September 12, 2023
5 min read
Throwback to Autumn 2019. The excitement around the fintech industry was wild. We witnessed the launch of new financial apps revolutionising their space every month: Instant cash advance, consumer credit in 24 hours, Buy now, pay later… The regulations around open banking have unlocked plenty of opportunities to fund, and incumbents have started to feel threatened by the new players, especially in lending. A standard narrative from newcomers on the market: promoting an “innovative” credit scoring engine as their primary asset while they were just starting their business. Sexy for sure, but how reliable is it?
We can define consumer credit scoring as a model that assesses individuals’ ability to repay a loan by detecting several patterns of solvency. To find those solvency patterns, we must understand how people can become insolvent… So, how do you launch a successful new lending business in an untapped and risky market? Making bold assumptions about insolvency roots or accepting initial (big) losses.
Indeed, a partial picture of consumers' data and advanced unsupervised learning models can’t replace the experience of getting various defaults. We can even see it as creating a valuable asset that will help lenders outperform the competition in the following decades. For instance, launching a freelancer credit offer but excluding platform workers automatically because they seem (at first sight) riskier removes an opportunity to increase knowledge about this population and build custom and profitable offers to address them.
This is an issue. Startups in the lending industry are asked to fill several boxes simultaneously. During a low interest rate period, BNPL companies grew massively without being super selective in their scoring process. The perception of their default rate was neutral as long as their number of users reached insane thresholds. However, their business model is questioned when money costs more and losses are seen as unfavourable. There should be a tradeoff in that situation. Startups in lending are funded based on the assumption that they can beat incumbents one day. The reality is that acquiring a high number of users thanks to a seamless user experience or providing innovative financial products is a good start. Still, it won’t replace the experience of getting defaulted loans and learning from them.
Building a world-class scoring engine has to be their priority to outperform competitors independently of the interest rate. A winning strategy lies in the combination of taking risks by increasing the acceptance rate while leveraging more datasets on consumers to optimise their decision engine.
At Rollee, we spend a lot of time getting to know the freelancer audience and their access to loans and mortgages.
Let’s assume you receive a hundred loan requests from a group of freelancers and plan to lend only to twenty of them. The initial step will be to rank them based on your collected data and select the 20 best profiles.
This is a rational way to proceed. The loan applicants answered some questions about their professional situation, uploaded documents, connected their bank accounts and proceeded with an ID verification. After a few months, you determine they have all repaid their loans on time, and your default rate is 0%. Is this a success? It depends on your ambitions.
Another lender will probably act differently by providing loans to the whole group, asking them to connect alternative data sources like their payroll account and figure out that among the 100 applicants, 40 have good profiles to lend to while the 60 remaining ones seem riskier based on hidden patterns they have detected. At scale, it significantly affects profitability and inclusivity for non-traditional profiles. Enriching your customer knowledge and refining your decision process with all the new patterns you detect will bring a more comprehensive picture of your target consumers and reduce your losses while growing.
At Rollee, we encourage risk teams to start as early as possible to enrich their datasets with our open data API so they can continuously feed their scoring engine to make it consistently better.