WTF Does a Risk Analyst Actually Do?

In partnership with

You're scrolling through Naukri. Half the entry-level data postings in banking and fintech say "Risk Analyst."

The salary looks decent. The requirements say SQL, Excel, and "understanding of risk frameworks."

You have no idea what that means. You apply anyway.

Let me save you from an embarrassing interview. Here is what Risk Analysts actually do, explained without the Wall Street jargon.

The simplest version of the job

A bank lends money to people. Some of those people don't pay it back. The bank needs someone to figure out: who is likely to pay us back, and who isn't? And they need to know this before they hand over the cash.

That is a Risk Analyst.

It's not just banks, though. Insurance companies need to predict who is likely to file a massive claim. Fintech apps (like CRED or Simpl) offering Buy-Now-Pay-Later need to decide who gets a credit line. E-commerce companies need to predict who is going to refuse a Cash-on-Delivery order.

Rule of thumb: Every company that extends something of value to a customer before getting paid back needs a Risk Analyst.

The 3 Types of Risk You Will Encounter

Credit Risk: Will this person repay their loan? This is the most common type. You are analyzing borrowers, their income history, and building rules to decide who gets approved and at what interest rate.
Fraud Risk: Is this transaction real, or is someone stealing? You are looking at transaction patterns to catch bad actors without blocking legitimate customers. (That ₹50,000 purchase at 3 AM from a new device in a different city? Probably worth flagging).
Operational Risk: What could go wrong internally? System outages, data breaches, human errors. This is less data-heavy, but analysts still measure it.

For a fresher, you will almost certainly start in Credit Risk or Fraud Risk. That is where the heavy SQL work is.

A real day as a Credit Risk Analyst

You work at a fintech company offering personal loans. Every day, thousands of applications come in. Your team has a set of rules that auto-approve, auto-reject, or flag for manual review.

Morning. You check yesterday's approval numbers. 5,200 applications. 2,800 approved. Your approval rate is 54%. The target is 50-55%, so you are in range.

But then you check something else: Of the loans approved a few months ago, how many have defaulted?

SELECT
    DATE_TRUNC('month', approval_date) AS approval_month,
    COUNT(*) AS loans_approved,
    SUM(CASE WHEN status = 'Default' THEN 1 ELSE 0 END) AS defaults,
    ROUND(100.0 * SUM(CASE WHEN status = 'Default' THEN 1 ELSE 0 END) 
          / COUNT(*), 2) AS default_rate
FROM loans
WHERE approval_date >= '2025-08-01'
GROUP BY 1 ORDER BY 1;

The Result:

Month	Approved	Defaults	Default Rate
Aug 2025	8,200	328	4.0%
Sep 2025	9,100	410	4.5%
Oct 2025	10,500	578	5.5%
Nov 2025	11,200	728	6.5%

The default rate is climbing. 4% to 6.5% in 4 months. That is not a small change—at a ₹50,000 average loan size, a 2.5% increase in defaults means an additional ₹14 Crore in dead losses per month.

This is a massive problem. Your job now is to figure out which segment of borrowers is defaulting.

Afternoon. You dig into the defaults by borrower profile.

SELECT
    CASE
        WHEN credit_score >= 750 THEN '1. Excellent (750+)'
        WHEN credit_score >= 700 THEN '2. Good (700-749)'
        WHEN credit_score >= 650 THEN '3. Fair (650-699)'
        ELSE '4. Poor (<650)'
    END AS score_bucket,
    COUNT(*) AS total_loans,
    SUM(CASE WHEN status = 'Default' THEN 1 ELSE 0 END) AS defaults,
    ROUND(100.0 * SUM(CASE WHEN status = 'Default' THEN 1 ELSE 0 END) 
          / COUNT(*), 2) AS default_rate
FROM loans
WHERE approval_date BETWEEN '2025-10-01' AND '2025-11-30'
GROUP BY 1 ORDER BY 1;

The Result:

Score Bucket	Loans	Defaults	Default Rate
1. Excellent (750+)	5,600	112	2.0%
2. Good (700-749)	8,100	324	4.0%
3. Fair (650-699)	4,800	384	8.0%
4. Poor (<650)	3,200	480	15.0%

The "Poor" bucket has a 15% default rate. That is terrible. But here is the critical question: Did we recently start approving more people in this bucket?

You check the policy logs and find that 3 months ago, the Growth team pushed to lower the minimum credit score threshold from 650 to 600 to "expand the market." The result was a flood of high-risk borrowers who are now defaulting.

Your recommendation: Revert the minimum score to 650. Or, if leadership wants to keep the lower threshold, add strict secondary conditions—like requiring 2+ years of employment history for anyone in the 600-650 bucket.

The vocabulary you need to know

You don't need to be a math genius, but if you drop these terms in an interview, you immediately sound like an insider:

PD (Probability of Default): The likelihood that a borrower won't repay. A PD of 0.05 means a 5% chance of default.
EAD (Exposure at Default): How much money was outstanding when they stopped paying? If they borrowed ₹50k but already repaid ₹15k, the EAD is ₹35k.
LGD (Loss Given Default): If they default, how much do you actually lose? If you can recover ₹20k through a collections agency, your LGD is 60%.

The Golden Formula: Expected Loss = PD × LGD × EAD (For a loan with PD 5%, LGD 60%, and EAD ₹35k: Expected Loss = 0.05 × 0.60 × 35,000 = ₹1,050. The bank legally must set aside ₹1,050 in cash reserves for this exact loan).

Fraud Risk: A different flavor

Fraud analysis is more detective work, less financial forecasting. You are looking for anomalies.

The SQL relies heavily on aggregations, lookbacks, and pattern detection:

-- Flagging accounts with unusual transaction spikes today
WITH user_history AS (
    SELECT 
        account_id, 
        AVG(daily_txn_count) AS avg_txns_last_30_days
    FROM daily_summary
    WHERE txn_date >= CURRENT_DATE - INTERVAL '30 days'
    GROUP BY account_id
)
SELECT 
    t.account_id, 
    COUNT(t.txn_id) AS txns_today,
    h.avg_txns_last_30_days,
    CASE WHEN COUNT(t.txn_id) > (5 * h.avg_txns_last_30_days) 
         THEN 'FRAUD_FLAG' ELSE 'CLEAR' END AS status
FROM transactions t
JOIN user_history h ON t.account_id = h.account_id
WHERE t.txn_date = CURRENT_DATE
GROUP BY t.account_id, h.avg_txns_last_30_days;

The challenge: Flag too aggressively, and you block legitimate customers (who angrily call support). Flag too loosely, and you let fraud slip through (which costs the company hard cash). Finding that mathematical balance is your entire job.

Become an AI expert :

Become An AI Expert In Just 5 Minutes

If you’re a decision maker at your company, you need to be on the bleeding edge of, well, everything. But before you go signing up for seminars, conferences, lunch ‘n learns, and all that jazz, just know there’s a far better (and simpler) way: Subscribing to The Deep View.

This daily newsletter condenses everything you need to know about the latest and greatest AI developments into a 5-minute read. Squeeze it into your morning coffee break and before you know it, you’ll be an expert too.

Subscribe right here. It’s totally free, wildly informative, and trusted by 600,000+ readers at Google, Meta, Microsoft, and beyond.

Why this career path is bulletproof

Banks, insurance companies, and fintech firms are heavily regulated. The RBI mandates risk assessment for every loan. These aren't optional vanity projects—they are legal requirements to stay in business.

Which means these companies will always hire Risk Analysts. The economy can slow down, marketing budgets can get slashed, but risk and compliance teams are the last to be touched.

In India specifically, HDFC, ICICI, Bajaj Finance, PhonePe, Razorpay, and CRED all have massive risk analytics teams, and they hire freshers at scale.

The Interview Question They Will Ask You

"If we lower our approval threshold to accept more loan applications, what are the risks? How would you monitor the impact?"

Weak answer: "We should use Python to predict the defaults." (Too vague. They want to hear you think about measurement, not just models).

Strong answer: "The immediate risk is a spike in default rates from the newly approved segment. I would monitor this by creating a separate cohort for the new 'lower-threshold' loans and tracking their DPD (Days Past Due) at 30, 60, and 90 days. I would also recommend setting a 'circuit breaker'—if the new cohort's 60-day DPD exceeds X%, we automatically tighten the threshold back to the original level."

See you next week.
—RAJ (Real Analyst Jobs)

P.S. Added new 25+ fresh remote jobs: realanalystjobs.com
Added new free projects : https://realanalystjobs.com/projects
Document your 2026 Journey: https://realanalystjobs.com/journey
Talk to me : https://realanalystjobs.com/raj