"In lending, you either have a growth problem or a fraud problem... "

Anytime you start a business or a company, you face many foundational questions. Of those, the most common is the “how-do-I-grow-fast-and-not-blow-up”. Depending on the type of business, the trade-offs can be significant and the consequences of making the wrong decisions can be existential. Let’s start with a hypothetical example. Assume you have an amazing product - say an AI driven chatbot that will essentially replace 75% of customer service calls (sounds very relevant these days ?? ). You have a prototype that works well against some common use-cases.You have done your customary roadshows, lined up your investors and even bagged a couple of POCs - you are off to the races. From that point on, it’s ‘grow baby grow’! It does not matter that initially every new contract you sign and every new customer you acquire likely has negative unit economics and that you are burning through cash like a drunk sailor in a Vegas casino! But who cares - as long as you can grow AND you have proven that you have a viable business, you have built a massive first mover advantage. The prevailing notion is that eventually the UE will work out. As you gain operating efficiency and scale, your CAC will go down and you will trim the waste around the margins and with a first mover advantage, you will eventually gain pricing arbitrage. This is your typical grow-first-make-profit-later model and it works for most consumer products.....or at least it used to till a few months back!?

Unfortunately though, this model never worked for lending. This is because in lending the losses are asymmetrical. In most non-lending businesses, the gains and losses are bounded and are often a fixed function of the volume i.e. you know exactly how much money you will lose for every unit you sell. In the above example, for every widget you sell, you lose X and only X and as long as your management and your investors are conformable with those losses, you can continue to grow.?However, in lending a bad actor will often take you out for the entirety of the loan amount - often leading to a very asymmetrical gain to loss ratio - often?20: 1 and sometimes as high high as 30:1 depending upon the product and segment. In other words just a few ‘bad accounts’ can literally wipe out the entire business. That is exactly the reason why the grow-at-any-cost model does not work for lending.?As they say “all lending businesses either have a have a growth problem or a fraud problem”??

With that context and background, let’s dive into the mechanics of launching and scaling a lending business. If you haven’t read part 1, I encourage you to read it here first to get a sense of some of the key decisions you have to make to set the stage for launching a lending business. Your next set of actions over the course of say 12-24 months are equally critical in building a sustainable, scalable and eventually profitable business.?

1. Go to market strategy - As a ‘start-up’ you and your investors likely live and die by how quickly you can grow your business. All things being equal, the steeper the growth curve, the better your business is doing. For most businesses, that would be the right approach. You grow the business, prove you have strong product market fit and then you optimize on the margins and improve the unit economics. Except that in lending this model does not work because of the asymmetrical downside explained earlier. Also, lending is one of the few businesses where the absolute worst customers initially look like your absolute best customers - they are engaged, they are spending, they are using your product, doing everything that you ideally want them to do -?till it comes time for them to make their first payment :-) Some will even make a few good payments initially and use your loan as a platform to build up their credit profile, apply for bigger loans with you and others before going ‘bust’ on everyone. As a lender, you are absolutely helpless when these seemingly amazing customers bust out, especially for unsecured loans like a personal loan or a credit card. This is also compounded by the fact that because these customers look so great initially, there is a natural tendency to double down on acquiring more of the same kind leading to often disastrous consequences down the road.?So what should you do? Well, the last thing you want to do is to launch with an open-market strategy on day 1. If you are new to lending or if you are a new player within a specific lending product or segment you have to understand that you do not have an established brand. Legitimate customers likely have not heard of you or your product to seek you out on day 1. Conversely, the people you will have absolutely no problem attracting on day 1 are the fraudsters and credit abusers ?? It’s almost like the fraudsters have their antennas up and are always scanning for the next bank or fintech to launch a new product. They all come out of the woodworks to attack you in droves as soon as you launch. Now, to be clear every business, including the big players have their share of fraudsters. However, because you are a new product/brand and you are not able to attract enough ‘good’ customers to offset the losses generated from the ‘bad’ ones AND your risk models (underwriting and fraud) are still nascent and learning, they fail to block many of these bad actors at the door (more?on this later). And because of this, your mix of good to bad customers gets completely out of whack. As a new lender, your GTM strategy should heavily involve strategies and marketing channels that allow you to pre-screen and pre-approve your audience. The flip side is that these targeting strategies and channels are expensive, time consuming and tend to be low volume at first. But given that it safeguards the business from ‘blowing-up’, it’s absolutely worth the effort and time. The second thing it does is it gives you some cushion to iterate and improve your risk models. As you collect the data and refine your models plus build your risk tools and infrastructure in parallel, you can gradually start expanding into marginal segments using a test and learn approach and grow the business.?
2. Targeting - So what IS effective targeting and how do you get better at it? You may be wondering - can we not just ‘block’?the bad actors at the door using our fraud and credit models vs investing time and effort to prescreen - after all we are in the digital age and we are a high growth company? I'd like to think of it this way -?say you want to throw a nice party at your house - fancy drinks, nice hors d'oeuvres and food, great music etc - you have it all. What do you think will happen if you suddenly decide that you want to invite everyone to the party - so you post the details of your party on your twitter/instagram/TikTok page for the whole world to see. Now, you do have a strict dress code and you have hired a few burly bouncers and positioned them at the door to enforce that strict dress code, check people’s id etc, heck,?you even make them solve math riddles at the door to gain entry ?? - thinking that all these steps will keep away the bad actors and ensure that only the ‘worthy’ ones gain entry. So what do you think will ensue - my bet is sheer pandemonium!! How difficult do you think it will be for people to fake their way in - wear seemingly fancy outfits, have fake ids, solve the riddles (with help from ChatGPT of course!) and eventually overwhelm your bouncers at the door with sheer volume - and once they are in, mayhem follows!?That’s exactly what happens when you ‘open up' your lending business to all on day 1. In reality when you do throw a party at your house, you carefully curate the invitee list based on multiple factors and only send out the invite to the people who you think will help make a successful party - and that’s how it should work in lending as well (at least initially). The exact mechanism of targeting will vary by product and the maturity level of your business. If you are a brand new business, you will likely start with a 3rd party prescreen vendor and give them your broad targeting criteria. The criteria is often a combination of a target segment, income and credit profile - e.g. say you want to go after near prime, with income above $75K and a credit score between 600-660 with no prior delinquency or defaults in the last 3 months. This is obviously massively simplified, but you get the idea. You then run your underwriting model on the selected population to further narrow down the target list. Once you pull a prospect’s bureau information and they meet your criteria (by law you have to give them a firm offer of credit), you will typically just send them a direct mail (yes, even in this digital age, direct mail is still one of the best channels to target even though there are other viable digital options coming up). Establishing this entire infrastructure takes time and effort but I cannot emphasize more how critical it is to have this in place before you go-to-market.?Obviously, as your business matures and you start amassing more data, you will continue to refine your targeting models by building complementary models like a response model, usage model and eventually a valuation model. Ultimately, the prospects you want to target are not just the one with the lowest risk, but the ones with the right balance of risk, responsiveness and usage - so that ultimately these accounts you book are all profitable (more on it later).
3. Risk models - There are four elements to managing your credit and fraud risk 1) the data you use for your models and how accurate and up-to-date they are 2) the model itself - moving from classic logistic regression based models to ML based models 3) the policy that leverages the output of the model and decides what to do with it 4) and the decision engine that runs all of this and spits out a decision. Having the right risk models is critical to having a viable lending business but it also happens to be the most difficult to get right immediately. Just to recap from my last post - as a new lender you do not have any historical performance data on your own book. No matter how sophisticated your modeling techniques are, at the end of the day a model is only as good as the data that it is trained on. I have heard people claim that they can build models on anonymized historical data. The fact of the matter is no matter how much historical data you collect, it will always be biased.. because it is built on someone else's performance data. There is no substitute for building models on your own performance data that factor in the strength and weakness of your current product and brand, the type of selection dynamics and the impact of the actual macro/micro environment you are operating in. So when you launch a new lending product, you just have to accept the fact that on day one, your risk models will be sub-optimal - be it your underwriting, fraud, pricing etc. I won’t go into the details of how you build your day-1 model in this post - but as I mentioned before, you will likely start with some anonymized data set for segments that closely mimic your target segments (most bureaus will give you access to that for a fee), use a conservative policy on the model output, layer some heuristic based hard cuts and then launch. If you are unable/unwilling to build your models, you could also use one of the many risk-models-in-a-box solutions available in the market and get to market quickly that way and then continuously iterate. An important point to note is regardless of which approach you adopt to launch, ultimately you (and your partner bank) will be on hook to ensure that your models are complying with the various regulations, particularly Reg-B or Fair Lending. Here is a post from Francisco Javier Arceo I came across recently that does an excellent job talking about how underwriting and risk models and decision engines work........Ok, so if you know that your models are suboptimal out of the gate, what do you do? First, you need to make sure that you enter with a test and learn mindset - i.e. design a bunch of small iterative tests around product value prop, marketing, underwriting and fraud policies and second build a robust mechanism consisting of reports and dashboards to monitor the performance as close to real time as possible. When you launch a new product, despite your best efforts, know that you will get a lot of fraudsters and bad actors. Your P&L should account for that and the business leader should set that expectation with the management and investors. You then focus on minimizing the damage by identifying and plugging the gaps iteratively. As a new entrant, I understand that your mandate is always to grow the business as fast as you can. But don’t give in to that temptation of doubling down on any particular strategy or segment prematurely based on early reads. Unlike other businesses, it is almost IMPOSSIBLE to undo bad lending decisions. You should wait at least 3-4 months to get reads on delinquency rates, payment behavioral data and other signals from the bureaus on the customers from a specific segment/strategy before you declare success and decide to expand. And as you collect more data from the test, you feed that back into your marketing and risk models and continuously refine them. In the first year of launching a new lending product, it is not uncommon to have risk and targeting models getting updated almost every 2-3 months.?So I am sure, at this point you must be wondering - if we follow this playbook, how the heck are we supposed to grow ?? In lending you need patience and discipline. Your first few years are all about finding the right segment and building the right capabilities. But once you build a strong foundation, growth and expansion will follow.
4. Valuation model?-?“Do you want to acquire prospect X and if so, how much are you willing to spend to acquire”. This is one of the most fundamental questions in lending. We’ve already discussed that when you launch a lending business, the biggest challenge you face is lack of data. Your situation is then further complicated by a near constant pressure to grow the business. So how do you make sure that the accounts that you are bringing in are actually the accounts you like and what is exactly a ‘right’ account? Depending on the lending product it’s usually a combination of 1) people who are willing to respond to your offer 2) people who are low risk (i.e. both willing and able to pay you back) 3) people who are willing to use your product on a consistent basis (for transaction products like Credit Card) and 4) people with low cost to serve. Quantitatively, this boils down to what we call in our industry the ‘valuation’ of the account - i.e. what is the future cash-flow from the account over a certain time horizon. You will likely also layer in different macroeconomic conditions to see how the cash flow changes in different scenarios, particularly in more stress scenarios. Ultimately, the idea is that you want to book accounts that are not only profitable in good economic conditions but could also be ‘resilient’ during stress scenarios. Based on your risk appetite, you will?narrow down your acquisition to the accounts that will likely be cash-flow positive over the chosen time horizon across different economic conditions. The next step then is to put together an acquisition strategy that will allow you to acquire as many of those accounts as possible in the most cost efficient way. The incumbents have a huge advantage here because they have very sophisticated valuation models - often to the individual account level.?Newer entrants barely have any data - so you will likely have to go to market with some heuristics and assumptions. Often I have seen businesses pick one of these dimensions and fixate on it somewhat disproportionately. If you focus on growth alone, you will likely get burnt by fraudsters - we discussed that before! If you focus only on low risk accounts, you will likely suffer from low response and/or low usage - because by definition, low risk accounts tend to have low need for credit. This is where a valuation model comes in. Early on it will be very challenging and will be as much an art as science. You will have very limited data from internal performance, so you will have to rely on some early indicators and extrapolate from there on. You will also have to rely on data from the bureaus to understand the typical performance of similar cohorts you booked and marry that with the actual performance of your accounts to get the longer-term performance curves which will help you build 3-8 years of cash flow. You will likely do this for very broad segments and key channels first and then continuously refine with each additional tranche of data to get more granular. I have seen many businesses put aside building a valuation model to focus on growth alone, only to realize down the road that the accounts they booked were basically junk!?
5. Portfolio risk management - While the concept of portfolio management is important for all lending products, it is critical for open-ended revolving loans like line of credit and credit cards. Any acquisition decision is probabilistic - you are relying on the data that you can get on a prospect to make a reasonable and informed decision on their credit worthiness. Post acquisition though, you tend to get a lot more information on the customer through his/her behavior that allows you to further refine your assessment. For instance, you now have payment history - frequency and amount, you get transaction data (for products like credit card) and you can continuously mine bureau data (real time these days) to see how the customer's credit profile has evolved. All of this data will allow you to make informed decisions on how to manage the customer through actions like managing the credit line, the authorization policies - i.e. which transaction you block vs which ones you let through, the over-limit policies, the payment hold policies etc etc. Each one of them is a massive undertaking in itself. Understandably, when you are a new player with a brand new business, this is often a lower priority - after all you do not have a portfolio to manage on day 1! However, as you scale your business,?having the right portfolio management strategies and tools will help you tremendously in both managing losses but also making sure your good customers stay engaged.?As with most things in lending you will need to fine tune and iterate your way into it but make sure you have the right foundation as quickly as possible. An important point I want to highlight is while portfolio management is critical, at the end of the day your acquisition strategy i.e. making sure that you acquire the ‘right’ customers will still be the most important lever - in lending it is impossible to entirely undo the impacts of bad acquisition decisions! Here is an amazing take by Alex Johnson on what makes a transactional lending product like Credit Cards such a beast!
6. Customer experience - On first glance, this seems like an easy one - and for some of the fixed term lending products, where the customer is expected to just make fixed monthly payments, this one is somewhat straightforward. On the other hand, for a product like Credit Card which is both a lending product as well as a transactional product, this one can be a beast. It is tempting to build a ground breaking and differentiated experience out of the gate and if you can, you should. Most people though with limited budget and limited resources will have to prioritize. For example in credit card, there are a few basic things that you need to prioritize - 1) Transaction experience - the card should work as intended- online, offline, domestic, international 2) Decline experience - in the event that a transaction does get declined, building the right customer experience 3) Fraud resolution experience?- this is certainly one of those ‘moments that matter” in the the customer experience journey because it is truly an anxious moment for the customer. Spend a disproportionate amount of time and effort in building a great end-end fraud resolution experience for the customer 4) The servicing Trifecta - a) checking balance b) checking on a transaction and c) making a payment. The balance should be clear and up to date - don’t make the customer do math. The transaction should be clear and intuitive - not of a bunch of codes and jargons.?Make it easy to make payments and use all the data and technology available at your disposal to clear the payments real time while safeguarding for fraudulent payments- nothing infuriates a customer more than having to wait days for a payment to clear. Once you get these fundamentals to work flawlessly at scale, you can start focusing on the other enhancements. Believe it or not, building a positive customer experience journey is also a risk management lever. If your good customers leave because of poor service (and they will because they also have other options), all you will be left with are the fraudsters and you can imagine what that will do to your loss rates!
7. Loss Mitigation - Like portfolio risk management, this is another area that is often ignored and marginalized at the time of launch. The argument is similar - "let’s first build a business before we build capabilities like collections and recoveries". While this is true, you will likely need collections within the first year and building a robust collection and recoveries capabilities takes time. Collections is not about taking a list of delinquent accounts and hounding them with incessant calls. Doing so will likely turn off the customer but also draw the ire of the regulators. Modern collection is about taking a customer-centric, data-driven, targeted and often a digital-first approach to offer customers a set of options that will allow them to get back on their feet as quickly as possible. Just like you did in underwriting, you will need to invest in the data, the models and the technology to offer the solutions to the customer and building all of these takes time. You want to be building these capabilities at a time of strength and not when the ship is on fire.?Also, collections is one of the few areas in lending which when done right, will have an immediate and positive impact on your P&L and your risk metrics. Also, collections is one of the most scrutinized areas from the regulators - so make sure you have the right people and processes in place to appropriately document and track every action, interaction and offer made in your collection shop.?Here is a great white paper from my friends 2nd Order Solutions that looks at collections in the digital age.

These are some of the foundational steps needed to build, launch and scale a lending business. There are many areas that I have not covered in-depth here that are equally critical - none more than the data infrastructure and the decision engines that leverage the data and help make some of the critical decisions in lending - underwriting, fraud, transaction, payment, pricing etc, many of them in real time. In my next post, I will tackle this area in more detail. In the meantime, please share your thoughts in the comments section or send me a DM either here or @ddas1992 on twitter.?