we left for the airport, was on a 2:22 pm flight on the same airline.
I went to the gate agent, and asked if the 2:22 flight was full. It wasn’t, plenty of seats available. Great, could I switch my flight to the earlier flight?
“Yes, but because you have no status on the airline, Mr. Shevlin, it will cost you $75.”
“Seriously? You can’t waive the fee?”
“No.”
“My boss on this flight is a Golden Supreme God with your airline. Couldn’t you waive the fee for me on behalf of him?”
“NO.”

Using Digital to Accomplish Business Goals
See how this bank increased application clicks by 115% and new users by 30%. Download the case study.

How to Build a Successful Business Case for CRM
Convincing leadership to invest in CRM can be challenging. Be certain to do the groundwork, show the data, and prepare for the questions.
Read More about How to Build a Successful Business Case for CRM
So I stayed with my 4:00 pm flight. And I will be sure to not choose American Airlines every chance I get (and I get lots of chances).
In the end, the airline didn’t get the $75, and the whole experience pissed me off. I missed out on a chance to spend an hour with my boss telling him how great I am (wait, maybe the airline DID make the right decision… for HIM).
Key Question: Would AI-driven customer service have made a different decision?
That is, could a bot access data about my flying habits, recognize that I fly a lot (and in particular from BOS to PHX, an American Airlines route), and have made a different decision than the gate agent?
Or better yet, would AI-driven customer service have done the right thing and said “sure, Mr. Shevlin, no problem” in a demonstration of good customer service trying to build goodwill with a customer… on a flight that had empty seats anyway??
Simply providing data-driven answers to easy questions is not a demonstration of AI’s potential to provide superior service. The test has to come from making hard (and often data-starved) decisions.
Read More:
- The Use of AI in Banking is Set to Explode
- Rise of the Machines? Consumers Say They Don’t Trust Financial AI
- Banking Must Move From Mobile-First to AI-First
The Data Challenge
There’s another thing that makes me question whether AI-driven customer service will truly deliver superior results.
AI needs data. For AI algorithms to self-correct/self-adjust over time, the tools must have access to the results of the decisions, recommendations, or prescriptions the tools previously made.
Tools like robo-advisors can access real market results to gauge whether or not prescribed advice was good advice. But where is that feedback loop in bank customer service interactions? How does the data that determines how well a service interaction actually solved, alleviated, eliminated, or even reduced a problem get captured and fed back into the AI tools?
Banks don’t have that data. I don’t even hear anybody talking about how to capture it. Nevertheless that doesn’t seem to stop anybody from claiming that AI-customer service is going to be the answer to all of banking’s prayers.
We need to do testing. More testing.