
Summary Analysis
Scale: 1 = Low, 5 = High
Responsible Banking Framework
We use Xempli’s Responsible Banking Framework to perform a high level analysis of this scenario.
Opportunities at Stake

Loss avoidance: Polidor predicts that it will reduce loan default loss by 50%
Margin: Polidor charges a risk premium for riskier loan applicants, thereby protecting overall margins
Conversion rate: by understanding risk better than other banks, Polidor is
able to make more attractive offers to lower risk customers, thereby increasing conversion rates
Social equity: Polidor argues that they are making it fairer for everyone: “why should people with sound financial attributes subsidise those that are less frugal?”
Potential Harm

Fraud: to what extent could social media be modified to game Polidor’s AI?
Price discrimination: is it OK for banks to offer different rates to different people based on their lifestyle choices?
Loss of anonymity: do people have a right to “be themselves” on social media
without worrying about the financial consequences?
Social equity: Polidor argues that they are making it fairer for everyone: “why should people with sound financial attributes subsidise those that are less frugal?”
Vulnerability to evil: what if the data fell into the wrong hands? For example, could evil agents use it to threaten customers with ransom?
Theft: doesn’t social media belong to me? What right do banks have to use this data?
Right to change: if my social media history contains things that I have done in the past but no longer do in the present, is it OK for Polidor to continue to judge me on past behaviour?
Strategies/Principles

Consent: Polidor can minimise ill-effects by first seeking consent from it’s customers. But what if customers don’t give Polidor consent? Should these customers be disqualified to apply for a loan? Or would these customers pay a risk premium?
Explainable: under GDPR, customers have the right to receive an explanation of automated credit decisions. How would Polidor achieve this without giving away
Accuracy: this is perhaps the greatest potential flaw in the product. Social media is patchy – some users are prolific, whilst others are non-existent. Social media data can be easily manipulated. Two strategies to consider: (1) use Social Credit Scoring data as an input to augment human decision-making (rather than full automation) and (2) train AI to recognise and report flaws in the data (e.g. a low confidence rating if the data is considered unreliable)
Data ownership: what if Polidor were to give customers the right to view their profile / Social Credit Score? Would this give them a greater sense of control?
Power & privilege: What if customers were given the right to “contest” decisions and request a human review?
Benefits sharing: what if Polidor offset the potential harms by doing some social good, such as helping people with financial literacy or allocating some of the gains to offer microloans to the less advantaged?
Social norms: should Polidor have run focus groups to test the social acceptability of Social Credit Scoring? And should Polidor take the path of gaining support from regulators and other influencers or should Polidor disrupt the market in the way Uber did?
Storage limitation: to overcome security concerns (i.e. data falling into the wrong hands), Poidor’s AI could generate the Social Credit Score on an as-needed basis and not store this data (i.e. it vanishes) after the credit decision has been made. Or at the very minimum, the data could be stored in an encrypted form.
Opinion
Social media shouldn’t be used to fully automate credit decision-making. It could be used to augment human decision-making so long as customers are made to feel in control e.g. must gain explicit consent; customers have the right to see how their credit score is calculated; and customers are given the right to ‘contest’ automated decisions to have it reviewed. Investment in a PR campaign, security and data access controls would be money well spent.
What do you think? Do you agree? Please share your thoughts on LinkedIn by clicking the LinkedIn logo below.