When it comes to judging a proposal, can a computer really trump human intuition?
For the latest Motor Finance round table, in conjunction with Transition Computing, 11 experts from the UK motor and asset finance industries were invited to the Cornhill Dining Room in London to get to grips with the pros and cons of automated and manual underwriting and the challenges of dealing with consumer credit information.
Go deeper with GlobalData
Fred Crawley: What particular work goes on in your underwriting division in terms of scale of operation and how you underwrite?
David Gibson: At Transition Computing we manage systems that acquire new business and take them through to contract. Our clients include subprime Private & Commercial, prime and captive finance, and Nissan and Renault finance. We have written systems for small-ticket leasing, wrote the Dell Financial systems across Europe. As far as brokers are concerned, we wrote the original Syscap system.
Gary Hill:We’re a subprime to near-prime finance company, predominantly in motor cars. We lend about £1m a month with a typical balance of about £4,000, funding average cars to average people. We estimate 90% of our business is introduced through brokers.
Oliver Mackaness: We manually underwrite all of our deals. We probably get between 1,000 and 1,500 props through each month. We write about 200 deals. They come through XML links and good old-fashioned faxes.
How well do you really know your competitors?
Access the most comprehensive Company Profiles on the market, powered by GlobalData. Save hours of research. Gain competitive edge.
Thank you!
Your download email will arrive shortly
Not ready to buy yet? Download a free sample
We are confident about the unique quality of our Company Profiles. However, we want you to make the most beneficial decision for your business, so we offer a free sample that you can download by submitting the below form
By GlobalDataPeter Minter: Moneybarn is a non-prime vehicle finance business. Being privately owned, we’ve grown gradually over our 20 years in a very risk-averse way. Our target market is in the full range of non-prime, from subprime all the way up to the prime-line.
For the last three years we have embarked on a programme of significant automation in our underwriting process and witnessed major benefits from it.
Amanda Basham: We have a mixture of system-automated and manually underwritten decisioning. We manually underwrite around a quarter of the applications we receive.
Mark Smith:We are a relatively new company. We’ve been funding people for three and a half years, currently lending about £2m a month, we may do £2.2m this month. We have an aggressive expansion policy at the moment. We will exit this year lending circa £1m a week, 1,200 deals a month; and, exit 2014 anywhere between 4,000 and 5,000 written units a month.
We operate dealer-direct business through a range of high street shops. Our underwriting is manual to a certain degree but we’re going to bring in a bespoke automated scoring based on loan performances we’ve seen so far.
We’re expanding overseas. We open our first international branch in Vancouver in September and, hopefully, in November, maybe January, one in Denver, Colorado, to get a foot in the North American market.
Fred Crawley: What is the single largest challenge facing the management of your underwriting function in 2012? Is anybody taking on more sales without making a proportionate increase in sales reporting and administration?
Peter Minter: We’ve been through this process over the past three years. We had an entirely manual underwriting system, which we actually quite enjoyed calling "artisan".
We’d look at a deal and lovingly craft a solution which the customer would take. We’ve gone from that to a portfolio basis of underwriting risk, fully automated, in the subprime environment, where, with a certain amount of information, a decision can be made programmatically. It’s as fast as a computer can work and the wires can transmit it.
The biggest change is leaving honest, excited underwriting. And I honestly believed I was much better than any computer could ever be.
Then you look at the results. The fact of the matter is you really can’t beat doing it the same way, day in, day out, using a set of defined rules.
Thinking about what those rules are and making it an automated process, takes time and it takes an awful lot of guts to just let go. On a portfolio basis, if you apply the same rule, day in, day out, you know some decisions are going to be wrong, definitely, but you know it’s only going to be 1%. You just don’t know which 1%.
The challenge now is that the move to such an underwriting approach is based on past experience, on credit scoring and so on. And past experience never is a particularly good guide of future performance.
However, we are in another recession and the nearest source of information is your own, pre-credit crunch book and records from Experian, Equifax, CRAs over the same period, way back, before analysis. The question exorcising us at the moment is: Are they up to the job when peoples’ use of credit changes?
Peter Nolan: I would just counter: Running by a set number of rules may not work. We may be asked to underwrite an asset, for example, an MOT station for a garage, costing £30,000.
The credit profile of the garage says they’re not good for £30,000 but if you understand the asset, you know it will make them £500 a week, minimum, and, therefore, pay for itself over a five-year period.
I can’t get my head around how a computer can comprehend enough about an asset that you don’t lose the business in the first place. MOT stations have got really good residual value and pay for themselves, provided the customer has some experience in their field. The rate you can charge for a subprime company, because people only look at the previous performance, is definitely well worth taking the risk.
If an asset looks quite poor but you understand it will get paid for, even if can’t be paid for now, once in use, the asset generates its own income and you can charge the rate accordingly because nobody else will finance it. Nobody else is looking in that much detail at what it is you’re actually financing. That’s definitely the benefit of manual underwriting.
Paul Caunter: With us, there are two common deals that we write on our own paper: poor asset in terms of asset security but low-risk hirer, or some historical credit impairment we ensure we are safe in terms of loan to value.
Typically we are not ‘strong and long’ lenders on our own book, especially if there is some adverse credit; 80% LTV over three years is our preference.
There are two cases where the car’s been voluntarily surrendered in the last couple of weeks. In both instances, we managed to sell the vehicles and clear the outstanding debt and costs without incurring any loss. Our bad debt last year was less than 0.1%. It might be said we’re not lending enough but we want to ensure we are lending responsibly to people we believe have an ability to repay.
If automated, we would miss the opportunity to do some deals thrown out by the system, for whatever reason. Instead, we’re looking at proposals ourselves and finding deals are possible with a bit more deposit or with suitable guarantors, that’s the danger in terms of automating, from our point of view.
Paul Sheedy: The more niche the deal, the harder it is to apply an algorithm to underwrite it. A person has to look at that proposal.
If you’re trying to move into a market where you want to be a sausage machine and most of the deals are exactly the same – for the sake of argument, cars – then it becomes a lot easier to do that.
Some of the deals that have been coming across our desk are so bespoke and unusual that it’s hard to imagine how you would come up with a computer solution to underwrite that particular deal.
Helen Reynolds: We have, because of the nature of our business, some fundamental black and whites when it comes to decisioning: Populations where we will definitely not lend to, populations where we are definitely happy to lend to, and then we have our population in the middle that we want to look at in more detail, which is why our manual underwriting function is key.
Fred Crawley: Does anybody have any other general points about challenges to underwriting functions in the coming months?
Mark Smith: Changing regulation: if regulations are changed without discussion you will have to second-guess the correct procedure.
Peter Minter: It completely changes the risk profile.
Fred Crawley: Has CCD had any impact on SMEs when you’re looking at sole traders? Will there be changes up ahead?
Paul Sheedy: Definitely, when the threshold of £25,000 was removed. It used to be that anything over above that amount meant you could finance a car in an individual’s name, on an unregulated basis and now, obviously, that’s regulated which, from a finance company’s point of view, isn’t so attractive.
These agreements are very much weighted in the customer’s favour. It creates a problem for the finance company if you’re not set up for that kind of business.
We don’t get involved in regulated deals. We do the odd one or two but only if we have to really. That took a chunk out of the marketplace for us when this happened. It restricted our market. Deals we were doing before CCD we couldn’t do anymore on an unregulated basis.
Fred Crawley: Is the scope of manual underwriting proportionate to the number of asset categories an underwriting department must consider? Are there ways small-ticket asset underwriting could be dealt with by a business flow similar to that in car financing?
Allan Ross: Yes, if you’ve got an asset in the SME market. Understanding SMEs are lifestyles businesses is one of the things asset finance companies tend to miss out on.
If an SME has been in business five, 10 years, the likelihood is that they are a partnership or family business. They have their lifestyle involved in it and are not going anywhere. They are known in a local area. They enjoy and are emotionally attached to the business. The downside of such a business is far less than people would perceive.
We’ve currently got a book with ING of about £30m and we’ve got one deal in arrears over 30 days. A lot of finance companies don’t understand that. Automating that marketplace is easier than you imagine. It’s about longevity of business. If somebody has been in business five years or more with a clean credit history, chances are, he’s going to be fine to underwrite.
Mark Smith: I’d agree with that, actually. I’ve spent the past 12 months really analysing behavioural patterns of people that don’t normally pay their debts on time. Credit history is almost an irrelevancy, I’ve found. Providing that they have the right stability and income to support the loan, they pay.
Paul Caunter: The thing is: if everybody went automated, most of us round this table wouldn’t have a business.
Oliver Mackaness: Certainly some of the small-ticket market could be automated.
What concerns me is that we’ve got three underwriters and, actually, if we lost two of them, we would be in a mess. We rely quite a lot on these three underwriters. If something happened to them, we’d be stuck.
But then I think that one of our selling points is our manual underwriting. We could perhaps introduce some automation, especially the declines, and save quite a lot of time.
Allan Ross: It’s interesting, this whole story about automation in the motor trade, starting with Mercantile Automated Credit Scoring in 1982 with Mercantile Credit. I remember them bringing in Mercantile automated credit scoring system, called Macs. Branch managers were going nuts because one Monday morning they were told they could no longer underwrite.
And then the 3% overrider was introduced; branch managers were allowed to underwrite 3% of business for the arguments going around this table, and that people can do things better than computers can.
Within a year, Mercantile Credit realised most of the defaults were coming from that 3%. Shortly before they sold it to GE, they realised the computer, in most of the cases, was the best way to go. The intervention of human beings very often brings emotional connection with the dealer or an introductory source that puts a slide on the work and actually introduces more problems.
From my experience, there is an algorithm that can be written for most cases, regardless of where you are in the market. It’s just the terms of that algorithm that need to be altered.
David Gibson: If you look at Private & Commercial, they reject automatically. This enables them to focus on the deals where they can make a difference. Look at automation for service improvement: We talked here about documents being passed backwards and forwards to the broker. Private & Commercial do that all electronically, and attach it all to the deals.
Talk about asset finance: If it’s a peculiar asset, I imagine you’re always going to look at it. The automation may actually be the improvement of capturing the credit data on the customer so people don’t have to do that manually.
Automation is not always used to make automatic credit decisions. Is there a process improvement? Could we be more efficient? Will it enable you to focus on the deals you can affect?
That’s what I’ve learned from the work we’ve done.
Fred Crawley: Peter, you mentioned letting go of the confidence in human intuition to solve most of these underwriting problems. What was the hardest component of your underwriters’ intuition to switch over responsibility?
Peter Minter: It was frightening because we had three underwriters, and I can relate exactly to what Oliver is saying. It was a part of the service we offered.
It helps that automation is a uniform product: You can say what the value of a car is and to set the amount of information you would need on a customer, to make Mark’s point, is not as great as you might think to underwrite a car loan.
The toughest thing was to decide what was important. It all comes down to one thing: affordability. You can play with rates according to the perceived credit risk but it is affordability which determines whether a non-prime customer will be good or bad.
Fred Crawley: Anybody want to contradict that view?
Mark Smith: I would wholeheartedly agree with it. We will decline Grade-A, prime business if we don’t think they can afford it. If people don’t have the ability to pay, they won’t pay.
Peter Minter: With consumers, there is a single source of income and, therefore, a single source of information about their financial status. SMEs or lifestyle businesses are a great deal more complicated for sure.
David Gibson: How did you determine the affordability?
Mark Smith:We ask for proof of income, or income subject to lending, before we underwrite a deal.
Gary Hill: We also look at affordability manually. Although our average deal is probably financing a £4,000 Ford Focus, a customer may owe, for instance, £200 on a credit card and £30 on a mail order, they are less financially active and therefore we look at affordability.
Someone earning £50,000 a year doesn’t necessarily have more disposable income than someone earning £15,000. Some people have mortgages and some have loans, with a wide variety of income requirements for the same loan because of the affordability.
All of which we deal with individually. If we look at a search and think it’s acceptable, we’ll go back to the broker with a tailored income requirement for that individual lend and person, which could be massively different for the same rate for someone else.
Mark Smith: It’s part of responsible lending. If you over-lend to somebody then the loan’s unenforceable.
Fred Crawley: Is anybody looking to upscale and manually underwrite? Where are you going to tighten things up to increase efficiency?
Paul Sheedy: We are looking to increase our share of the market and we are seeing more proposals. The problem with growing and having manual underwriting is that it’s hard to maintain a good consistency of service of the brokers that you are dealing with.
I suppose, being an underwriter, you don’t want to be replaced. Therefore, it’s always going to be something on the backburner. I’m sure there are a percentage of the deals we see that could be underwritten by a computer quite easily. I can see the advantages, especially when you’re growing.
From our point of view, we would probably look to do a mixture at some point, depending on how many proposals we were receiving. We have three underwriters ourselves; I and my colleague do most of the underwriting, with the chairman involved in the larger transactions.
That’s the other thing: Some of the deals we’re underwriting now are up to £500,000. We wouldn’t want a computer to make a decision of that size.
Paul Caunter: All of our underwriters are owners of the business as well, so we’re not using our sales administration or taking the broker or dealers trying to twist our arm to do the deal. If it goes wrong, it affects us in our own pocket.
Fred Crawley: How would you go about enlarging that resource if you ended up with more proposals?
Paul Caunter: The internal accountant managers manage the new business proposals effectively, and are of a standard to put the proposal together and pass to the directors to underwrite if they believe we should be writing the deal.
All of our searches, the CAP valuation and the bank statements are in one place. They will put the deal together to a standard where they can decline a deal if it comes in. They can put it to a position where, yes, we should do the deal, it’s just a case of us performing searches and making sure all the boxes have been ticked.
Fred Crawley: Filtering out the non-starters, then, is probably the place to start with, rather than the auto-approvals: Is that something people would generally agree with? Is it easier to build automation in to the declines?
Paul Sheedy: That would be the first place you’d start.
Peter Minter: It’s probably the first step.
Peter Nolan: I’d still want to take an hour out of my day to review all the declines. It’s so hard for me to let go of them.
It’s having the faith in the technology. You would never have used it before.
Oliver Mackaness: I see what you’re saying, I used to underwrite every single deal that came in. It’s better now that I’m not. I was either too cautious or sometimes I’d think ‘go on, let somebody else have a go’… But, now, having three dedicated people doing it, they’re doing a better job than me because I haven’t got time to.
That was when we were quite small, we broker quite a lot now, and it was a big step to let go of that process but I’m glad we did. The next step would be to automate it even further.
Although we’re manually underwriting, everything around that manual underwriting is streamlined as much as possible. We took a huge step in accepting fax payments or email payouts two months ago but it saves so much time. Those three underwriters write so many more deals now than they could three years ago, or two years ago. Everything else is now as automated as possible.
Gary Hill: You want your underwriters underwriting, not sorting through rubbish proposals. We try to have good relationships with the brokers and although it’s manually underwritten we do have rules set in stone, mortgage arrears being the number one example. If somebody’s not paying the mortgage we will not finance them.
That’s our company policy but there are other proposals which we could decide not to accept under any circumstances.
We try to feed that information back to the brokers. If they’re performing a search on a customer before it comes to us they’ll know if there’s no point sending the proposal, we’re not going to touch it.
If you put a basic, initial test in place, if it cuts out a lot of these instant declines, and something passes it, in theory, the guys should only be looking at proposals they can do something with.
Fred Crawley: How have you increased the efficiency of staff working with the information they are given? What’s the most efficient way in which people affect consumer credit information? What’s changed in the way you manage your department in the past year?
Helen Reynolds: We’ve not made any fundamental changes to the way we manage our manual underwriting function.
Our manual underwriting resource is there to decision those cases that require more information or closer review. This process works well for us at the moment.
Peter Minter: Have you changed any of the automated process? When was the last time you changed your scoring engine?
Helen Reynolds: We constantly review our scoring techniques, our scoring and decision break points, our policy rules, it’s a constant, evolving piece.
Peter Minter: It’s a fairly essential part of automation when you automate a process like this. If you think all you do is replace everything that’s done manually with something that does it automatically, it doesn’t work. You have to make sure you maintain the feedback loops.
It allows you to build better volumes, better relationships and better results.
It’s a tough thing to say but I think it is completely inevitable if you replace your decision system on car finance from manual to automatic your decisions will improve.
Helen Reynolds: A key thing for me is that feedback loop. For us, fundamentally, we rely on feedback from underwriters.
You need the feedback from that human eye to help you identify whether there are areas of your automated decisioning you may want to review to ensure that you are trying to make the best decisions that you can for your customers.
Mark Smith: I’d agree with that. I’ve looked at the dynamics of underwriting manually compared to automation; we should be fully automated to about 90% of the business by the summer this year.
The alternative is: I employ another 50 underwriters and 50 underwriting assistants this year, and the same again next year. A whole warehouse full of people; none of them will make the same decision on the same deal.
It’s just an opinion on manual underwriting: You could not possibly deal with any volume on a manual basis.
Peter Minter: Some things you have to do manually. Affordability is one area which can result in manual work but requires noticing the corrections on credit reference agency records where the customer has said: "I know it looks like I never pay my bills but obviously I don’t normally behave like that." Small brackets: "Actually, I do but I don’t want you to think it." Thus, you have to look at those.
You have to look at any warnings on it, as well. Certain things will get looked at. You’ll never automate it completely.
Gary Hill: Mark, you’re expanding quite quickly and aggressively, so that’s a different view to ours. We want to expand but we can’t expand at the rate Mark is. What may suit us doesn’t suit another company. There’s not a right or wrong answer, it’s dependent on appetite for business as well.
Mark Smith: There are a lot of really successful, manually underwriting companies like Southern Finance who have manually underwritten for 25 years.
Fred Crawley: How has the way you use credit information in underwriting changed over the last year? Is everybody here using Experian?
Peter Minter: We use Equifax.
Andrew Murphy: We use Creditsafe, and Experian for CAIS information, which allows us to make much better decisions.
Mark Smith: I changed supplier, to Callcredit, and found a significant difference between the customer data sets. I looked at all the three of the "major suppliers". Callcredit’s an emerging one. But most of the main lenders send their data to them.
What I’ve found is that Callcredit get a lot more data that the others don’t bother collecting, or don’t have the ability to collect, such as door-step lending, the provident-type loans, the pay-day lending, rent payments, utility payments, white good payments, which people are getting more and more of.
They provide this data set which is useful for us. We lend in the subprime car finance market, if somebody had a default five years ago, it’s unlikely they’ve had any tier-one funding since then but they are likely to have borrowed in other places such as BrightHouse, Provident, Wonga, etc…
You can see the credit profiles. If somebody’s been paying a doorstep loan for the last five years and had 10 different doorstep loans, they have a proven ability to pay credit, whereas another referencing agency might suggest they’ve had nothing, which is clearly not the case. It’s a much better data set for us.
Fred Crawley: Is there any data anybody wishes they were getting but can’t?
Peter Minter: Income data would be really useful. I’m not sure how we’d collect it but in the US the availability of income data is much freer.
Fred Crawley: That’s not available from any source here?
Peter Minter: Well, it is claimed to be. You really can’t base any decision on what’s available. We’ve just started looking at Call Credit. They seem to be more proactive.
We moved from Experian to Equifax and it struck us as extraordinary that they are all working, except for some of the things you’re talking about, from the same sorts of data and have fundamentally different records against each person. It’s hard to imagine how that can be the case.
Oliver Mackaness: They should be sharing the same information.
Peter Minter: They do share information but it will still come up fundamentally different. We were turning proposals down based on CAIS data from Experian and we’d be sent an Equifax report on the same person showing fundamentally different results.
Mark Smith: We broker a small amount of business to Billing Finance and sometimes we have completely different data searches on the same customers; same address and conflicting information.
David Gibson: We’ve just implemented something that we centred on finance. We couldn’t reconcile the test environment with the live environment.
The testing was complicated by Experian configuration parameters and the fact we were using a new Experian infrastructure.
