How a credit score is actually calculated

What the calculation actually is

A credit score is a statistical model run over one bureau's copy of your borrowing record. Lenders send the data, the bureau assembles it into a report, and a model turns that report into a number. The scale it lands on is fixed by regulation, and so is your right to read the report behind it. How the model weights what it reads is not.

The definition of the score itself — what it predicts, and why never having borrowed does not make it good — belongs to the credit score explainer and is not repeated here. This article is about the machinery: which fields the model reads, how quickly each one can move, and which of the familiar rules about the calculation are rules at all.

Hold on to that last distinction through everything below. Two of the claims in this article come with a paragraph number attached — the common scale every bureau must score you on, and the free full report you can demand from each of them once a year. The things people most often call rules do not: the weights, the 750 cut-off, the seven-year retention period. Almost every confusing thing written about credit scores comes from someone reporting the second kind as though it were the first.

One scale, several models

The old framing — each bureau has its own range, so the numbers are not comparable — is obsolete. Every credit information company operating in India is now required to calibrate its consumer score to a common scale of 300 to 900, under RBI's Credit Information Reporting Directions 2025. It is no longer one bureau's range that the others copied.

Read what that does and does not fix. The scale is standardised: 700 means the same position on the ruler wherever you read it. The model is not. Each bureau builds its own, from its own data, and none of them publishes the formula.

So a common scale removes one source of confusion and leaves two intact. Your numbers can still differ across bureaus, for two reasons that have nothing to do with the ruler:

• Different data. Two bureaus' reports on the same person are not copies of each other. Accounts go missing, one report carries a balance a cycle staler than the other, and the same borrower can end up matched to two partial records under slightly different identity details. Whether a lender is obliged to furnish the same file to every bureau is a question about the rulebook; whether it arrives intact everywhere is a question about plumbing, and the answers are not the same. An absent account is not a neutral fact — it is a stretch of repayment history that one model never sees.
• Different models. Given identical data, two bureaus would still produce two numbers, because the weightings are proprietary and were fitted on different portfolios.

The practical consequence is worth stating plainly, because it is the one people argue with. Two bureaus can hand back numbers far enough apart to sit either side of a lender's cut-off, and neither of them is the wrong one. There is no official score. The one that matters on any given application is whichever bureau your lender happens to pull, and you do not get to choose it.

Three stages, and the lag built into all of them

Nothing about your finances reaches a score directly. It travels through three stages, and each one is a place where reality and the number can part company.

1. A lender reports. Your bank or non-banking finance company submits what it holds on you — the account, the limit, the outstanding balance, the payment status — on its reporting cycle.
2. The bureau assembles a report. Every account you hold, from every lender that reports to that bureau, plus a dated list of who has enquired about you.
3. A model scores the report. The number is a function of the report and nothing else.

Two things follow immediately, and between them they explain most of the complaints people have about their scores.

The score can only change when the report changes. Paying off a card does not move it; the lender reporting the lower balance moves it. Until that submission lands, the model is scoring a balance you no longer owe. This is not an error to dispute — it is the pipeline working as designed, and the fix is timing rather than complaint.

The model cannot see what was never reported. The report is built from what lenders submit about credit they have extended to you. Your salary, your savings balance, your investments, your rent and your UPI history do not arrive through that channel — no lender is submitting them — so on an ordinary consumer report they are not there to be scored. Whether any Indian bureau folds alternative data of that kind into a consumer score through some other route is not something we have been able to establish, and we would not assume it. The mechanical point holds either way: the number is a function of the file, not of your finances. A lender assesses income separately, at the eligibility stage, which is a different calculation with different inputs and is the reason a high score and a rejected application are perfectly compatible.

The five inputs, and the speed each one moves at

Every bureau describes broadly the same five families of input, in broadly the same order of importance. What none of them publishes is the weights. The precise percentages that circulate — payment history 35%, utilisation 30%, and so on — are not an Indian bureau's disclosure, and quoting them as though they were is the commonest error in writing on this topic.

The ordering is documented. The arithmetic behind it is not. What is far more useful than a guessed percentage, and is genuinely derivable from how the report is structured, is how fast each input can move.

Set the second column against the third and the interesting property appears. These five inputs are not five dials on the same machine. They run on completely different clocks.

The fast lever and the slow lever

Here is the part that changes what a person does next.

The asymmetry is in the data structure, not in a weight — which matters, because the weights are the part nobody outside a bureau has seen and the structure is the part you can look at on your own report.

Utilisation is a state, not a ledger. What the report carries is the balance each lender last submitted against the limit on that account — a current figure that the next submission overwrites. There is no dated utilisation series sitting alongside it the way there is for payments, so the 90% you ran last March is not a row anywhere; it is simply a number that used to be there and is not any more. Fix the reported balance and the input the model reads is fixed, in one cycle. (Some overseas bureaus sell lenders a trended view built from past balances. Whether any Indian model reads one is not something we can confirm, so treat this as the structure of the report rather than as a promise about the model.)

Payment history is the opposite. It is a ledger. Every late payment is a dated row that stays, and a run of on-time months does not delete anything — it dilutes. There is no action available that removes a correctly reported delinquency. You can only add clean months next to it and wait for the arithmetic of proportion to do its slow work.

Now the mistake, which is extremely common and completely rational given what people are told. Someone whose score is being dragged down by a missed payment eighteen months ago goes looking for a lever, finds the utilisation advice, clears their cards, and waits for a jump that does not come. The lever they pulled was real — it just was not connected to the thing that was wrong.

The same asymmetry explains why the score is slow to build and fast to damage. The input that accumulates is the heavy one; the input you can change quickly is the light one.

What a late payment physically is

“Payment history” sounds like a judgement. It is a data structure.

For each account, the report holds a period-by-period record of how far behind you were — nothing owed, or a number of days past due. The model reads three things off it: how recently you were late, how late you got, and how often it happened. A single 30-day mark from three years ago and a 90-day mark from last quarter are not the same event scored twice; they are different values on all three dimensions.

Bureaus describe recent behaviour as weighing more heavily than old behaviour, which is mechanically sensible — the model is predicting the near future and last quarter is better evidence than 2022. How steeply the weight decays is not published, so anyone telling you a delinquency stops mattering after a specific number of months is describing a model they have not seen.

Two statuses sit outside the days-past-due ladder and are worth recognising in your own report, because they are read as outcomes rather than as lateness: an account marked settled, where the lender accepted less than the full amount, and one marked written off. Neither is a count of days late, so neither improves with time the way a stale delinquency does — they describe how the account ended. The pillar works through the settled-versus-closed distinction, and why paying the full outstanding is usually worth more than the amount a settlement waives.

Utilisation, and what the denominator does

Utilisation is the reported balance divided by the limit, and it is measured both on each card individually and across all of them together. The trap in it — that banks generally report the statement balance, so paying in full every month can still show heavy usage — is covered in the pillar and is the first thing to check if your score seems unfairly low.

What is worth adding here is the denominator, because it moves in ways people do not expect. Your limit is not a constant. An issuer can raise it unasked, which lowers your ratio without you doing anything, and can cut it — which raises your ratio without you spending a rupee. A ₹40,000 balance on a ₹1 lakh limit is 40%; the same balance after the limit is trimmed to ₹60,000 is 67%, on identical behaviour.

The same arithmetic is why closing an unused card can hurt. The balance stays where it is and the limit leaves the calculation, so the ratio rises on what remains. It is also the tension that sits underneath a consolidation: clearing card balances with a loan drops the reported utilisation to near zero and hands back the full limit on the same day. What it does not do is reduce what you owe — the debt has moved, not gone, and the restored limit is the thing that gets spent again. The score improves on the day the balances are reported as nil; whether the borrower is better off depends entirely on what happens to that limit afterwards.

One thing utilisation is not: a measure of whether you pay interest. A person who clears the statement in full and a person who revolves the balance at a punishing rate can report the same ratio. The model is reading dependence on available credit, not the cost of it.

Enquiries, and the one thing the model cannot see

When you apply for credit, the lender pulls your report and that pull is recorded, with a date and the lender's name. That is a hard enquiry. When you pull your own report, the access is logged as your own — a soft enquiry — and is not read as an application. Checking your own report as often as you like costs your score nothing. Screening for pre-approved offers is generally described as soft as well, but that is the industry's account of its own process rather than something we can point at a rule for — worth knowing, not worth relying on.

The mechanism that makes hard enquiries matter is a gap in the data, not a punishment. The report records that a lender looked. It does not record what the lender decided.

So five applications in three weeks look identical whether you were approved five times or refused five times. The model cannot separate the two, and the conventional reading — which follows from what it can see rather than from anything a bureau has published — is that a burst of enquiries is more often somebody being turned down repeatedly than somebody shopping carefully.

Which produces a named failure mode worth avoiding deliberately — call it the rejection spiral. An application is declined for a reason that has nothing to do with the score, perhaps income documentation. The applicant tries a second lender, then a third. Each attempt adds an enquiry, the enquiries themselves make the next lender warier, and the original problem — the documentation — is still unfixed. Every step is reasonable and the sequence makes things worse.

The way out is to find out why the first refusal happened before making the second application. Some overseas scoring models de-duplicate several enquiries for the same product inside a short window, so that comparing home loan offers is not penalised. Whether any Indian bureau's model does that is not something we have been able to establish, and it would be unwise to assume it.

How long things stay — and where the seven years came from

We could not find the seven-year figure in any rule. “Defaults stay on your report for seven years” is repeated across Indian personal finance writing as though it were statute, and it is not in RBI's Credit Information Reporting Directions or in the governing Act.

No retention period for adverse information appears in the material we checked, which leaves two possibilities: either the period is a bureau display policy that everyone has been quoting as law, or it sits in an instrument we have not found. Both are live, so we are not going to state a number, and neither should anyone who has not read the provision they are citing. If you see the figure quoted again, the useful question is which paragraph of which instrument it comes from.

What can be said without a citation problem is structural. A closed account does not vanish from the report — it remains, with its opening date, and continues to contribute to the age of your credit history. That is why closing an old card is worse than leaving it dormant: closure removes the limit from the utilisation denominator immediately, while the history it built keeps counting for as long as the record is displayed.

If a specific line on your report matters to you — a settlement, a write-off, an old delinquency — the answerable question is not “when does it drop off?” but “is it accurate?” An inaccurate entry can be disputed with the bureau, which takes it up with the lender that reported it. An accurate one is not going anywhere on request, and no service that says otherwise is describing something a rule permits.

The free reports, and how to spend them

The entitlement is one free full credit report a calendar year from each credit information company, and each bureau must carry the link on its own homepage. The load-bearing words there are from each: this is a right against every bureau separately, so what you hold in a year is one full report per bureau rather than one in total. It is the only thing in this article that a rule hands you outright, and almost nobody uses it.

Most people use none of them, and the few who do use them together in January. Spread them instead: pull one bureau, then the next a few months later, and so on through the year. The cost is the same and the coverage is continuous — an error introduced in March surfaces in the spring pull rather than the following January. What you give up is the one thing the January habit is good for, which is seeing every bureau's version of you side by side on the same day; if you are about to apply for a large loan, that comparison is worth more than the calendar coverage, and it is the one moment to spend them together.

There is a second reason to spread them across bureaus rather than repeatedly checking the same one. Reports diverge, so an error can sit on one and not another, and a report you never pull is one you cannot correct.

What to read when it arrives, in order: the list of accounts, checking that each is yours and that closed ones say closed; the reported balance and limit on each card; the payment record for anything marked late; and the enquiry list, where an application you did not make is worth taking seriously.

Note the word full. A free score on an app is not the same product as the full report the rule entitles you to — the score is the output, and the report is the input you can actually check.

Working out what moved your number

Since no bureau publishes its weights, there is no formula to substitute into. There is still a method, and it is the one an analyst would use on any black box: difference two of your own reports.

Pull the same bureau twice, a few months apart, and lay the reports side by side. List every field that changed — a balance, a new account, a closed one, an enquiry, a payment status. Whatever moved the score is in that list, because nothing else was different. It is a smaller list than people expect, usually two or three lines.

This will not give you the weights, and it should not be dressed up as though it did. What it gives you is attribution: this change, on this report, alongside that movement in the number. Do it across a few periods and the ordering of the inputs becomes visible from your own data instead of from a percentage someone copied from another country's scoring model.

One discipline makes it work and its absence makes it useless: change one thing at a time where you can. Clearing three cards, closing a fourth and applying for a loan in the same month produces a score movement that cannot be attributed to anything.

What the number is for

A closing distinction, because it is the one that decides how much any of this matters.

No rule sets a score at which a lender must approve you. There is no regulatory threshold, no floor below which lending is prohibited, no figure that entitles anyone to anything. The familiar cut-offs — the 750 that appears in every article — are individual lenders' credit policies, and they differ by lender, by product, and by how much risk that lender wants this quarter. Treating a lender convention as a regulatory line is the same category error as treating a bureau's weightings as published rules.

What the score does reliably is set your price. Most lenders quote a rate that varies with how risky the applicant looks rather than one rate for everyone — risk-based pricing — so the same loan is offered at a lower rate to the stronger score, and on a long-tenure debt that difference is not cosmetic. The amortisation arithmetic shows why: interest is charged on the outstanding balance every month for twenty years, so a small difference in the rate compounds into a very large difference in the total, in exactly the way compounding works for you when investing and against you when borrowing.

That is the honest case for caring about the calculation. Not a number to collect, but a price you are quoted on borrowing you were going to do anyway.

FNOTrader is not a lender, a credit bureau or a credit adviser, and nothing here is a recommendation to take, refinance or repay any particular debt.

Common questions

Continue reading

More in Loans & Credit · App: Mutual Funds · Definitions: glossary · Free tools: calculators · All: every article