"Computer says no." Why are Credit Decisions So Opaque?
Updated: Jan 29
I’ve often wondered what would happen if every lender were to make their credit algorithms publicly known.
Would people try to game it?
How? And what effect would it have, on, well, everything?
My hunch is that the reason credit algorithms are a black box is, well, mostly because at this point there actually isn’t a single ‘person’ who knows what’s in the algorithm.
Imagine trying to explain a complex financial model to someone. There’s literally no way to do it. You can either give the model away - or you can just ask for the inputs and run the numbers for someone. The latter is basically what MoneySuperMarket and other comparison websites do when they ‘pre qualify’ you.
Credit algorithms are incredibly dense. Someone once told me about a bank they used to work at - this particular bank used 999 data points to generate a credit decision! Imagine that. Imagine trying to make all of that data public - it would be a nightmare.
What WOULD be helpful, however, is if banks and lenders open up decisioning APIs that let third party developers build apps on top that give consumers visibility into whether they’d be accepted or rejected; why; and what variables they need to change to make it over the line. All of this - without running a credit check.
To be honest, I don’t think it’ll ever happen. Partly inertia. Partly lack of financial incentives for banks/lenders. But mostly because the algorithms and data they use are so dense and heavy, that they’d be totally inconsumable by anyone who isn’t a data scientist.
My personal view is that over the next few years, we’re going to see a resurgence of ‘trust’ based lending. That is, lending to people based on the reputation or character they demonstrate in their communities. Specifically, in their online communities. Uber driver with a rating of 4.9? Approved. 4.1? Denied. It sounds nuts, but remember that for most of the 20th century, this is how lending decisions were made - based on trust and character assessments.
Though I do think the uproar over Apple rejecting one computer programmer’s wife is a bit ridiculous, there is a point in there. The point is this: somewhere along the way, we (lenders) lost the plot. We forgot that a dataset can never achieve what a handshake and looking in someone’s eyes can. The human brain is still the most powerful computational machine ever invented - it processes millions of datapoints in milliseconds. Yet, we don’t use it any more when we make or break people’s lives by giving or denying them access to borrowing.
Yes, there is an imbalance right now. Machines need to do less, and humans need to do more. Let’s bring back some humanity into lending. It’s about time.