Latest / The Best of the Money Show / Will AI replace human judgment in loan approvals?
Transcript
- 0:00Speaker 1: Well, more confirmation over the last few days of the
- 0:02Speaker 1: difference AI is making to business practices in a week
- 0:06Speaker 1: in which Anthropic is talking more about its initial public offering.
- 0:11Speaker 1: Reuters reporting Deutsche Bank says decisions about lending that used
- 0:16Speaker 1: to take several weeks now just taking a few moments.
- 0:18Speaker 2: Dr.
- 0:19Speaker 1: Mark Nassila is the chief data and analytics officer at
- 0:22Speaker 1: FNBAI Strategy. Mark, good evening. Good to chat again. I mean,
- 0:26Speaker 1: this is exactly the kind of thing that AI is
- 0:29Speaker 1: good at. It's maths, it's risk. I would imagine all
- 0:32Speaker 1: the banks are using AI for exactly this kind of decision.
- 0:35Speaker 2: Good evening, Stephen, and to the listeners who are joining us.
- 0:39Speaker 2: So this is exactly the same applications we're seeing locally.
- 0:46Speaker 2: the fact that Deutsche Bank is cutting loan decision times
- 0:49Speaker 2: from weeks to minutes. And we've got similar applications locally
- 0:53Speaker 2: at FNB. It highlights its ability to ingest, process, and
- 0:58Speaker 2: analyze massive amounts of unstructured and structured data almost instantly. Now,
- 1:04Speaker 2: if you just review traditionally, loan processing or credit processing
- 1:10Speaker 2: was very manual traditionally from manual data entry to historical reviews.
- 1:16Speaker 2: What this technology is doing, it's shifting this to a
- 1:19Speaker 2: proactive model where risk is assessed dynamically, allowing banks to
- 1:23Speaker 2: serve clients at the speed of their business needs. Essentially,
- 1:27Speaker 2: what we're seeing from these applications and what we have
- 1:30Speaker 2: at F & B is not just about doing things faster.
- 1:33Speaker 2: It's about removing friction from banking and making sure we
- 1:36Speaker 2: unlock efficiencies and even more economic opportunities for our clients.
- 1:41Speaker 1: Do you believe, or is it possible to know, if
- 1:44Speaker 1: the decisions made by AI in these lending processes are
- 1:49Speaker 1: the same as the decisions that would have been reached
- 1:53Speaker 1: when humans were more involved in the process before?
- 1:58Speaker 2: So, banks and other organizations, and I can attest to
- 2:01Speaker 2: what we do at F & B, we rigorously test
- 2:04Speaker 2: these models before starting to use them. And we make
- 2:08Speaker 2: sure that responsible AI and The nature of these decisions
- 2:13Speaker 2: are at the center of driving trust. While we know
- 2:19Speaker 2: that AI scales, so does the need for governance and
- 2:23Speaker 2: rigorous monitoring. Because if we treat AI as an afterthought,
- 2:27Speaker 2: it's likely to even lead to more reputational decisions. So,
- 2:31Speaker 2: for example, we have to make sure that the AI
- 2:36Speaker 2: models are as good as the data they're trained on.
- 2:39Speaker 2: So we audit data sets to ensure that they are diverse,
- 2:43Speaker 2: they are representative. We try to make sure we don't
- 2:46Speaker 2: have historical biases in the way decisions are made. We
- 2:49Speaker 2: continuously monitor and draw a threshold to detect whether these
- 2:53Speaker 2: models are working the way they were expected or they
- 2:56Speaker 2: are drifting. But most importantly, human in the loop and
- 3:00Speaker 2: explainability are part of the strategy. So it's a human-led
- 3:04Speaker 2: enabled by AI. We know that... with high-stakes decisions, credit
- 3:10Speaker 2: or even risk modeling. Some of the applications include providing
- 3:14Speaker 2: advice to our clients, using AI for risk management to
- 3:18Speaker 2: enable our due diligence analysts to make decisions. And therefore,
- 3:23Speaker 2: trust is the currency, Stephen. It's not just about even guardrails.
- 3:28Speaker 2: It's about engineering safety to ensure our AI is transparent
- 3:31Speaker 2: and biased and gives us decisions that we expect on
- 3:35Speaker 2: an ongoing perspective.
- 3:38Speaker 1: Would it be able to tell you why a loan
- 3:40Speaker 1: decision was made in a particular way? In the past,
- 3:43Speaker 1: if I went to a bank manager a very long
- 3:45Speaker 1: time ago, they might have been able to say it
- 3:48Speaker 1: was turned down because of this or turned down because
- 3:50Speaker 1: of that. Will the AI model tell you that?
- 3:54Speaker 2: So we've got techniques to make sure that models have
- 4:01Speaker 2: explainable outcomes. When we talk about human in the loop,
- 4:07Speaker 2: we're not just talking about a procedural arrangement where human
- 4:12Speaker 2: beings just rubber stamp decisions. So people are empowered to
- 4:17Speaker 2: make sure that they can explain these decisions, what types
- 4:21Speaker 2: of variables or attributes led to a decision, whether it's
- 4:24Speaker 2: a fraud recommendation or a marketing or a credit. They
- 4:30Speaker 2: are empowered with technical insights. And this is also done
- 4:35Speaker 2: through fluency programs we run because for people to understand
- 4:39Speaker 2: and explain these decisions, they have to understand how AI works,
- 4:44Speaker 2: its limitations, the risks that come with the use of AI.
- 4:48Speaker 2: So definitely that is part and parcel of maturity in
- 4:53Speaker 2: adoption or else it becomes very difficult for us to
- 4:56Speaker 2: understand the risks we're getting into.
- 4:58Speaker 1: The people who were involved in making the decisions before,
- 5:01Speaker 1: if it took so long, there would have been several
- 5:03Speaker 1: of them. What happens now? Those jobs no longer exist.
- 5:06Speaker 1: They might supervise. Do you see actually they're as busy
- 5:10Speaker 1: as they were before because the volume of business increases
- 5:14Speaker 1: as a result of AI?
- 5:17Speaker 2: So for us, you know, our engine is contextual banking.
- 5:24Speaker 2: And contextual banking, Stephen, is not just about the algorithm.
- 5:28Speaker 2: It's about complementing the algorithms with a human touch. And
- 5:33Speaker 2: this involves making sure we are repurposing people to things
- 5:39Speaker 2: that AI cannot do today. And there's a lot of that.
- 5:42Speaker 2: So there's things like angst or things that require people,
- 5:48Speaker 2: our customers want to speak to a human being. You
- 5:51Speaker 2: just talked about explainability, and that is well done by
- 5:55Speaker 2: a human being, even though it's recommended by an algorithm.
- 5:58Speaker 2: We talked about, for example, closing out fraud investigations. At
- 6:03Speaker 2: the end of the process, it's a human in the
- 6:06Speaker 2: loop doing that because a human being is well-placed. We
- 6:09Speaker 2: also understand that we have to manage this algorithm. So
- 6:13Speaker 2: there's so much opportunity that I've opened up in the
- 6:16Speaker 2: AI development and management space that people are being repurposed to.
- 6:21Speaker 2: And again, this is being powered by our literacy and
- 6:23Speaker 2: fluency programs that we're running across the group.
- 6:27Speaker 1: Thank you very much indeed, Dr. Mark Nassila. Really do
- 6:30Speaker 1: appreciate that. Chief Data Analytics Officer at F & B
- 6:34Speaker 1: AI Strategies.