Can Credit Scoring replace Judgemental Lending?

‘Let us Parametrise’, just like the temptation to automate everything, is another articulation we see clients getting excited about. Get data, feed in the system, create a multi variable algorithm and loan is approved! While this may work in stable, digitally savvy (data-rich) environments, it is not a solution for everyone or for every situation. Historically, results for some categories like Consumer Lending, Mortgages and Supply Chain Finance, show much better performance, attributable as much to adequate security and clear sight of cash flows. However, for large ticket loans, which generally involve complex assessment and require longer time-line projections, over reliance on parameterisation may be perilous. A deliberate bias towards human judgement, even with its subjectivity and flaws, may still be a better bet.

Parameterisation has limits. A typical scoring model is a mix of Application and Behavioural data, from intra (e.g. account transactions) and inter-organisational (e.g. Bureaus) sources. Added to this are incidentals like management and industry risks. The models may work well in steady state but in case of any disruption or material changes in macro factors, relevance of behavioural data may drop drastically.

Adjustments, in any model based on past performance, comes with a lag and need sufficient data to be retrained. For example, unexpected weather changes may impact agri-related or garment industries or similarly, increase/decrease in tariffs may have a short-term impact across many industries, but these factors may remain under-weighted in the model based on previous averages. For watershed events like COVID-19, the whole model and assumptions may need to be amended. But without any precedence or any supervised data set to train the model, there has to be much reliance on experts. So, is scoring an inefficient way of evaluating loans?

Of course not. It has its strengths- reduced TAT, better objectivity, reduced manpower requirements, reduction in operating costs et al. However, the beauty lies in balance. And balance is achieved by using the best of both worlds. Our advice to Banks: use scoring as check point and leave room for human judgement. The size of the room for human intervention depend on

>>Risk-Reward: governed by customer segments, ticket size, security

>>Stage of business: startup vs established, large portfolio vs small portfolio

>>Macro factors: Dynamics of policy/tariffs, demand-supply, seasonality, disruptions

>>Availability of Skill-sets

No matter how impressive the academic paper or statistically significant the results, the fact remains that parameterisation has limits and organisations need to take a judicious call on ‘when’, ‘for whom’ and ‘how much’ they should rely on it. In the light of recent events, where the premise ‘past data is a predictor of the future’ is upended, it may just be better to err on the side of human judgement!