How we can use explainability to improve FinTech platforms

Earlier this year, Facebook introduced a fantastic tool called “why am I seeing this”, that allows a user to look behind the curtains of its news-feed algorithm. To quote Facebook:

We’re introducing “Why am I seeing this post?” to help you better understand and more easily control what you see from friends, Pages and Groups in your News Feed. This is the first time that we’ve built information on how ranking works directly into the app.

This is in addition to its older tool, “why am I seeing this ad” which allowed users to see the selection criteria used by advertisers to target users.

To be clear, the tool here answers the question, “what were the criteria used to show me this post or ad”, and not necessarily “why am I seeing this ad”. The answer to the latter could end up being a philosophical one that would question the very purpose of Facebook’s existence or advertising. There’s also a personalisation element referring to the fact that news-feed posts (and ads) are customized for each of the users. That said, the information on what steps lead to me seeing this ad or post gives me sufficient information on FB’s selection mechanism behind the scenes. Laying them all out makes it possible for the users and the makers to flag off unintended consequences of a single minded pursuit of metrics like clicks or shares.

Explainability in FinTech:

Explainability has been usually spoken about in the context of opaque AI algorithms that can at times be too complex to understand. However, I think there’s great scope for applicability of this idea in other areas as well — especially in the case of new-age FinTech platforms.

If you go through any of the Play store reviews of the new-and upcoming online lending platforms, you’ll come across several like these:

Really getting frustrated of using this app. I applied for loan. I uploaded all the documents and even fields officers also visited to my office address. Since, after the verification from field officer I didn't get any response /update from ###. I keep on sending mail to #### but I received system generated acknowledgement mail only. Kindly look this review and be aware of this.

*Name of the company has been redacted since it isn’t important in this context.

The above example of a user frustrated due to their loan application being rejected without being offered any reason behind the rejection is characteristic of the methods of the financial industry, a sector that has historically operated like black box.

Loans have been traditionally given to “credit worthy”, determined using a metric called the credit score. The new platforms have sought to come up with a new set of metrics that are better than the older one.

A loan could make or break someone’s life. This is one of the reasons why loan sanction officers at rural banks, where access to credit is hard, make up to 40% of the loan amount as kick-backs.

By not indicating the reasons behind the rejection, loan providers offer no way for the users to work on improving their credit worthiness legitimately. And consequentially, a whole bunch of companies have showed up that offer to improve the credit score through various means.

Now, I understand that as a lender, you are always at the risk of being gamed by the users. However, indicating the selection criteria broadly, like salary, line of work, neighbourhood, etc could certainly improve the user experience without giving away too much information.

A crude mock-up of what a loan application rejection screen could possibly look like

Explainability doesn’t have to be limited to loan applications alone. The primary revenue model of most players in the financial industry involves tapping into the origination fee (fee from cross-selling) in some way or the other. Insurance sales, for example, earn the sellers 25–40% of the yearly premiums as commissions, and mutual funds earn 1–2%, and ULIPs can earn a lot more. Over reliance on a revenue model predicated on commissions sets up a value chain where the incentives aren’t necessarily aligned. Expectedly, one study conducted by Monika Halan & Renuka Sane concluded that mis-selling is rampant across the Indian financial ecosystem.

PhonePe recommends two Tax Saving Funds. However, it offers no explanation on how these funds were chosen. Adjoining figure offers a possible (?) explanation.

Introducing an Explainability tool to breakdown the reasons behind the credit card or insurance banners will also go a long way in reducing mis-selling and offer ways for the organisations to improve their targeting.

NiYO cross-sells a deposit product to its a Global Card customers. It would be nice to know the criteria applied in selecting the target audience for this direct mailer.

Ignorance, the real devil

While to many outsiders tech companies may seem like hyper-rational entities exercising great care and diligence in their actions, the reality is that tech is often built with surprising ignorance of the users. And although most of the people in the industry do have the right intentions, it’s easy for them to get metamorphosed into something else.

Tools that can offer a way for the people in tech to explain their actions to themselves, and reflect on whether their good intentions are indeed translating into action, are essential. These will in many ways ensure that we get the knowledge that we need need to prevent harm before it happens.