How is a CIBIL score calculated? From what lenders report, by unpublished weights.

The short answer

How is a CIBIL score calculated? From the data lenders send the bureau about your loans, cards and applications, run through a scoring model that the bureau owns and does not publish. The factors are known, and they are the same at every bureau: payment history, how much of your limits you use, how long you have held credit, the mix of account types, and recent applications. The weights are not known, the output is a number between 300 and 900, and the only thing that changes it is a change in what gets reported.

Start with the data, because the data is the whole input.

A bureau knows nothing about you that a lender did not tell it. Under the RBI's directions on credit information reporting, every bank, NBFC and card issuer must report each credit account it holds in your name: the type, the sanctioned amount or limit, the balance, the repayment record month by month as days past due, the status (standard, written-off, settled, closed), and every enquiry it made when you applied. Since 1 July 2026 that file goes to all four bureaus as of four dates a month, the 9th, 16th, 23rd and the last day. Your salary, your savings, your UPI spending and your own score checks are not in the file, so they cannot be in the score.

The five factors

What the bureaus say the model reads.

CRIF High Mark lists five factors in its FAQ, and the other bureaus describe the same family in their own words. In the order every bureau puts them:

• Payment history. Each account's record of payments made on time or late, and by how many days, shown on the report as the days-past-due grid. A single 30-day late is a mark; a run of 60s and 90s is a pattern; a written-off or settled status is the heaviest entry of all. This is the factor every bureau ranks first.
• Credit utilisation. Your card balances as reported, divided by your total limits. A ratio that stays high reads as dependence on credit. CRIF High Mark's own advice is to keep spending within 30 to 35% of the limit; the ratio the bureau sees is the one on the lender's reporting date, which is why paying before the statement closes matters more than paying on the due date.
• Length of credit history. How long your oldest account has been open and how long your accounts have been reported. A long, clean history is evidence; a new file is a guess. This is why closing your oldest card can cost more than it saves.
• Credit mix. The spread of secured loans, unsecured loans and cards. A file that shows you have handled more than one kind reads as experience. It is the minor factor, and the one most misused by advice that tells people to open a loan they do not need.
• New credit. The enquiries lenders make when you apply, and the accounts opened recently. Several applications in a short period read as distress. Your own score checks are consumer enquiries and do not count; what enquiries really cost is its own guide.

The missing formula

Why the percentages you see online are guesses.

Search for a CIBIL score calculation formula and you will find confident breakdowns: payment history 30%, or 35%; utilisation 25%, or 30%. They disagree because none of them comes from a bureau. CRIF High Mark lists its factors without weights, and no bureau page we could read publishes one; the breakdowns are borrowed from descriptions of American scoring models and repeated. The honest statement is the one the bureaus make themselves: these are the factors, payment history matters most, and the model is proprietary. Any page that gives you an exact percentage is telling you what it assumes, not what the bureau does.

Four numbers

Why CIBIL, Experian, Equifax and CRIF give you different scores.

Each bureau holds its own copy of the lenders' files and runs its own model, so the same facts produce four numbers. They usually sit within a few points of each other because the RBI requires every bureau to calibrate on the same 300 to 900 range and the inputs are the same. The gaps come from timing, a lender's file reaching one bureau before another, from a correction applied at one bureau and not yet at the next, and from the models themselves. A large gap is information: it usually means one bureau is carrying an entry the others are not, and that entry is the thing to read first. Credit bureaus in India compares the four.

The quiz question

"How to calculate CIBIL score from the data source": you cannot, and you need not.

The phrase turns up in search because it is set as an exercise, and the correct answer is that the computation is the bureau's. What you can do with the data source, your own report, is more useful: read the days-past-due grid under every account, check each balance against its limit, confirm every account is yours and every closed account is shown closed, and count the enquiries. Those fields are the model's entire input. How to read your CIBIL report walks through them section by section.

What actually moves it

The number follows the record. Change the record.

Because the score is a function of the reported data, it moves only when the data moves: a payment reported on time, a balance reported lower, an enquiry ageing out, or a wrong entry corrected. In our published cases the last of those is the fast one. A 768 that fell to 727 with zero missed payments was restored in seven days once the entry that caused it was corrected; a CRIF score of 510 reached 754 when four accounts that were never his came off. No one, including this firm, can add points to a number the model computes; CIBIL correction changes the inputs, and the model does the rest. What can and cannot be fixed says which inputs are correctable.

Documented cases

Documented on this topic.

Real cases from our files, published with the clients' consent. Every case depends on its facts; none is a promise of the same result.

• 768 to 727 with zero missed payments: the wrong entry found and the score restored in seven days
• CRIF rejected his dispute; the right escalation removed the wrong accounts anyway, 510 to 754
• Five Shriram Finance loans on his report, one was his: four removed, 742 to 818

Quick answers

Answered first.