Defaults are inevitable.
For the past few years, we have been hearing about predictive analytics that can identify a borrower who is going to default before the default actually happens.
Time and again, we have seen the limits of this idea.
Defaults aren't a prediction problem that technology will eventually eliminate. Defaults are an uncertainty that lending produces, by design. What lenders need is not machinery that assumes this uncertainty can be removed, but machinery designed to operate effectively under it.
You cannot predict at an individual borrower level who is going to default. You can, at best, predict, over a sufficiently large cohort of borrowers, what the default rate might be.
The behaviour of large groups can be predicted statistically. The behaviour of a particular individual cannot.
If you have one million borrowers, you can make remarkably good statements about what will happen to the portfolio. Narrow it to 100,000 borrowers in Karnataka, and you can still make useful predictions. Narrow it further by repayment history, geography, employment, credit score and other risk characteristics, and you can create increasingly meaningful cohorts.
But as the cohorts get smaller, the certainty of those predictions starts decreasing.
Eventually, you arrive at an individual borrower, say Ashok.Ashok's salary didn't arrive yesterday because his employer had a cash-flow problem. Your model cannot know that. It cannot know that unless you invade the privacy of Ashok and everyone he interacts with. Which you absolutely should not do.
This is the fundamental limitation of trying to predict defaults at an individual level.
For example, if we look at borrowers onboarded in Karnataka who have paid 60% of their EMIs on time and have a credit score of 650 or above, we may be able to say with reasonable confidence that 90% of them will repay on time in April.
But we cannot tell you which 90%.
And lenders have to operate with this uncertainty.
Borrowers do not fit into a neat Excel formula. Even if they did, you would not know all the variables that go into calculating it.
Some of these variables are macro.
In the western region, for instance, collections can soften during August and September around Ganesh Chaturthi. We also see periods where household cash flows come under pressure around the start of the school year, when families are managing school fees, admissions, books and other education-related expenses.
This information is useful.
It can tell us that collections might be lower during a particular period and help us anticipate how a portfolio could perform.
But it still does not tell us who is going to miss their EMI.
Then there are micro circumstances.
A salary gets delayed. A business has a bad month. There is an unexpected expense. A school fee needs to be prioritised. Something happens in the family. Or someone simply forgets to maintain enough balance in their account.
It is impossible for a lender to know every individual's circumstances at every point in time. This is why lending is ultimately a game of large numbers.
Risk models can be extremely effective at estimating how a sufficiently large cohort will behave. But every individual in that cohort is a different person, going through a different set of circumstances.
None of this is to say that predictive analytics is not useful. In fact, it is the opposite.
Predictive analytics is a powerful tool when used for the problem it is actually capable of solving. It can tell you where a portfolio might end up. It can tell you where to focus effort. It can identify groups of borrowers that are more likely to need intervention than others. It can help lenders allocate their collection resources more intelligently and set realistic expectations for portfolio performance.
The mistake is in assuming that this translates into certainty at the individual level. It doesn't.
There will always be uncertainty at that level. And that means defaults will always exist.
The more interesting opportunity for AI, therefore, is not to predict every default before it happens. It is to make lenders better at responding to defaults when they do happen.
When a borrower misses an EMI, the system should be able to help determine what happens next. Who needs an immediate intervention? Who is likely to self-cure? Which borrowers require a human conversation? What is the appropriate channel and timing? How should collection teams prioritise their limited resources?
This is where predictive analytics, combined with AI-led decisioning and communication, can make a meaningful difference.
The objective should not be to chase every borrower harder. It should be to make every intervention more relevant.
That distinction matters, particularly as credit expands to millions of new borrowers across India. At that scale, collections cannot rely entirely on manual follow-ups, nor can it rely on the assumption that an algorithm will perfectly predict human behaviour.
The future lies somewhere more practical: using data to understand portfolios, AI to prioritise decisions, and human judgement where individual circumstances matter most.
Trying to predict every default before it happens is a game you cannot win. And it is a game you should not play.
The real competitive advantage will come from building a system that accepts uncertainty, and is intelligent enough to deal with it.
The author is Co-founder & CTO, DPDzero, an AI-led debt collection platform.