How credit scores work and what moves them

A credit score is the output of a statistical model. It reads the data in one credit file and returns a number that ranks you against everyone else by the odds of a serious delinquency, usually defined as falling about 90 days behind on something within the next couple of years. It is not a measure of wealth, of character, or of how much a lender is willing to give you.

Two structural facts explain most of the confusion about it.

There is no single score. FICO and VantageScore are separate companies, each ships several model versions, and a version stays in use for years after a newer one exists. A mortgage underwriter, a card issuer and a car dealer often pull different models of you on the same afternoon. The free number in a banking app is usually a different model again.

There is no single file. In the United States, Equifax, Experian and TransUnion each hold their own record, and a creditor that reports to two of them leaves the third stale. So the same model run on three files gives three answers.

The scale most United States consumer models use runs from 300 to 850, with higher meaning lower predicted risk, though older and industry-specific versions have used other ranges. The cutoffs matter more than the scale, and lenders set those themselves and move them as credit conditions change. Band labels published online are a rough map, not a rule. What does travel between models is the direction of each factor: behaviour that helps one score almost always helps the others.

FICO publishes an approximate breakdown of its classic model, and it is the closest thing to an official ranking anyone gets.

Those shares are averages across a whole population, and the model builder is explicit that the weight of any one factor shifts with what is in an individual file. On a thin file with two accounts, opening a third matters far more than it does on a file with twenty years of history. Other models divide the same ground differently, and some newer ones add trended data, meaning whether your balances have been rising or falling over recent months rather than just where they sit today.

The second column is the one most summaries leave out and the one worth reading. Payment history dominates the total, but nothing you do this month changes what happened last year. Utilisation is the second largest factor and the only large one that responds to an action taken this week. That combination, large weight plus fast response, is why every practical answer to "how do I raise this number" starts in the same place.

A payment a few days past its due date is a late fee and an internal problem. It is not yet a credit event. In the United States, lenders report delinquency in 30 day buckets, so the first thing that reaches the file is a 30 day late, then 60, then 90 and beyond.

That first mark is the expensive one. Among routine events the largest single drop on most files is the step from never late to once late, and further marks on the same account add less than the first did. Bankruptcies, foreclosures and charge-offs are heavier still, so the first missed payment is the most damaging thing that happens to an otherwise ordinary file rather than the most damaging thing that can happen to any file. Three things then modify the damage: severity, meaning how far behind, recency, meaning how long ago, and frequency, meaning how many accounts and how often. A single 30 day late from four years ago is still on the file but carries a fraction of the weight of one from last month, because the models are built on how well recent behaviour predicts the near future.

Beyond the buckets sit the serious entries: charge-offs, accounts sold to a collection agency, repossessions, foreclosures and bankruptcies. In the United States the Fair Credit Reporting Act caps how long most of these may be reported, with bankruptcy allowed to stay longer than an ordinary delinquency, and the clock generally runs from the date the account first went bad rather than from the date you settled it. Paying a collection does not delete it, although several newer model versions ignore paid collections while older versions still in wide use do not. If an entry is wrong, dispute it with both the agency holding the file and the lender that reported it, and the obligation to verify it sits with them.

Credit utilisation, spelled utilization in American usage, is the reported balance on a revolving account divided by its credit limit. Revolving means a card or a line of credit rather than a fixed installment loan. A card is unsecured borrowing, priced on the risk this number helps measure, while a home equity line is revolving and secured, so revolving and unsecured are overlapping categories rather than the same one. Models look at utilisation two ways at once: the aggregate figure across every card, and the worst individual card. Both are scored, which is why one card near its limit can hold a number down while the total looks reasonable.

The worked examples below run the arithmetic on one file. Balances of $4,100, $970 and $430 against limits of $4,500, $3,000 and $2,500 come to $5,500 against $10,000, or 55 percent overall, while the first card alone sits at 91.11 percent of its own limit.

Utilisation also carries almost no memory. Mainstream models read the file as it stands, so a balance that was near the limit last year stops counting once a lower balance is reported. That is why it moves faster than anything else on the list: pay the balance down and the effect lands on the next report, usually within one cycle, rather than aging out over years. Payment history works the opposite way, which is the asymmetry worth carrying around. Utilisation is a photograph of today. Payment history is a recording of the last several years.

Here is the part that catches careful people. Most issuers report the statement balance, not the balance left after you pay. Clear the card in full on the due date every month and the file can still show a balance close to the limit, because the snapshot was taken when the statement closed. Someone who never pays a cent of interest can still look, on paper, like someone running hard against their limits. Paying before the statement date, or making a payment mid-cycle, changes what gets reported without changing what you owe or what it costs you.

Two other mechanics move the same ratio without a payment. A higher limit lowers it outright, which is why a limit increase can help, though the request itself may be recorded as a hard enquiry. And closing a card takes its limit out of the denominator, so the same balances read as a higher percentage the day the account closes, which is the strongest argument for leaving an old no-fee card open and lightly used.

The often-quoted 30 percent line is a rule of thumb rather than a cliff. Nothing snaps at 29 or 31, lower is better nearly all the way down, and a small reported balance tends to score marginally better than zero reported across every card. Treat it as a marker to aim under, not a target to sit on.

None of this changes what the debt costs you. The card's APR applies to the balance you carry, not to the ratio, so the cheapest card to clear first and the one that moves the score fastest are not always the same card. Where they disagree, the interest is the certain number and the score effect is the estimate.

Length of history counts both the age of the oldest account and the average age across all of them. Opening anything drags the average down, which is the real reason a new card can nudge a score lower for a while even when nothing has gone wrong. In the United States a closed account in good standing does not vanish from the file straight away, and it keeps contributing to age while it is there, so closing a card is mainly a utilisation decision rather than a history one.

Credit mix means the model would like evidence that you have handled more than one kind of borrowing: revolving accounts such as cards alongside installment accounts such as a car loan, a student loan or a mortgage. It is the smallest of the five factors. Taking on a loan you do not need in order to improve it is a bad trade, because the interest is certain and the score effect is small.

New credit is measured through hard enquiries, recorded when you apply for something. Checking your own report or score is a soft enquiry, is not visible to lenders in the same way, and is not scored at all. In the United States models, hard enquiries fade quickly: they carry weight for a matter of months and stop counting well before they drop off the report entirely. Rate shopping is handled specially, because otherwise comparing five mortgage lenders would look like applying for five mortgages. Enquiries of the same type inside a short window are collapsed into a single event. The length of that window depends on the model version, and other countries handle repeated applications by their own rules, so the practical point is that a tight cluster of shopping is treated differently from the same applications spread across a season.

The score answers one question: how has this person handled credit in the past? Debt-to-income answers a different one: can this person afford the payment now? Your income is not in the credit file, so no credit bureau score can read it. That is narrower than the usual shorthand. One of the United States agencies also runs a separate payroll database that lenders buy employment and income verification from, and a lender's own internal application scorecard can use income directly. What none of that does is feed income back into the bureau score. A lender works the ratio out separately from your application and your documents, then applies its own ceiling. That is why the tool at the top of this page is a debt-to-income calculator: it is the half of the decision your credit report cannot tell you.

Take the same borrower from the examples. Required monthly payments of $1,450 for housing, $380 for the car loan, $220 for a student loan and $137.50 of card minimums come to $2,187.50. Against $6,000 of gross monthly income that is 36.46 percent, and a 43 percent ceiling would be $2,580, leaving $392.50 a month of room.

That 43 percent is a widely quoted United States benchmark rather than a fixed rule. It came out of one version of the mortgage regulations and has outlived it as a habit. Individual lenders and loan programmes set their own ceilings, several will go above this one where the rest of the application is strong, and the only figure that decides anything is the one the lender in front of you actually applies. It is used here because it is the number readers arrive with, not because it is a line in law.

Now watch how differently the two react to the same act. Paying $2,500 off the cards takes utilisation from 55 percent to 30 percent as soon as the new balances are reported. That changes the input the model reads at the next pull, though how far the number itself moves depends on everything else in the file, so the direction is dependable and the size is an estimate. The ratio barely moves, because it counts required payments and the minimum only falls as the balance does. Run it in reverse and the asymmetry is sharper: a new car loan takes its bite out of the room in the ratio in full on day one, while on the score it usually shows up as a modest dip from the enquiry and the drag on average account age, and that dip fades over the following months. The loan payment calculator shows what a given loan size and term does to that monthly figure.

They also fail differently. A thin file with nothing missed can produce an unimpressive score alongside a spotless affordability picture. A high earner with a recent default has the opposite. Underwriting looks at both, plus the deposit, the reserves and the employment history, which is why neither number on its own is an approval.

Everything above describes one system: three national agencies, private model vendors, and a scale a consumer can look up. Other countries are built differently, and United States rules of thumb give wrong answers when applied to them.

• United Kingdom. Experian, Equifax and TransUnion each publish a number on its own scale, so a figure that looks strong at one agency has no fixed translation at another. Lenders mostly decide with their own internal models, built on their own customer history plus the file and the application, so the consumer-facing number is an indicator rather than the thing being used.
• Canada. Closest to the United States pattern, with two agencies and a scale that looks similar without being the same one, so a number carried across the border does not mean what it appears to. Reporting rules and the handling of enquiries differ in detail as well.
• Australia. Files carry repayment history month by month as well as defaults, and each agency scores on its own range.
• France. There is no consumer credit score at all. The central bank keeps a register of payment incidents, and lenders assess affordability from documents instead.
• Germany. A private agency scores individuals using account and contract data as well as defaults, and states its headline result as a probability of repayment rather than as a points total on a consumer scale, with separate scores for different industries sitting underneath it.

One pattern is worth knowing before you move country: where a file records only defaults and judgments, years of paying on time earn you nothing visible, and arriving with no local file at all can look the same as arriving with a bad one. Two habits do generalise. Pay on time, and do not run balances hard against limits. And check your own file wherever you live, because it is free or close to it and it is the only way to see the record a lender will be working from. It is not the whole of what the lender sees. The model they run it through, and any history you already have with them, stay on their side of the counter.