Is cash-flow based decisioning finally ready to turn the corner?

In one of my previous posts, I discussed the basic framework of doing underwriting - the data, the model, the policy, the infrastructure. Since then I have gotten a lot of questions around one specific topic - cash flow based underwriting (sometimes used interchangeably with the term open-banking) and its potential. I recently also had the opportunity to attend a symposium hosted by Nova Credit on the same topic and engage in a robust exchange of ideas and views from industry experts and practitioners (thank you Misha Esipov and Ellyse Q. ). In this month's edition of Coffee and Credit, I wanted to recap and share some of my POVs on this topic. 

So what is cash flow based underwriting or decision-making? 

At its core, a successful lending business for both consumer and SME lending usually rests on successfully predicting i) ability to pay, ii) willingness to pay and iii) and future stability of cash flow. Historically lending businesses have relied heavily on bureau data for underwriting. As you can imagine, when a customer has a reasonably long history with the bureaus (typically >2 years), the data is pretty accurate in predicting willingness to pay. Basically, given enough length of history, a customer or SME's past pattern of how they have handled loan payments is usually a good predictor of how they are going to behave in the future. However, accurately predicting ability to pay is an area where many lenders face gaps. Historically lenders have used self reported income combined with a picture of debt obligations from the bureaus, plus any other self reported obligations (like alimony payments, child support etc) to calculate ability to pay. For some higher ticket loans (like mortgages), there may be a more involved (albeit high-friction) doc-upload process to verify income. They also factor in the bureau data with the implicit assumption that if you had a stellar history of meeting your debt obligations in the past, you will likely also be good on your ability to pay. As you can see, all of these approaches have gaps. Not to mention, most lenders have limited or no line of sight to the stability, frequency and size of future cash flow. Enter cash-flow data! Having an accurate picture of a borrower’s cash flow - how much money is coming in, how much is going out, the destination, the frequency, the timing etc can be huge in determining ability to pay and stability of income and can help in improving both direct and marginal lending decisions, especially when you also consider that even when available, bureau data is very ‘laggy’. When used effectively, cash-flow data can be used in all types of lending decisions - approve/decline, determining the size of the loan, line assignment, pricing, account management policies like credit limit management, payment hold strategies etc. In my experience, when this data is available, it has definitely proved to be a powerful risk splitter.

So why hasn’t cash-flow based underwriting and decisioning really taken off?

Despite its seeming usefulness, less than 5% of all underwriting decisions use true automated cash flow data. In his most recent post (a must read), Jason Mikula also talks about the much broader point of how lending, despite all the tall claims about the emergence of alternate data, machine learning etc, still seems to overwhelmingly rely on bureau data for underwriting for the most part. Why is that? To understand this, let’s first start by painting a picture of the lending landscape as it stands today with various types of FIs and their relationship to cash-flow data.

1. Big Banks - While the big banks often have a big deposit base and with it access to primary banking relationship and cash flow data, they often don’t play in the segments that could benefit most from it, like new to credit, recent immigrants etc, for a host of reasons, primary of which are economics and reputation risk. This often leads to the investments needed in building the infrastructure for cash flow decision-making to be deprioritized over competing priorities (and I have presided over some of these painful but economically pragmatic decisions myself). 
2. Credit Unions and Regional Banks - The credit unions and small and mid-size regional banks fall somewhere in the middle. They have the bank data but not the technology or the infrastructure wherewithal to easily incorporate new data sources and policies. 
3. Neo Banks - They have some of the data and most the technology but do not have the scale to effectively monetize. They typically do one thing very well (their wedge!) and don't have the second or the third product upgrade needed to make these segments economically viable.
4. Non-Bank lenders - They are probably the most disadvantaged of the lot. Because they do not have a direct access to any deposit data (without using a friction-filled customer-permission based process), they also find it hardest to test and implement it. 

So let’s dive a bit deeper into the specific challenges that FIs have to deal with when it comes to cash flow data

Data - Getting cash flow data to build models is not easy - why? Because unlike bureau data where you can go to bureaus and get an anonymized data set per your specifications and then use that to build your V.1 model, you can’t do that for cash flow. There is currently not a central repository (at least not at scale yet; though it is starting to get addressed by the likes of Akoya , Financial Data Exchange , Early Warning® etc). Also, any effort to validate or retro-score a model post build is very difficult with cash flow data. Secondly, unlike the bureau data, cash flow data is not standardized yet. This creates challenges in data ETL and running models on the data when you consider that most of the underwriting and decision-making infrastructure today is built to ingest a standard format typically from the bureaus. 

Customer Experience - In the age of speed and instant gratification, asking customers to provide bank data, even simple log-in credentials, creates a lot of friction for most of the loan applications leading to drop-off in the funnel. Also, unlike bureau data where once you build the pipes and have seamless and ongoing access to the customer’s bureau data for portfolio management decisions in the future, you simply cannot do that with cash-flow data without causing significant friction.  

Policies - Building policies around cash flow data may create some unintended consequences. As one of my peers recently mentioned - seniors have limited cash flow but high assets. A policy that exclusively looks at cash flow scores may inadvertently discriminate against seniors.

Infrastructure - Current data and decision infrastructures are based on bureau data and ingesting a completely orthogonal data source and integrating it with an existing infrastructure creates its own set of challenges. 

And finally, the ultimate catch 22 - The Economics - Big banks have the data and the infrastructure to use cash-flow underwriting effective today and they also have a full suite of products to cross-sell and up-sell. But for them, the economics of investing in the infrastructure and product development to support segments like new to credit, recent immigrants who tend to benefit most from cash flow underwriting do not make sense - not to mention come with a lot of reputational risks. These underserved segments tend to be constantly in the radar of regulators and public officials and any missteps howsoever unintended, can cause severe damage to the reputation. On the other hand, while smaller banks and non-banks focus on serving this segment, they do not have easy access to the bank data and/or do not have the full suite of products to cross sell / upsell eventually to justify the economics. 

So what does the future of cash flow based underwriting and decisioning looks like:

In the short term the cash flow will continue to see use in the following areas:

1. Segments with no or limited bureau data - FIs catering to this segment have been extensively using cash flow data and my bet is that it will continue to grow in prominence. It is easy to understand how most of the challenges that I have listed above either are non-existent or FIs are able and willing to surmount them because cash-flow data provides the only real opportunity to underwrite some of these segments. Also, from a customer’s standpoint, if you have very limited options to get a loan, you will be much more likely to go through some of that additional friction.
2. Marginal decisions - Marginal decisions are areas where lenders made a certain decision based on existing data and policies but are willing to take a ‘second look’ on a marginal segment if additional data were made available. This is where policies around second look approvals -  “we cannot approve you now but if you provide us with your bank data, we will take another look…..” -  or giving someone a higher loan amount/credit limit than what was originally approved come into play. One word of caution here though - regulators often are not huge fans of too many second look programs:  “If you are willing to approve someone or give someone a favorable term eventually as a second look, why didn’t you incorporate the policy and data into your first look decision in the first place”… is often something they ask. 

In the long term though, I see a few things happening - mostly leading to the wider and pervasive use of cash-flow data:

1. The CFPB kicked off the work on implementing section 1033 of the Dodd-Frank Act - making a set of policies towards the consumer’s right over their own personal banking data - details here. It is widely believed that the outcome of this will lead to greater portability and access to consumer-permissioned financial data. When you combine this with the rapid strides made in the open banking space through API enabled data transfer from the likes of Plaid , MX , Finicity, a Mastercard Company etc, I think we are now on the cusp of addressing the access and standardization issues. 
2. Next we are also starting to see the emergence of consortiums of FIs of various shapes and sizes - the likes of  Akoya, FDX, EWS etc mentioned above - who are trying to create a bureau-like repository of #cashflow data. This will definitely go a long way in addressing the user friction in accessing cash flow data - not to mention ease of model development and scoring. 
3. We are also starting to see many of the decision engine providers like Alloy and Provenir start to integrate with the above data providers and allow for the easy integration of cash flow data with traditional data in their decision engines. Here's some more details about the recently announced partnership between Prism Data and Provenir (Congrats Jason Rosen and Erin Allard !!)
4. Finally, the TAM and use-cases for this data will only grow. After the decline during and right after the pandemic, the level of immigration is back up to pre-pandemic levels and likely growing. Custodial checking accounts like Greenlight and Step etc are becoming ubiquitous and proving to be a new and valuable sources of data for underwriting future lending products for a new generation of customers (as a parent of a preteen, I love these products and can see how potential lenders can leverage the data to 'pre-approve' them for viable lending products when these kids come of age)

Suffice to say, I am bullish about cash-flow data and optimistic that in the next few years, it will become more mainstream. Companies like Nova Credit and Prism Data who have not only invested in building cash flow based scores, attributes and solutions but are also rapidly expanding their partnership with FIs on one end and third party platform providers on the other side are going to be at the forefront of making this transition. It feels like the stars are finally starting to align. If you have or are using cash flow data, what has been your experience - what are the challenges you have faced and do you agree that the next few years will see cash-flow based underwriting becoming more mainstream?