Quick answer: Influence of behavioural biases in Indian credit decisions are more often than institutions admit. Even when lenders use models, scorecards, policies, and committees, decisions can still be shaped by recent losses, vivid incidents, comfort with familiar segments, and group dynamics that do not show up clearly in dashboards.
The meeting where you really feel behavioural bias is shaping a portfolio never has that label on the agenda.
It looks like a normal Thursday.
Risk, business, collections, and analytics are in a room to discuss a specific proposal:
a small expansion of unsecured lending into a new salaried segment,
modest ticket sizes,
bureau based filters,
sensible pricing.
The analytics team walks through their work.
Sample size is decent.
Score distributions look similar to existing pools.
Vintage curves on a pilot are slightly worse, but still within an acceptable band.
Stress scenarios have been run.
Then someone mentions one case.
“There was a fraud in this segment last quarter.
Salary slip was fabricated.
Customer went to social media when we declined.
We had to spend time with compliance and PR.”
You can feel the room tilt.
Questions move from:
“What is the long term risk profile of this segment?”
to:
“How do we make sure that never happens again?”
“What will we tell the regulator if there is another case like that?”
“Is this worth it for such a small incremental book?”
The conversation slowly rearranges itself around that one story.
At the end, the decision note reads:
“Given recent fraud or case experience, recommend we go slow on this segment for now and revisit after more data.”
On paper, nothing is obviously wrong.
No policy was broken.
No metric is lying.
No one behaved irresponsibly.
Yet a single vivid incident has quietly outweighed months of boring pilot data.
If you ask later why the segment was deprioritised, the answer will be:
“The numbers were not very compelling.”
That is the polite version.
The real driver was something else we rarely name in credit meetings: how human minds handle fear, recent pain, and vivid stories.
For banks, NBFCs, fintechs, and credit teams across India, from Mumbai and Pune to Delhi NCR, Bengaluru, Jaipur, Lucknow, and Indore, this matters because behavioural bias can shape not only individual loan decisions but entire portfolio direction over time.
Because models, policies, and governance frameworks do not remove human judgement. They simply operate inside it.
Credit decisions may look structured and data led, but people still decide which segments to test, which risks to avoid, which incidents to remember, and which policies to tighten or leave untouched.
That is why behavioural biases in Indian credit decisions often show up not as obvious mistakes, but as repeated patterns of caution, overreaction, or comfort seeking.
If you listen to how senior people in Indian banks, NBFCs, and fintechs talk about decision quality, there is a familiar line:
“We used to rely on gut and relationships.
Now we have bureau data, scores, rule engines, and governance.
Human bias is much lower. The system protects us.”
It comes up in many forms.
In a Board Risk Committee, someone says:
“Our frameworks and policies ensure consistency.
Individual biases in Indian credit decisions may exist, but they cannot move the portfolio much.”
In a model validation discussion:
“As long as we rely on scorecards and cut offs, we reduce judgement error.
Over time, data will correct any residual bias.”
In a strategy offsite, a senior leader jokes:
“We all have intuitions, but thankfully we have the numbers now.”
Underneath is a comfortable assumption:
Behavioural biases are an issue for retail borrowers, such as impulse spending, over confidence, and anchoring on offers.
For lenders, professional structures neutralise most of that.
Whatever bias remains is noise around the edges, not a driver of systemic outcomes.
It feels true because:
we have policies,
we have committees,
we have audit trails,
we have model documentation.
The uncomfortable reality is simpler:
the models and policies sit inside human narratives, not outside them,
and behavioural biases in Indian credit decisions show up less in single wrong decisions, more in the portfolio of choices we never quite make.
If you watch actual credit decisions over quarters rather than hours, you do not see dramatic failures.
You see small, repeated patterns.
In many discussions, the team starts with one case.
Sometimes it is:
the worst fraud they have seen in a new segment,
the loudest social media escalation,
or the cleanest success story from a digital pilot.
Whatever comes first sets the anchor.
Then every subsequent point is measured against that story.
You see this in:
SteerCo decks where the first page of a new segment update is key incidents seen so far, with two or three narratives,
followed by pages of aggregated data that everyone nods through quickly.
Two months later, someone describes the segment decision as:
“We tried that, but it is messy.”
or
“We have seen it works very well.”
The sample size they are mentally using is three or four incidents, not the full cohort.
The models and MIS are still there.
They are just being weighed against whichever story came first.
Loss aversion is simple: we dislike losses more than we like equivalent gains.
In credit portfolios, this shows up as overreacting to recent pain.
A familiar pattern:
A new segment or product starts decently.
After a year, one or two bad vintages come through, maybe driven by macro, maybe by a partner issue, maybe by pure timing.
Those losses hit P and L and internal credibility at the same time.
In the next quarterly portfolio review, the questions change:
“Why did we go here at all?”
“Who sponsored this segment?”
“Was the upside really worth the noise?”
The fact that:
70 to 80% of that segment is behaving as expected,
and the under performance is localised and fixable,
gets less airtime than:
one graph suddenly spiking red,
and the political cost of defending something that just hurt.
Over the next 12 to 18 months, you see:
limits and budgets quietly cut,
credit appetite statements rewritten,
internal enthusiasm moving to safer places.
If you compare risk reward on paper, the segment may still make sense.
If you compare emotional memory, it does not.
The decision is not irrational.
It is just loss averse in a way that models did not factor for.
When institutions enter new areas, such as new cities, new occupational groups, or new partners, they carry unspoken priors.
You see it in offhand lines at the start of projects:
“This city has always been tricky.”
“This income group is very aspirational. They tend to over borrow.”
“This partner’s customers are more disciplined. They are used to formal credit.”
These priors shape what teams look for.
If the initial belief was this is risky:
early delinquencies are read as confirmation,
clean behaviour is seen as we got lucky this time.
If the initial belief was this is premium:
early stress is explained away as one off incidents,
good behaviour reinforces the story.
You can see confirmation bias in the language of early pilot review notes.
For a segment seen as risky:
“As expected, roll rates are higher. Need caution.”
even when the actual numbers are only slightly worse than core.
For a segment seen as attractive:
“Early performance broadly in line. Some noise in X and Y channel.”
even when indicators are signalling a structural issue.
The model outputs are the same.
The story we tell ourselves about them is not.
Status quo bias is the tendency to prefer the current state, even when change is rational.
In credit, it often wears the clothes of prudence.
You hear lines like:
“Our current mix works. Why disturb it?”
“We understand salaried prime in these cities. Let’s deepen that first.”
“NTC and semi formal segments are important, but perhaps we should wait 12 to 18 months.”
On any given day, each of these is defensible.
The cumulative effect over years is:
a book heavily skewed to segments that felt comfortable,
under exposure to segments that are riskier to learn but critical for long term relevance,
growing dependence on a narrow set of employer types, geographies, and channels.
The irony is that teams then speak about this narrowness as:
our risk appetite,
or our strategy,
when in many cases it is just:
we did not want to live through the discomfort of learning something harder.
The models reflect that history.
They do not question it.
Most important credit decisions in Indian institutions are not taken alone.
They happen in:
Credit Committees,
Product SteerCos,
one level up escalations.
Behavioural bias here is less about any one person’s flaw, more about how the group behaves.
Common patterns:
The most senior person speaks early, and the rest calibrate opinions accordingly.
The business head and the risk head both know that pushing too hard on one point will cost them capital in another space, so they trade compromises silently.
A member who raised a concern last time and was overruled hesitates to speak again.
On the minutes, the outcome looks like consensus.
In reality, it is often anchoring on whoever framed the issue first, plus a layer of self preservation.
The models and policies are still referenced.
They just operate inside a group psychology that no dashboard captures.
Given all this, why does behavioural bias not show up more explicitly in the way we run credit?
Because our artefacts are built to catch technical errors, not human tendencies.
Core systems and documentation capture:
score and policy outcomes,
exception approvals,
who signed what and when.
If a higher risk loan was approved, the system will show:
exception reason code,
approver ID,
maybe a short note.
It will not show:
the story that dominated the room,
the fear that drove caution,
or the incentive someone was quietly optimising.
So when internal audit or compliance reviews decisions, they see:
whether the right boxes were ticked,
not the biases that shaped which options were ever presented.
Portfolio MIS is built for aggregation:
disbursals by segment and channel,
GNPA and roll rates,
score distributions,
collection effectiveness.
These tell you what happened.
They do not tell you:
which segments were never piloted because someone had a bad experience years ago,
which partners were avoided because of one escalation,
which pricing decisions were driven by fear of comparison rather than risk.
There is no chart labelled:
opportunities avoided due to one vivid incident,
or segments neglected because of status quo comfort.
So biases exert their influence mainly through what never appears in the pack.
Most biases are not policy breaches.
They are baked into policy itself.
Examples:
A long standing rule that certain professions are not to be touched, based on an incident from another era.
A geographical appetite grid that is more about institutional memory than current data.
A long held reluctance to revisit score cut offs upward or downward, regardless of changing macro.
These norms feel like institutional wisdom.
In truth, many are yesterday’s fear frozen into rules.
Because they are in the policy, not in the exceptions, they are almost never questioned.
Teams that handle behavioural bias better are not more moral or more academic.
They behave a little differently around decisions.
A few patterns I have seen follow.
In more grounded rooms, when someone narrates a vivid incident, you hear a follow up like:
“That is one strong story.
Let’s also see what the full distribution says before we anchor on it.”
Or:
“This is clearly shaping how we feel about this segment.
Can we park the fear for a minute and see whether it is representative?”
It does not turn into a workshop on psychology.
It is just a small habit:
acknowledge the emotional weight of a case,
then deliberately look at the data without letting that case dominate.
In decks, this shows up as:
one page for notable incidents,
followed by a separate section on overall performance,
with someone in the room guarding against conflating the two.
When a decision has clearly been shaped by recent pain or comfort, some teams ask a simple reversal:
“If this recent incident had not happened, would we still decide this way?”
or
“If we had just seen a big success in this segment, would we be as cautious today?”
The point is not to change the decision on the spot.
It is to surface that behavioural context exists at all.
Often, it leads to small mid course corrections:
not killing a segment entirely, but tightening channels more surgically,
not expanding an area just because of one success, but committing to a proper pilot.
In some institutions, buried near the back of the unsecured risk pack, you find charts that make people shift in their seats.
Things like:
share of book in top 10 employer categories over a five year trend,
exposure by state vs broader market opportunity,
products and segments never piloted but present in peer books.
These pages rarely produce immediate action.
They perform one quiet function:
they remind the room how narrow or path dependent the book has become,
and make it harder to disguise status quo bias as prudence.
Nobody likes those slides.
That is often a sign they are doing their job.
Behavioural bias amplifies when people are afraid to look difficult.
In more seasoned teams, you sometimes see:
a deliberate invitation to quieter voices in credit committees,
a practice of asking the most junior person in the room, “What are we not seeing?”,
a willingness from senior people to say, “I may be overly influenced by X. Convince me otherwise.”
This is not about being democratic for its own sake.
It is a practical hedge.
If only one or two senior narratives dominate, you get highly correlated biases.
Allowing structured dissent reduces that correlation.
It does not make anyone infallible.
It makes the portfolio of decisions a bit less exposed to one person’s fears or enthusiasms.
A small but important habit:
once in a while, they pick one long standing rule and ask:
“What original story led to this?
Is that story still valid?
If we were writing this rule fresh today, would we write it the same way?”
This does not lead to dramatic overnight shifts.
It does lead to:
occasionally relaxing a rule that no longer fits the world,
tightening one where complacency has crept in,
or at least updating the rationale so future teams know why it exists.
In those moments, behavioural bias moves from being an invisible driver to a discussed input.
That is often enough to limit its damage.
It means decision quality is not only about better models, better bureau data, or tighter governance.
For Indian banks, NBFCs, and fintech lenders, credit portfolio strategy, segment expansion, policy design, and committee discipline are all influenced by behavioural tendencies like anchoring, loss aversion, confirmation bias, and status quo comfort.
The practical challenge is not eliminating bias completely.
It is recognising where human judgement is shaping portfolio choices more than the institution admits.
The bias I see most often in Indian credit discussions is not optimism or fear.
It is the quiet belief that, because we sit with:
data,
models,
policies,
committees,
we are largely insulated from the cognitive shortcuts that affect ordinary decisions.
The truth is less flattering.
Our tools make some errors harder, yes.
They also make other biases easier to hide:
in which incidents we choose to remember,
in which segments we quietly avoid,
in which rules we never revisit.
If there is one useful question to ask at the end of any important credit decision, it might be:
“Looking back a year from now, if this decision turns out to be wrong,
will we say the model failed,
or will we recognise that we let one story, one fear, or one comfort zone weigh more than it should have?”
The teams that sit with that question honestly do not become free of bias.
They simply stop pretending that scorecards and approvals live outside human behaviour.
And that, in a system built on judgement under uncertainty, is usually as much realism as anyone gets.
Behavioural biases in Indian credit decisions are mental shortcuts or emotional patterns that affect how lenders interpret data, assess segments, or make portfolio choices. These include anchoring, loss aversion, confirmation bias, status quo bias, and group influence.
Not fully. Models can reduce some judgement errors, but they do not remove the human stories, fears, assumptions, and institutional habits that shape which segments get funded, which policies get tightened, and which risks are avoided.
Because teams often remember recent frauds, escalations, or losses more strongly than slow moving portfolio averages. A single vivid event can anchor discussion and shift risk appetite more than a full data distribution.
Lenders can reduce behavioural bias by separating incidents from aggregate data, using reversal questions, reviewing uncomfortable portfolio concentrations, allowing structured dissent, and revisiting policies that may be based on outdated fear rather than current evidence.
