NAB Runs Loan Deferral Calls Via Voice Analysis – Finance – Cloud – Software
NAB has run tens of thousands of recordings of outgoing calls through voice analysis models to verify that customers received the correct information on loan repayment deferrals during Covid.
Data manager Glenda Crisp revealed the project at Melbourne Business School’s business analysis conference last week.
While speech analytics was first applied to a Covid-related use case, Crisp said the bank has been evaluating the technology since before Covid, but the onset of the pandemic has moved work forward. .
One of the first challenges was that the bank’s preferred technology, Deloitte’s TrueVoice, operated as software as a service, but could not be packaged and run inside a client’s own cloud presence. .
Crisp said she and NAB were uncomfortable sending large amounts of call recording data to a platform as a service.
“I believe data has gravity, and so we really need to bring models to the data,” she said.
“I don’t think it’s viable, at least not for a bank, to send large amounts of data out of the organization to a model.
“Considering the kind of data we hold, I’m generally not comfortable with sending large amounts of it outside.
“TrueVoice started out as a software as a service offering, which would have meant we should have shipped our data.
“But we’ve worked very closely with them to basically create a containerized deployment outside of the Deloitte environment that runs in our environment on AWS.
Crisp said Deloitte had been “remarkable” in effectively modifying TrueVoice’s operating model to meet the needs of the NAB.
“The processing engine they have has proprietary algorithms to convert audio to text and to perform sentiment analysis. They use machine learning algorithms and apply voice analytics for feature extraction, but now it’s all done securely on our NAB Discovery Cloud (NDC) platform on AWS, ”said Crisp.
“By running it in our environment, we also ensure that all of our governance processes, ethical and privacy frameworks are applied, all security controls, etc. are all enforced and that it works in a container in our space. . “
Voice recordings flow from the platform they are captured on to the NAB’s data lake – the NAB or NDH data hub – and to the NDC where the TrueVoice model is applied.
TrueVoice uses an AI engine to transcribe and analyze the content of a customer’s call.
“It takes unstructured data that gets trapped in recordings, emails, chat transcripts, etc. Said Crisp.
“Then we go through a voice recognition process that transforms sound into text.
“This technology really automates a range of business processes for us, and it includes several types of technologies, from call recording to voice recognition and transcription, keyword research, information retrieval. actionable, conversation quality management, behavioral analysis, etc.
“There is a wide range of abilities that need to come together to really do speech analysis well. “
In the past, similar processes, although more manual, would have been applied to a sample of calls; TrueVoice provides the NAB with assurance that it has met its compliance obligations for all calls, not just those it sampled.
Several models have been developed and trained so far, and more are in the works.
TrueVoice has so far been rolled out in NAB Assist, the part of the bank that handles appeals and tough cases.
“We implemented it late last year, and the first components of TrueVoice have been deployed with over 60,000 outgoing Covid registration calls,” Crisp said.
“NAB Assist used voice analytics to support compliance monitoring of conversations with our clients who had suspended mortgage payments during Covid.
“The compliance check was done on what we called the registration calls, which are halfway through the postponement, and the expiration calls, just before the end of the postponement.
“The purpose of this scenario was to ensure that the proper disclosures were made by our bankers to our clients.
“We were also looking to understand their feeling and make sure the right treatment plan was being followed.
Using TrueVoice, Crisp said the bank was able to identify some missing postponement disclosures from its outgoing calls.
“This then meant that we had a shortlist of clients that we had to call and make sure we had given them that information correctly,” she said.
“Basically because we identified these customers, they got a higher priority score for a return call.”
Registration calls were also passed through a vulnerability model also built on TrueVoice.
“This signaled a few hundred calls from customers that we felt we had to listen to manually because the customer might experience additional distress,” Crisp said.
“Once we listened to them manually, it resulted in a few customers that we had to reach out to and really give them additional support. “
Achieving the same level of compliance assurance using sampling and traditional methods would have taken around seven years to be performed manually, Crisp added.
TrueVoice should also bring benefits by being able to recognize exemplary examples of customer service and learnings that might be of use to other bankers and agents.
“Where we can see calls that have gone pretty well, compliance has been met and there have been good [customer] feeling, so we’re looking to capitalize on those learnings and coach others so that they can benefit from it, ”said Crisp.
“If we can find, say, a banker who’s really good at ticking all the compliance boxes but also good at customer engagement and positive sentiment, then we’re going to want to learn from that. conversational style and approach and apply it more generally to others so that we can both respect compliance and have a good customer experience.
Crisp said what the bank has achieved so far with TrueVoice represents a minimum viable product phase (MVP) for the technology.
“We are finishing all our industrialization [of the technology] this month, and I have a backlog of additional use cases to apply to the model, again primarily from a quality compliance and improved customer experience perspective, ” she declared.
Crisp also said the bank will conduct additional training to train contact center staff and data scientists on when and how to leverage TrueVoice, reducing the bank’s reliance on outside help.