Home Bots & Business Report: Do big claims for machine learning success in Banking have big proof point backing?

Report: Do big claims for machine learning success in Banking have big proof point backing?

by Gary Flood

Real-world confirmation of extensive penetration of machine learning into the FS sector seem to back up claims of significant ROI by automation vendors.

Just a few of the ways machine learning is being used in banking now include automation of document handling and processing for managing the entire journey of a loan or mortgage application. Financial services organizations are also said to be employing the techniques to manage increasingly more complex journeys requiring cognitive skills, with one investment bank’s junior banking team said to have used its help to cut the time and manual effort needed to produce the pitchbooks they need to offer clients by 65%.

From consumer and SME banking all the way to capital markets and customer onboarding, it seems at least some customers are harnessing AI in some way to help, with millions of dollars claimed to have been won back by investing in such approaches; there’s even the possibility of slashing the time wasted on dealing with email through automation, which could give human colleagues back at least 2 minutes per email triaged and administered on their behalf.

It’s an amazing and inspiring list. And we’re not saying they’re not possible, or even used daily in some environments… but when it comes down to the true ‘art of the possible,’ users tend to want to hear more of the real-world war stories and experiences of financial services peers like themselves.

To help surface some of these use cases, last week we tuned into an online discussion, Machine Learning in key areas of Banking: Streamling processes and leveraging a business-led approach. Hosted by RPA leader UiPath, there was no shortage of high-level thinking best practice shared by the discussion’s host, the company’s Senior Director – Banking Industry Practice, Nitin Purwar—but the insights offered by his two banking industry guests that we really locked on to.

These were offered by Alvaro Castillo Alsina, an Innovation Manager at Barcelona-based Spanish banking giant CaixaBank, and Onur Deniz, Development Manager, Natural Language Processing at one of Turkey’s biggest national commercial banks, Yapi Kredi.

Alsina explained that for the least three years, his institution has been using a process improvement methodology utilizing a manual and internal subject matter expert approach analyzing workflows end-to-end to try and spot areas of inefficiency.

Why ‘La Caixa’ is now a process mining convert

This was first tested out, he said, in two key CaixaBank back office processes–mortgages and customer complaint handling. CaixaBank was already using AI and machine learning in separate areas like ID, customer services, and HR. But because of the investigation, he said, automatic data extraction from legal documents for mortgage application processing was immediately introduced.

The results, he said–in both use case—have delivered valuable cost savings, employees time reduction, as well as better user for both internal and external clients. However, the methodology that had identified potential applications for machine learning optimization were not suited for repeat usage or scaling, so the team has instead switched to use of process mining to find the next wave of target processes.

“We’ve performed several proofs of concepts with different players, including one for consumer launch. Again, the results seemed very promising, with potential to capture value in terms of cost reduction, increased efficiency, and a better user experience mainly by reducing the ‘time to yes’.” Machine learning also identified potential sales off better customer leakage analysis.

Based on his experience, Alsina is now convinced process mining is a quicker way of finding opportunities for optimization and automation. “Nevertheless, in life nothing is perfect,” he cautioned. “Implementing process mining has difficulties or implication that must be addressed and analysed for every use case: for example, information availability, as well as homogeneity, unfortunately is not the case in all applications.” Other areas he says banking organizations need to be careful with in working with process mining include cross-team data interpretation, while the right choice of tool can be affected by the volume of data you want to use on it.

Still, “We concluded that process mining solutions are really amazing, work very well and they bring a lot of advantages, but its industrialization must be progress through identifying areas with specific opportunities,” he warned. “For each use case, there are several steps to be achieved with business case.”

CaixaBank will continue to use AI for processes analysis and improvement, while also exploring equally advanced technologies likes digital twins, knowledge graph “and any new disruptive technology that we identify”.

‘Significant AI investment since 2003’ at Yapi Kredi

In contrast, Turkish financial services leader Yapi Kredi is much more into the speech analysis end of AI than process mining. Deniz explained that his part of the tech team is looking at the application of both natural language processing (NLP) and applied data science, with both working on applications of AI machine learning and deep learning in banking areas.

“At Yapi Kredi, we have applied many machine learning to many applications, in operations like helping with customer orders and signature transaction processing, business documents, ATM optimization and check processing,” he revealed.

That happens via a range of programs, from chatbots to RPA, he added, including In its call center in a few speech-to-text and text-to-speech uses. The bank is also using RPA from UiPath, where information extraction and classification models detect document type and if they can, extract information required for the operations and put the required information into an application, so it gets triggered.

But that’s far from the whole picture of AI at his company, he said:

“In core banking, we need customer segmentation, income prediction, cash needs prediction, credit risk models and financial analysis. We use machine learning to help there, and we also apply machine learning in different areas like in audit and control, where we’ve developed QA audits that process customer documents using AI-powered permission. We also use it for internal help desk tickets, where it provides leads for internal control operations,” he said.

Other daily use cases for AI at this organization include question-answer systems to support branch employees—a system used in over of its 800 branches by 16,000 users and which are estimated to provide answers to 35% of customer queries and so prevent the need for support tickets or human intervention.

Finally, the bank is employing machine learning help in fraud detection for its credit card operation, while security monitoring applications using machine learning tools is also “on the way.”

“The overall picture of machine learning at Yapi Kredi is that we’ve invested heavily in artificial intelligence and machine learning, in both internally and externally developed solutions. And that’s been going on with things like text document classification and information extraction since 2003,” he claimed.

Based on successes like CaixaBank and Yapi Kredi, that amazing list of machine learning help and all those possible millions of dollars in savings seems more like a beginning of this tech in banking—not the end.

Watch the full webinar on machine learning in key areas of banking here

Misschien vind je deze berichten ook interessant