Complaints management is a big, expensive headache. Banks and lenders have thrown so much money, time, and technology at it without any ROI.
But good news - AI is bringing us the answer.
“It's the tool I wish I had when I was in banking. And it's exciting to show it in action and see people's responses and talk through what's possible.”
We chatted with Scott Hamilton, who spent 30 years in banking with Bank of America, JP Morgan, Chase, and Capital One and is now Prodigal’s Banking Strategy Executive.
Here’s a quick quote overview of our discussion, or feel free to check out the whole conversation on our LinkedIn page (and follow us while you’re there - we love new friends!).
“Everybody has a slightly different definition of complaints, and they generally all lean on their frontline associates to execute and apply judgment on every interaction they have with every customer every day. And when you multiply those nuanced judgment calls across thousands of agents across millions of interactions, the ability to do that consistently is extremely difficult.”
“I was talking to one particular bank and they determined they were only capturing about half of what they defined to be the definition of a complaint. So about a third of those that were being captured were probably in the gray area that may or may not need to have been captured. But far more importantly, about half of the complaints that should have been caught weren't.”
“If you can uncover the root cause of complaints, you can definitely improve your revenue, your costs, and your regulatory standing.”
“Across your call centers, your text chat, CFPB, your office of the president, there's six or seven different ways a complaint can come into your organization. Then there's this aggregation process, there's a cleanup process, then a root cause process. So, by the time the complaints are aggregated and trended, it can be a month, in which case, if there was something broken in the organization, it would have gone on and festered for a month. So the pain in terms of revenue or cost or regulatory impacts is substantial.”
“What most of the Consent Orders and MRAs are based off of is simply identifying whether a complaint needs to be captured, period. So it's a yes/ no judgment call. But by far the biggest issue is capturing it to begin with. Because if you don't capture it, you can't dig into it.”
“The worst case scenario is if it gets to a certain level, from a regulatory standpoint when they do exams, they can have findings. If you have repeat findings, they graduate to a friendly MRA. And, unfortunately, if MRAs are not dealt with robustly, they can mature into a Consent Order. And they take years to address and in some cases can cause millions and millions of dollars worth of penalties. I've seen consent orders for complaints at multiple banks, all rooting back to this issue of really essentially capturing them consistently, and then getting to the root cause efficiently.”
“About 10 years or so ago, it was strictly training the agents on the front line to identify complaints, then when it was subsequently determined that one was missed, go back to the agent and do a little hand-holding and retraining. And then you repeat that process.”
“Then there's technology out there that assesses, in the voice case, transcriptions and looks for certain words. So ‘I'm mad,’ or ‘I want to talk to your manager,’ or ‘This transaction got declined,’ or whatever these phrases are. And that was helpful in identifying, finding sort of needles in a haystack. But they're very, very specific needles.”
“However, what most firms have found is those searching for those words, or that set of two or three words, created a whole bunch of false positives. And more importantly, false negatives, they're still missing, even with those tools, 50% of the complaints, so the technology really hasn't been there. And a few banks are continuing to optimize on that exact model.
“If you and I got together every Tuesday for two hours, we could come up with more and more phrases that would map to complaints. And that's exactly what happens. These tools convert voice to text into a transcription, and you tell it what to go look for. And every time you do that, you very quickly realize that you aren't catching everything. So you put more words or phrases into that list. And then the next week, you do it again. But at the end of that cycle over a year or two, you still land at only being able to identify half. It's just a ton of work. And at the end of the day, the tools just flat out aren't getting you what you need.”
“Where I came from, I was wrestling with the same challenge that most firms are wrestling with today. So what Prodigal has done and why I joined was we've built pretty advanced natural language processing and machine learning models, ones that only have come into existence three or four years ago.
“So we've built the models, we've passed a quarter of a million of these exact creditor lender banking conversations through the models, and have been able to train them to not just the words or phrases that are in question, but the context around which those phrases are spoken. And that's the secret sauce, when you can understand the context of the conversation or the tonality or the sentiment, you can really begin to build machines to apply this very targeted judgment on top of these conversations, and that's essentially what protocol does, in this case around complaints.”
In case you couldn’t tell, Scott knows a heck of a lot about how to manage complaints, and he loves learning what folks are doing about it.
Want to share what you’ve tried and what you’re working on now, or see how AI can work for you? Drop him a line, or if you’ll be at the ABA Risk and Compliance Conference in San Antonio next week, schedule some time to chat.
.svg){kind=icon}
-min.png)
.jpg)
-modified.png)
.png)
.png)
.png)
.png)
.jpg)
.jpg)
.jpg)
.jpg)
