Artificial intelligence could be used to hunt rogue brokers

by Miklos Bolza13 Dec 2016
A new prescriptive algorithm has been created which could one day be used by ASIC to spot fraudulent behaviour by rogue brokers.
 
Launched by tech start-up Veriluma, the platform can be embedded behind an aggregator’s CRM system to track important data such as client details and product information. It is designed to spot any subconscious bias the broker may exhibit when delivering their services.
 
“Bias can be by omission; bias can be hindsight; bias can be familiarity,” Richard Howard, advisory board member for Veriluma, told Australian Broker.
 
“What happens if I have dealt with a particular bank several times? Bang, bang, bang – three loan successes. Immediately I throw the next loan to that bank but is that the suitable product?”
 
The firm has already talked to ASIC about new tech that gives the regulator better oversight over the 23,000 brokers and advisors in Australia, said Veriluma CEO Elizabeth Whitelock.
 
This would ensure advisors are acting in the best interest of clients with the products they are recommending, allowing ASIC get on top of individual brokers before they go rogue, she added.
 
“From a strategy perspective, we may not know who the rogue individuals are but what we might be able to spin up is a model which actually shows us what the behaviour may look like.
 
“We can take real data and ingest that from different scenarios. This might start to give us a flavour so that when we start to see that behaviour coming through, we can stamp down on it before it gets out of hand. That way, we can be more proactive.”
 
A better fit
 
The algorithm may also help brokers find the best products for clients by feeding information about the individual and certain products into the system, Whitelock said.
 
“That could sit in the background doing a quick assessment based on what we know about the client and what they’re looking for – are these the right fit?”
 
This means the broker no longer has to sift through hundreds of individual loan products one-by-one, she added.
 
“It’s not only the suitability of the product,” said Howard. “That’s an idealistic scenario – which products am I most suitable for – but then what’s my probability of approval?”
 
The background assessment would be looking at both of these factors, Howard added, and would compile a list of suitable products ranked by the likelihood of approval.
 
“That’s a process which could take a broker days,” he said.
 
However, this system was not meant to be a replacement for broker-led decisions "but it can be a check,” he added.
 
Veriluma tackles individual problems – such as which loans will be most suitable for a client – and then examining what that problem looks like.
 
“What’s the question you want to answer and what are the elements that make up all of that?” said Whitelock. “Underneath all of those elements, there may be a dozen further elements, so we may end up with a hundred different issues that become information points. That’s where we start.”
 
The flexibility of Veriluma’s prescriptive analytics allows it to be tailored to the needs of each broker’s clients.
 
“The algorithm’s already patented so we don’t touch that. The only thing that changes is the clothes that you want to wear,” said Whitelock. “It becomes a layer on top of the engine.”
 
This means that, for mortgage brokers, Veriluma could train some of the firm’s analysts to build a model specifically for the type of clients being catered to.
 
The company is currently gathering research in the financial services sector by talking to brokers, aggregators and other parties in the value chain to see how the algorithm can be used in the mortgage space, Howard said.
 
Related stories:
 
Fintech platform launches new digital loan push
 
Aggregator pioneers new digital platform
 
Five lenders share loan automation strategies

COMMENTS

  • by Brado 13/12/2016 8:45:30 AM

    the problem with this statement; "but is that the suitable product?" is that we don't have to find the 'suitable product' but find one the is 'not unsuitable'... so is this algorithm defunct before it starts?

  • by chrisc 13/12/2016 10:09:35 AM

    We seem to be being pushed into socialist regimes by the day - not enough that the banks dictate everything to us, now to be controlled by computers (or the people who invent the programs - with their / their sponsors flavours in built / who is the sponsor for this latest one); the brokers and the clients will have reduced say or subjectivity or choice - how many times have we seen where the current credit code (consumer protection) favours the bank over the consumer - the banks grow even stronger in then being able to fit their programs to the algorithm which in time they will - its all computers Vs computers - is this being tried on all Government Depts too - that may have many more and better far reaching outcomes than picking on Brokers all the time or does sponsorship and politics get in the way of it there.

  • by Brett from Brisbane 13/12/2016 10:22:47 AM

    Of course, the algorithm will have analysed your business plan and target market, interviewed the client and analysed the qualitative and quantitative data accumulated at the interview stage before determining “a suitable product”, which we all know we can’t do. Then the overlay of urgency on SLA overs combined with approaching contract clause requirements which could deem some products unable to achieve delivery by the specified timeframes, which causes disharmony and anxiety at a client level. This is something which the algorithm may have a difficult time coping with.
    A ‘robo advice’ model as an audit check can only be the continuation of a draconian and naïve understanding that programmers and officials seem so keen to push around in wheel barrows.
    If the industry wants to look at rogue brokers, maybe look at the percentage of application volume for that broker and question why the lodgements seem skewed towards on lender. This algorithm (not digital) has been in the industry for many years at an aggregator level.