Graham Whitley is turning scores into revenue

Last week, we set the scene, and in doing so we allowed ourselves to keep the conversation high-level and somewhat academic in nature. In the corporate world, we’re seldom afforded such a luxury, however.

Instead, if questions of lending strategy are to be given their due attention from the Board, they need to be shown to be delivering quantifiable value. In episode two of How to Lend Money to Strangers, I address this with Graham Whitley of Quid Pro Consulting.

Graham has been building scorecards and scorecard-driven lending strategies for over twenty years in developed and developing economies. Crucially, though, he is also willing to stand behind the revenue-generating power of those strategies by linking his consulting fees to measured improvements.

I speak to him about what he looks for when he builds a new lending strategy, how scorecards and business strategy interact, how he adapts when working in less than ideal data environments, and because it was so important to our shared experience, we also spend some time speaking about how to effectively implement and leverage champion/ challenger strategies.

If you have any feedback, questions, or if you would like to participate in the show, please feel free to reach out to me via the contact page on this site.

Regards,

Brendan

You can read the full transcript with timestamps here:

Graham Whitley 0:00

Some engagements that I'm busy with at the moment. I'm so convinced that, not a brilliant, but just a good scorecard can bring value to business that I'll engage with the client and the payment of the contract will only come through if certain uplifts in revenue are achieved.

Brendan Le Grange 0:46

Welcome back to How to Lend Money to Strangers the podcast about consumer credit strategies around the world. In my first episode, I spoke to Raymond Anderson about the long history of risk assessment, and the more recent history of credit risk scorecards. But scorecards, like everything else in life, must pay their own way. So today, I welcome friend of the show, 20 year veteran of scorecard building, and Chief Risk Officer for hire Graham Whitley of Quid Pro Consulting - we're talking about the sometimes messy process of turning scorecards into real and measurable business benefits, and the role test and learn strategies play in this.

You've spent a lot of time building scorecards and building strategies, but you've also spent a lot of time thinking about how to turn those into business value. That's really what I want to talk about today but let's start with a quick introduction of your background in credit, risk analytics, what you've done, your experience, and then a little bit on the type of work that you're doing now.

Graham Whitley 1:52

I graduated in actuarial science, and then I joined the CapitalOne/ Nedbank Alliance in South Africa. And with them, I moved to Nottingham in the UK where I worked for just under two years in the CapitalOne credit card business. When I returned to South Africa, I resumed my role mainly focused on acquisition.

And then I saw an opportunity and I consulted independently. And I haven't really looked back to be honest. And that was about 16 years ago now. From acquisition, credit policies, scorecards, profitability models, business case derivation, customer lifecycle management, line management, a lot of stuff in collections, and across all the different retail products. There's a lot of data work, because a lot of clients don't really have the experience or the in depth knowledge of their data, so there's a lot of value in understanding the data and how that drives the business, because that's how I think you can add the most value: what phases of the business does your scorecard sit in? And how good are the things like your profitability models and your implementation strategies, and those kinds of things. I land up spending a lot of my time evaluating how to generate the maximum amount of value for the business from a scorecard?

Brendan Le Grange 3:06

In my introduction, I referred to you as a Chief Risk Officer for hire because the difference between you and a pure scorecard consultant is that aspect of it, that although you - now giving away our ages a bit - but, you know, 20 years of experience in building and using scorecards, your focus is not on what is the best way to get the math right to maybe gain a tiny bit more of gini, it's actually what is the better gini? Where is there a compromise that is needed to be made, where we might want to take away a little bit of predictive power in exchange for being cheaper to get the data or in exchange for improving a ratio that the rest of the business might need?

Graham Whitley 3:46

It's quite funny say, as you say, like a Chief Risk Officer role for for hire, which is a role that I do take, but a large part of it is companies not really understanding the need themselves. So they know that a scorecard is important. They don't really know what a scorecard does, you've got to actually inform them or, I don't want to use the word educate but it is right in this context, you would educate them that scorecard is just a tool. So I'd spend more time getting people to understand how the scorecard is used and finding ways that the scorecard can add value and identify different customers, rather than focusing on the scorecard itself.

It is also quite funny, because scorecard building is seen as a science, you've generally got a tool like SAS, which helps you create a scorecard, and it's often seen as a black box activity. People give the information to an analyst, they come back and they present a scorecard. The guys use the scorecard and they walk away. And they have the opinion that no one inside the business can create a scorecard. But the business also sits there and goes 'we can use the scorecard and we can create strategies around the scorecard because it seems easier and it's not a black box, etc.

Creating a scorecard, once you understand it, is actually pretty straightforward. And you don't need to know the math behind the scorecard, SAS does it for you. So we don't want to get into building the scorecards here but the scorecard isn't as complicated as it seems to be. But what I think does need the expertise is the implementation of the scorecard. And, and the way that you can take a scorecard for what it is and the power that it gives you, and really drive through maximising the value within the business. And often they they land up with strategies that are suboptimal because they haven't engaged enough in in terms of how to use the scorecard and the power that the scorecard will drive, not only in just a singular decision of approve and decline, but how you can use a scorecard to inform your strategies and how you can implement things like risk based pricing off of it, and how your risk scorecard can help you identify response rates.

So, you've got two different scorecards that operate right at the front of your business, which is around customer acquisition, the one would tell you what customers to target, there are more likely to take up a lending product and the other one will tell you who should you approve - those scorecards generally work in opposite direction, because customer that is going to be more likely to respond to a credit offer is often going to be a customer that could be desperate for credit, and that generally aligns with higher risk customers. So as an example, you could be in a situation where your response model says, out of a population of 10,000 customers, your response model says, target maybe these 1,000 customers because they are more likely to respond. However, they do respond, you might find that out of those customers you land with 100 approved loans. So you've got to try and find a compromise between the two scorecards.

So you actually, you almost create a matrix of the two scorecards. And then what you do is you target the segment of the customers that maximises profit.

Brendan Le Grange 7:14

Yeah, because that should be informing the approach that you go out with, the messaging that you're using to market would be very different. If you are knowingly targeting the population, knowing that many of them will be declined, you're gonna have to have a different messaging that makes that clear to protect the brand. You don't want to go out sounding all confident, and then decline a whole lot of people. Or likewise, you mentioned risk based pricing. That's the kind of bridge between those two where, say, these are the people I want, we're going to approve all of these that we can get. But we know that this is a population where everybody wants them. So there's a lot of competition, I better give a discount. Or maybe I've got to give a sign-on gift. This is a population that doesn't need credit so the messaging, maybe it's something more luxury based or more emphasising the rewards your credit card offers. Maybe it's been done by marketing, maybe it's been done by a team entirely divorced from the credit side, nevermind the analytics of the scorecard. But the thinking that the scorecard builder put in there is relevant and should be made available to these teams.

I did some work with a client who had a big, impressive, expensive scorecard upfront. And the guys that we were working with, there was a little bit of grumbling under the breath, because once this big, expensive scorecard had gone in, they were still not really seeing a great uplift in the quality of customers that we're bringing on board. Because I was working at a credit bureau at the time, I could look at 'what's the credit score of the people that you did an inquiry on, and what's the credit score of the people you approved'. And I could see that the degree of separation from applied to approved was amongst the highest in the industry. So you could actually say that scorecard was doing his job very well. But when we looked at those same data fields it was very clear that the quality of the through-the-door population, the quality who were applying at them was really, really low.

So we're saying you've spent a lot of money on a scorecard, but actually a lot of your problem stems from the people you're attracting to apply. So yes, the scorecard's good, you probably need a new scorecard, it's probably worth the money you've spent however, you haven't addressed a big part of the problem: and that's your marketing. So perhaps it's your branding, why are so many high-risk people applying to you? Is there something about your brand that's turning off lower-risk customers, and it's that big picture that, seen together, which would have been a lot more effective use of money than just building a new scorecard with a better gini.

Graham Whitley 9:39

I agree with that 100%. And I think that also comes back to the point that we made earlier, which was around the tools that support and surround your scorecard. If you can identify good customers that you want to target and bring on board you're generally going to have to convince that customer, you can have to give them some kind of offer. And the danger there is I often find myself in a situation where you're having a meeting and people say, OK we've got to make it more attractive for these customers to take a product. So let's offer them a 2% or 3% rebate on the interest rate. And they really get excited about that, and off you go. But that could then make that product unprofitable for those customers, despite being low risk.

And this is where the understanding of your performance and your profitability is actually vital. What can you do to attract a customer, if you pre approve the customer, what, you know, what is the interest rate that they that they should be on, or if you give an interest rate discount, or a fee discount, all of those kinds of things become very, very relevant. Likewise, in that example, that we spoke through earlier around the 1,000 high risk customers that come aboard. On the face of it, as we described it earlier didn't sound like it was a good decision for the for the bank. But if you could say to those customers, we'll give you a product, but bearing in mind, your interest rate is going to be 5% higher than our standard offer and that switches it around into a profitable nature, then that's good for the bank. And that that's the risk based pricing truly comes in, you push in that decision and that profitability earlier in in the lifecycle of the account.

Just the one thing on the scorecard that people often overlook is, people think that the remit of a scorecard is 'how you approved customer'. That's not always the case. A scorecard also helps you identify areas where activity shouldn't be conducted. So, as an example, a collection scorecard is is often overlooked and de-prioritised. It's something someone does off the side of the desk. But a collection scorecard can add huge value by identifying customers that don't need to be contacted, customers that are going to pay without intervention. So good customers. And then you reduce your required capacity. And you focus on the sweet spot, which is the customers that have a chance of paying, if you contact them. Then, as I said, a collection scorecard is actually one of the easiest scorecards to be brought in. I mean, you can probably build it in Excel. And, I mean, I've seen impacts on collections where collections have jumped by 30% to 40% because you better focusing your activity on the customers who really need it.

Brendan Le Grange 12:28

Yeah, because you've got the, you know, the straight operational savings but also then the ability to better work the customers where there's a chance of recovery.

Graham Whitley 12:37

But the strategy and getting the buy-in, it's not an easy sell within a business, often.

Brendan Le Grange 12:43

Yeah, the collections topic as well as the pricing one we mentioned earlier are both conceptually fairly easy, but practically hard to measure unless we are doing champion challenge or test and learn. I was going to pick this up later, but I think it is probably worth diving into now. If you are going to implement risk based pricing, as you said, simply saying 'we'll give a 2% discount' is very unlikely to be the way to do it, we're going to want to make sure that the discount we're giving is essentially as little as possible, but still big enough to change your decision. And we're also going to think through things like what is the best format - when I was in the credit card space, you wouldn't really want to talk about interest rates reductions because, (1) frankly, they're so high, you often don't want to tell a customer 'hey, you'll only paying 25%' when they didn't realise that we're paying 27% in the past amd (2) practically speaking, for a lot of people, particularly in the low risk space, the interest rate on the credit card is not something they think about because they intend to fully pay their balances each month. So maybe the discount is a sign-on bonus. So we have to find what's the optimal discount, the optimal form of a discount, or in collections, does not phoning this person change their behaviour, because I've identified them as low risk but if I let them run too long, do they become high risk? Or simply to prove it to give confidence to a manager? You might say, Okay, this is what the numbers show, but we're going to do it in a small way to control the risk. All of that can only really be done properly with champion challenger approaches.

So can you maybe talk a little bit about how you use champion challenger approaches, either to prove to an executive, to a client is maybe a little bit unsure whether they trust the numbers enough to do something bold, or whether it is to measure and optimise a strategy?

Graham Whitley 14:23

I mean, everything that I look to implement or any strategy that I look to roll out has to be done through a champion challenger approach, because there's just so many moving elements at any point in time that you have to understand the impact that your strategy's having.

Having said that, there are challenges that come up. It's difficult to have two scorecards implemented at the same time from a decisioning point of view. And the other one, it lands up in a funny situation because at some points in time you spend time convincing the executives that the new strategy is the best strategy. And then you finish up the conversation by saying 'but we're only going to put 70% of their accounts through the new strategy and 30% are going to go through the old strategy, which we believe is substandard'. In growing businesses or businesses that are new to scorecards and things like that, they're not a massive fan of that. They, they want to go big, and they want to drive the value as quickly as possible. It's definitely a discipline that we have to get into. And we have to force ourselves to stick with because, you know, you have to operate in the realms of the known and not in the realms of what we think this happened. And that's why you absolutely have to have the champion challenger strategy, as difficult as it may be. And even sometimes do things manually, make sure that test and learn becomes a staple and a integrated part of the business,

Brendan Le Grange 15:48

I get that. I worked for a big established multinational bank, who had champion challenger in the credit card issuing systems. But the two parts were: 80% went down strategy A, 20% went down strategy B, as they had done for the last two years. That's not champion challenger, you can't just do two strategies for two years. Yes, we could compare one or two things. But it's test and learn, champion/ challenger, it's about always updating the new strategy, it might be great, it might be best. But slowly over time, people change their behaviour, because they're getting declined, or because they're getting approved, because their limits are being increased. If we're not continuously measuring that and getting ready to change and adapt to what's happening in the current day, we're always going to find ourselves slowly moving off track.

And even if you're not big enough, or sophisticated enough to implement multiple marketing campaigns at once and monitor for the exact perfect price, or the best way to word an offer in a letter, even if you've just got that simple champion challenger: okay, here's the system, we're replacing the score, the strategy we're replacing, versus a new one. And we're going to check in after a month, two months, three months, and we're going to prove that the new one works, that 100% of people, in theory could go to a new one. But don't we have a better idea now, maybe we want to try something else, let's just put in a new challenger, and to create that atmosphere, or that culture of continuous learning, of not sitting on the status quo. You can't do that with full confidence unless you real life testing. Because everything we measure, we change. So you're changing people's behaviours. And unless you accounting for that, they're just going to be having to get the consultants back in a year to adapt the strategy.

Graham Whitley 17:31

Test and learn isn't something that you do once to tick a box, test and learn is a culture of evolution and improvement and continual development, to come out and deliver the best possible and potential strategy. And it's something that you've got to do all the time. And you've got to be continually asking the question and pushing and trying to find those improvements. What I do find really helps is, when you're putting in a new strategy based on a scorecard, or a new scorecard, is make sure that the success criteria are well documented. So you're looking for an improvement in a key metric, let's say customers who've missed three payments after six months, let's say that's a critical metric, and your conditions of success are a 10% improvement in that metric. That's an agreed outcome whereby if that condition is met, actions are taken, because often find it and especially with the smaller companies, data is sporadic data is spiky, all of those kinds of challenges exist. But if you sit up front and you say, right, when we see this, then this is the action that we're going to take. That's very powerful. And likewise, the conditions of failure, because everybody wants the new strategy to be a success but things don't always work out as we planned.

Brendan Le Grange 18:49

Yeah, I think what's interesting there, as well as the test and learn does create its own data. So you might be operating in a world where data is hard to acquire right at the moment but by doing test to learn, you can create it quickly if you've thought through what you're actually testing and what those success and failure criteria are.

In one of the other episodes I'm talking to a FinTech in the Philippines, and one of the things they're working with is a client base that doesn't have a lot of traditional credit data, and so they do lots of champion challenger, lots of test and learn campaigns very quickly. They'll take a small risk on a customer, and then try and learn as much as they can about the customer early on. Because they see that as a way to create the data they need, where it doesn't exist.

Graham Whitley 19:31

The low-and-grow is a strategy that works well when there's there's little or no data and you're effectively pushing back your scorecard from an upfront decision to a three month or six months decision.

Brendan Le Grange 19:45

I've heard the term low-and-grow used where, yeah, we give everybody a small credit limit and then at six months, we see if they been paying everything back and if so we increase the limit. But if it's done right, you're not so interested in learning a lot more about the specific client, you're trying to find ways that that client can teach you about everybody. So you want a thought-out campaign, so that you're improving your scorecards over time. It's building those feedback loops. So which customers which type of customers made it through to six months and look good and got a credit limit increased? What might I know about them? Now? Could I rebuild my scorecard if I did it today, because particularly in this example I'm referring to in the Philippines, they're bringing on a lot of new data fields, you know, or the sort of smartphone data fields that exist today that haven't got a long history of being modelled in, they can keep all this data 1000s and 1000s of fields. And as they get more customers on board, as they get more performance data, they're continuously thinking through have we learned anything that's enough to build the scorecard, to improve the scorecard. And they will know that so much faster because they're checking all the time, than if they just waited for a two year development cycle or tracked gini until two unique or too low. And they did a an audit pushed us scorecard build.

So yeah, champion challenger, we talk about it, and it's something that most people know. But it's that culture, as you said, a culture of learning. When you boil it, what are we testing? What are we looking for? What would success mean? What would failure mean? What time frame would we need to look at those, so depending on what you're looking at response rates, you know, those happen from within a day or two; defaults, you know, take several months, it does take time, it's naturally it does sound inefficient if someone might be very keen just to get everybody on the new scorecard, on the new strategy, but if we can build that culture it allows us to learn more accurately

Graham Whitley 21:32

Yeah, I think you've touched on two really important aspects there. The one is just the time to learn is not something to be understated or underestimated, especially your risk performance side of things, it does take time to get significant learnings and to make proper decisions. And I think the other thing, when you're talking about the low-and-grow type of strategy is: good decisions need data. And throughout the entire process of everything that we do, is having good data. And that flows through from making sure that the data is accurate on the systems, to have an extra small systems, but also to making sure that you capture as much data as you can, whenever you can. So my default position is always to try and... just give me as much data as is potentially possible to gather and then let me sorted out later on, and figure out what's important and what's not. So what you're talking about the 1000s of columns of data in that's brilliant. A lot of companies that you engage with don't fully understand the value of capturing data upfront. I can't tell you what data is going to be important. I'm only going to know the answer to that a year from now. And then then we can trim down the application form or the data gathering exercise. That's how you build a good scorecard.

There is a lot of pushback against that. Everything knows about making things quicker and better and faster. But the worst situation that you could be in is having implemented a low-and-grow strategy. And the year down the line, you're sitting there going, actually I don't have enough information or don't have the right information that helps me identify who's a good customer or not. And we need another year of information and performance. And let's do it right, this time. So I'd rather be in a situation of maybe losing a couple of customers through the application process, than delaying our development by a year, if that makes sense.

Brendan Le Grange 23:22

Yeah, it does. It's also why I've got interested in the sort of new alternative data fields, in developing markets in particular. In big, developed markets, there's normally a credit bureau score you can refer to and you can get the data you need on the customer quite easily. In the developing world, you'll often have your none of this legacy of data built up, there's often not really a strong credit bureau, or where there is a bureau, only a very small portion of the population is represented in there. So in the Philippines, where that example was from, but the same is true in places like Kenya, there's a strong credit bureau, sometimes multiple credit bureaus, but they're talking about a very small high income portion of the population. But when we look at now, what smartphones in particular can bring us, is these smartphones are storing all this data anyway. Things like your home address can be inferred from the data, the activities that's happening on the phone, how many relationships exist, payment status and things, if we can start pulling these in from other sources that are storing it, we can sometimes work around that fact that the lender might not have institutional capacity to build and manage databases, but maybe somebody else has got it. And we're entering a world where you don't have to have a handwritten application form that's seven pages long, the smartphone can maybe pull that through with a few bits of consent given

Graham Whitley 24:44

That's the key part, just around the concent. So with my clients in Kenya, I mean, there's a huge amount of focus in terms of how social data and open data can help slope risk. It does have huge potential but then you do have regulatory concerns and privacy concerns that come along with it.

Brendan Le Grange 25:05

The risk that it gets turned off by the gatherer, at some point is all controlled by you. So it's got value in some fields, but there's some complications to get through. I know we sort of gone off a bit on the champion/challenger, test and learn, but if we circle back to when you're linking a strategy to revenue, what is your approach to make sure that the work you're doing is driving genuine, measurable and sustainable business results. And not just giving a scorecard with a very high gini?

Graham Whitley 25:41

As an example, I was assisting one of our clients in Africa, and my engagement at that time had nothing to do with the upfront credit policy. But I did in conversations with the CEO understand that they'd invested quite a bit into developing a scorecard, an application stage scorecard, but weren't fully understanding the benefits that the scorecard could bring to them. And there was a lot of push back, they were second guessing themselves and doing a lot of manual reviews of the scorecards decisions. This was problematic because a they were they had a lot of capacity that was being used in making subjective decisions to slow down the application time, all of these negative things.

And at the same time, from what I could see, my analysis was that they were declining potentially good accounts. So I spoke to them and engaged with them, said I can help you out here, I'm willing to put myself in a situation where the payment of the contract will only come through if certain uplifts in revenue are achieved. So the first thing was to actually understand the account performance and link that through to a profitability model that people understood. And then to transition that into the scorecard and link it through and say, well, now that now we've got the scorecard and these are the profits for different scores within the scorecard. So if we were in a in a world where the manual vetting department was offline for a couple of days, and we moved fully onto this new scorecard, this is the value add incremental value that will be generated.

And we worked it out to be about 15% of the bottom line of the business, from moving to an automated scorecard without manual vetting. So it was a huge thing for the business and that itself generated a lot of excitement, you don't often come across a time and a place when you can add 15% to the bottom line with business through, you know, one single implementation of a scorecard or strategy. A number of engagements after that had taken on a view of saying, I'm so confident that we can create value, that payment of the engagement is conditional on value being achieved.

Brendan Le Grange 27:49

I like that idea of instead of upfront, we can agree that we're going to look at the same things. It gives everyone a reason to track the success. You mentioned a profitability model, in my mind that can range from a few lines in Excel to something as complicated as a scorecard. Can you just talk a little bit about what a profitability model was in this context, and maybe some broader thoughts on on modelling profitability?

Graham Whitley 28:15

I think, for me, it's got to be understood, so can't be too complex, it's got to be accurate. And there has to be a way for the business to tie back or understand the performance, you can't just have a a low-level profitability model in isolation that no one's got any reference to. So I build an account-level profitability model to go through and look at the performance of one average account that represents the portfolio and then you multiplyup to the portfolio size at a tranche level. And then that should reflect the portfolio as it is now. And if you can do that, you can build a lot of buy-in to the fact that the profitability model itself is accurate, at a loan level.

And then you can start modelling out different scenarios, what happens if the interest rate is 10%, higher, what happens if risk is 15% lower. And you can do this on an average account level to create a very specific NPV on which specific decisions can be made. So if you can sit down you can say bureau score 600 translates to a risk experience of x over time, 5% of the accounts default in months six. So that really gives you the ability to make very, very powerful decisions within the business and lots of gaming scenarios.Like how do we model out low-and-grow? How do we model out changes in interest rate? How do we model out bringing on board a fee? How do we model out changes in risk experience? What happens if we believe that in a year from now we're going to have a better collections department and better collections strategies? How's that going to impact the present value? All of those things become relevant. It doesn't have to be this massively complex thing. It's just got to be... people have to be able to to look at it and understand it, and it's got to be flexible, flexible and accurate. And then then it's probably the most powerful tool that you've got within the business profitability model and the scorecard are, in my mind, the two most important tools within a business.

Brendan Le Grange 30:14

Yeah, I think I could agree with that, because I think that a lot of the value of the profitability model, it's not an easy word to say, the profitability model, is from its ability to drive conversations and to get people thinking about 'how does this connect to that?' Where's the money coming from? Where we're losing money?

The scorecard is very much 'this is the risk situation', 'how many people are going to meet the bad definition or the target variable'? But then, what are we going to do about that? How do we use that? What are the next steps? What's all this complicated business stuff that sits on top of it? What are interest rates, pricing, approval rates, what are your operational costs, all of that can be folded in and it can build up over time. I always like them as an exercise to get everybody to see that one picture. And I built some models for clients that are in an Excel sheet with a few lookups and and matrices and things built in. And in those cases, it was to drive just that conversation. I remember doing a client engagement in Sweden, where I built the model on the train ride over in a couple of hours and it wasn't okay, this is your actual profitability yet, but for that ability to sit down and say this is what it could mean,

Graham Whitley 31:20

Ja, correct.

Brendan Le Grange 31:21

Thank you, Graham. As always, it's been a pleasure catching up. I look forward to having you back on the show another day.

Geog Steiger is the co-founder and CEO of First Digital Finance, a FinTech in the Philippines that's using advanced AI modelling and bu- now-pay-later product offerings to increase access to finance in that country. I'll be speaking to him next Thursday about how to lend to thin file or new to credit customers in developing markets. You can join us on Spotify, Apple podcast player, or wherever you're listening to this.

Graham Whitley 32:31

Look, to be honest with you, I'm not too stressed about getting the consultants back in because that's what keeps me employed. But your fundamental point is correct.