Ask a corporate credit officer and a retail risk analyst how they judge a borrower, and you will hear two very different answers. The corporate officer will talk about a credit rating model, audited financials and a long meeting with the CFO. The retail analyst, on the other hand, will talk about score cut-offs, bureau pulls and approval rates. Yet both are trying to answer the same question: how likely is this borrower to default?
The difference lies in who the borrower is. A rating assesses a company, a bank or a government, where each file is large and each borrower is different. A score, by contrast, sorts thousands of individuals applying for credit cards, personal loans or two-wheeler loans. As a result, rating leans on analysis and judgement, while scoring leans on statistics and automation.
Credit rating model vs credit scoring: how borrower type, data, method and output differ
You need to understand both, because most banks and NBFCs run them side by side. Both also feed the same decisions on pricing, limits, provisioning and capital. And if you are preparing for a credit risk modeling role, expect the credit rating vs credit score question in interviews.
A credit rating model places a borrower on one of a fixed set of ordered risk grades. Think of it as a ladder: the higher the rung, the lower the chance of default. However, the phrase covers two different things, so it helps to separate them early.
External ratings are opinions that credit rating agencies publish on issuers and their debt. In India, the largest agencies include CRISIL Ratings, ICRA, CARE Ratings and India Ratings. Globally, S&P Global Ratings, Moody’s and Fitch lead the market. These agencies rate bonds, commercial paper and bank loan facilities, as well as the issuers behind them.
Indian agencies rate long-term debt on a scale from AAA, the highest safety, down to D, which means default. Anything rated BBB- or above is investment grade. These grades also matter well beyond bond markets. Under the Standardised Approach, for instance, a bank’s risk weight on a rated corporate loan depends on that rating.
An internal rating model is a bank’s own grading system for corporate, SME and institutional borrowers. Banks need one because many of their borrowers have no agency rating at all. For example, a small auto-component maker in Pune may borrow from two banks for years without ever approaching an agency.
These models blend numbers with judgement. On the quantitative side, they use financial ratios such as leverage, interest coverage, debt service coverage (DSCR), liquidity and profitability. The qualitative side, meanwhile, looks at management quality, industry outlook, market position, group support and how the account has behaved with the bank.
For corporate, sovereign and bank exposures, Basel’s IRB rules require at least seven grades for performing borrowers and one for defaulted borrowers. Each grade then links to a PD band through a master scale, a single table that every business line shares. That is why “Grade 4” means the same risk for a hospital as for a cement plant.
Timing matters too. Agencies aim for through-the-cycle ratings that stay fairly stable through booms and slowdowns. IFRS 9 models, by contrast, need point-in-time PDs that move with the economy.
Judgement sits on top of all this. A credit officer can override the grade a credit rating model produces, although usually only within set limits. That flexibility is useful, but only with documented reasons and regular review of override rates. Otherwise, overrides slowly erode the model’s ability to tell good borrowers from bad ones.
Credit scoring turns a borrower’s characteristics into a single number. A credit scoring model learns from past repayment data which traits go with default, and then ranks new applicants on that basis. Because there is little case-by-case judgement involved, lenders use it where volumes are high and loan sizes are small.
The credit scorecard is the workhorse of retail lending. An application scorecard assesses a new applicant at the point of origination. Typical inputs include age, income, employment type, existing obligations and bureau history.
A behavioural scorecard, in contrast, looks at customers the lender already has. It uses internal account data such as limit utilisation, payment patterns and days past due. Banks rely on it for limit increases, cross-sell offers, collections priority and early-warning alerts. Some lenders also build collection scorecards to predict which overdue accounts are likely to pay up.
Credit bureaus build generic scores from repayment data that lenders report across the market. In India, the TransUnion CIBIL score runs from 300 to 900, and a higher score means lower risk. Experian, Equifax and CRIF High Mark also publish scores, while FICO scores play a similar role in the United States.
Most lenders use the bureau score as one input into their own application scorecard, rather than as a replacement for it. After all, the bureau only sees how a person has repaid across the market. Only the lender’s own data reflects its product, its pricing and its customer mix.
Scoring dominates credit cards, personal loans, consumer durable finance, two-wheeler loans and microfinance. For example, a digital lender can approve a small personal loan in minutes, largely on a score cut-off. Scores also support risk-based pricing. A lender might offer its best rate above one score, decline applicants below another, and send the band in between to manual review.
The table below shows how the two approaches compare across seven dimensions.
| Dimension | Credit Rating Model | Credit Scoring Model |
| Borrower segment | Large corporates, mid-corporates, SMEs, banks, NBFCs, sovereigns | Individuals, micro-businesses and other high-volume retail segments |
| Data sources | Audited financials, projections, industry data, management meetings, external ratings | Application data, bureau records, internal account behaviour |
| Methodology | Ratio analysis plus qualitative assessment; weights set by experts or estimated statistically | Statistical models, typically logistic regression on WoE-binned variables |
| Output | Ordinal grade (e.g., AAA to D, or Grade 1 to 10) mapped to PD via a master scale | Numerical score (e.g., 300 to 900) mapped to odds or PD |
| Update frequency | Annual review plus event-driven re-rating | At application, then monthly behavioural refresh |
| Level of judgement | High; overrides and committee approval are common | Low; mostly automated, with limited overrides |
| Regulatory use | Standardised Approach risk weights (external); IRB PD (internal); IFRS 9 SICR triggers | Retail IRB pooling; IFRS 9 staging and ECL; credit policy cut-offs |
The biggest difference is depth versus breadth. A rating analyst studies one borrower closely, whereas a scorecard learns patterns across a population and applies them to each applicant. So a rating can pick up a pending lawsuit or the loss of a key customer, which no scorecard variable would capture.
Data availability also shapes the method. Statistical models need plenty of defaults to learn from, and retail books provide them. Large corporate books, however, are “low-default portfolios” with very few observed defaults. That is why corporate models rely more on financial analysis, expert weights and agency benchmarks than on pure regression.
Judgement adds flexibility, but it also adds inconsistency. For instance, two experienced officers can rate the same company a couple of notches apart. Banks therefore manage this through credit committees, override limits and annual validation. Scorecards, meanwhile, face a quieter risk: their accuracy fades as the applicant population changes.
A rating is ordinal. AA is safer than A, but the letters alone do not tell you a default probability. That is why agencies publish default studies showing how often each grade has actually defaulted. Similarly, a raw score means little until the lender maps it to odds. Both therefore need calibration before they feed ECL or capital, because a model can rank borrowers well and still get expected losses wrong.
Under Basel, external ratings set Standardised Approach risk weights for rated corporates. Internal ratings matter under the IRB approach, where approved banks use their own PD estimates. Retail scorecards, on the other hand, usually feed IRB through pools of similar exposures rather than individual grades. India is different, though. All scheduled commercial banks here currently compute credit risk capital under the Standardised Approach. Internal ratings therefore shape approvals, pricing and limits today, and they will also feed ECL staging from April 2027.
Both routes end in a PD, but they travel differently. Our guide to PD estimation methods covers estimation in depth, so here we focus on how each model is put together.
Development starts with financial ratios. First, analysts pick ratios that have historically separated strong borrowers from weak ones, such as debt-to-EBITDA, interest coverage and DSCR. They then score each ratio against sector-specific thresholds. For example, leverage of 3x may be comfortable for a regulated utility but stretched for a commodity trader.
Next come the qualitative factors. Management track record, promoter support and industry cyclicality each get a structured score. The credit rating model then weights the quantitative and qualitative blocks and combines them into one total score. Some banks set these weights by expert judgement, while others estimate them statistically where enough default history exists.
Finally, the bank calibrates scores to PD through the master scale. Calibration lines up each grade with observed default rates. For Basel purposes, banks anchor these PDs to long-run averages. With low-default portfolios, they also benchmark against agency default and transition studies.
Scorecard development follows a well-worn sequence:
Both model types need validation, although the tests differ. For a credit rating model, teams back-test default rates by grade against master-scale PDs and study override patterns. Scorecard teams, by contrast, track Gini, KS and PSI every month or quarter. In both cases, an independent team should review the model before use and at least once a year after that. Above all, the people who build a model should not be the ones who sign it off.
A single grade tells you about one borrower. Thousands of grades, however, tell you which way the whole book is moving.
Rating migration tracks how borrowers move between grades over time. A transition matrix sums up those moves as one-year probabilities, as in the illustrative example below.
Illustrative one-year rating transition matrix for a credit rating model portfolio
Read it row by row. Of the borrowers who start the year in Grade B, for example, 85% stay there, 9% slip to Grade C and 2% default. When the downgrade share starts creeping up, the portfolio is weakening, often well before defaults show up.
Agency data also shows how stable top grades normally are. According to the Crisil Ratings default and transition study (up to fiscal 2026), one-year stability rates for Crisil AAA and AA have consistently exceeded 98% and 96%. That is exactly why the IL&FS slide, covered below, came as such a shock.
Banks roll individual grades up into an exposure-weighted portfolio PD. They also slice exposure by grade, sector and business group to spot concentration. Limits then cap how much the bank can lend to weaker grades, single sectors or single groups. Similarly, retail books get the same scrutiny through score-band distributions and vintage curves.
System-wide asset quality looks healthy right now. RBI’s Financial Stability Report (June 2026) puts the gross NPA ratio of scheduled commercial banks at 1.8% in March 2026, a multi-decadal low. Yet the same report’s severe stress scenarios take that ratio to 3.8–4.1% by March 2028. In other words, migration data is how a bank sees that kind of turn coming.
Under the IRB approach, a bank’s internal PD drives risk-weighted assets (RWA), alongside LGD, EAD and maturity. Our analysis of LGD recovery rates covers the loss side of that formula. The Basel Committee also tightened this framework in its December 2017 Basel III reforms. In particular, it removed the advanced IRB option for corporates with annual revenue above €500 million. It also raised the corporate PD floor from 0.03% to 0.05%.
Grades drive provisioning too. Under IFRS 9 and Ind AS 109, a significant increase in credit risk (SICR) moves a loan from Stage 1 to Stage 2. The loan then carries lifetime ECL instead of 12-month ECL. In practice, a multi-notch downgrade is one of the most common SICR triggers, alongside the 30 days past due backstop. NBFCs covered by the Ind AS roadmap already report under Ind AS 109. Meanwhile, RBI’s final ECL Directions (April 2026) bring staging to commercial banks from 1 April 2027. Crisil Ratings estimates the shift will have a one-time net impact of up to 120 basis points on banks’ CET1 ratios.
IL&FS shows how fast a rating event can travel from one borrower into whole portfolios. Infrastructure Leasing & Financial Services (IL&FS) sat at the top of a large infrastructure development and finance group. As of March 2018, the group owed more than ₹91,000 crore.
Until August 2018, ICRA rated IL&FS’s loans and debentures AAA. That month, it trimmed the rating one notch to AA+. Group entities then delayed payments, and on 8 September 2018 ICRA cut the long-term rating to BB. It also cut the commercial paper rating from A1+ to A4, and CARE Ratings downgraded IL&FS group papers too. Finally, on 17 September, ICRA moved IL&FS to D after it missed obligations due on 14 September.
The fallout came quickly. Debt mutual funds holding IL&FS paper marked down their NAVs, while banks and insurers held the largest exposures to the group. On 1 October 2018, the NCLT allowed the government to replace the IL&FS board. Later, in December 2018, SEBI allowed mutual funds to side-pocket debt hit by a credit event.
What should a risk team take from this? For a start, the rating fell from AA+ to D in nine days, which left little time for anyone relying on external grades. An internal credit rating model therefore needs to capture liquidity and refinancing risk, not only leverage. Migration data also has to reach limits and staging quickly. Finally, exposure to any single business group needs firm limits, however strong its rating looks.
The right tool depends on the segment, the data and the number of decisions. A credit rating model fits when borrowers are few, large and complex, and defaults are rare. Scoring, by contrast, fits when applicants are many, products are standard and decisions must be quick. Cost matters as well. For instance, a full rating for a ₹20 lakh loan rarely justifies the analyst hours. Equally, deciding a ₹500 crore corporate loan on a score alone would miss risks that only analysis reveals.
Most banks therefore end up using both. SME lending is often a hybrid: a scorecard for small-ticket loans and a full internal rating above a size threshold. The bureau scores of promoters and guarantors also feed into SME ratings. Ultimately, the master scale holds it all together, so a score band and a corporate grade can speak the same PD language.
Credit rating and credit scoring attack the same problem from opposite ends. A credit rating model goes deep on one borrower and leans on analysis and judgement. Scoring, on the other hand, goes wide across a population and leans on statistics and automation. Still, both end up as a PD that drives pricing, limits, provisioning and capital.
For analysts in India, the timing matters. From April 2027, ECL staging will tie rating migration and scorecard outputs directly to bank provisions. IL&FS is also a reminder that even top grades can fall quickly, so monitoring matters as much as the first assessment. If you can build, validate and explain both approaches, you will be well placed for credit risk modeling roles.
Explore Dexlab Analytics’ Credit Risk Modeling certification program to build PD, LGD, and EAD models from scratch, work through IFRS 9 ECL frameworks, and learn model validation techniques used by practicing risk teams.
.
Credit and Market Risk, Credit Risk, credit risk analysis, Credit Risk Analytics And Modeling, credit risk analytics training, Credit risk certification India, Credit risk modeling course India, Credit Risk Modelling, Credit Risk Modelling Using SAS, Credit scorecard modeling, Expected credit loss ECL, IFRS 9 modeling course India, Logistic regression scorecard, PD LGD EAD modeling, Python credit risk modeling
Comments are closed here.