What a credit score actually is — and how the number gets made

Personal Money 3 min 80 sources

A credit score is a single number that predicts how likely you are to repay. Five things build it, and two of them decide most of it.

Key takeaways

A credit score is a 300–850 number that predicts how likely you are to repay, and it sets whether you're approved and at what interest rate.
Two factors decide most of it: paying on time (about 35%) and keeping balances low against your limits (about 30%).
Checking your own score never lowers it, and one late payment fades over time rather than haunting you for seven years.

You apply for a card, a loan, or a flat, and someone you’ve never met pulls a three-digit number and decides yes or no — and at what price. That number is your credit score. Most people know it matters and have no idea how it’s built. Here is the machine.

A credit score is a number, usually between 300 and 850, that predicts how likely you are to pay back money on time [1][6]. Lenders use it to decide whether to offer you a mortgage or a credit card, what interest rate to charge, and how big a credit limit to give [5]. FICO is the most common brand of score; “FICO score” and “credit score” get used as if they’re the same thing, though there are several versions, each weighing your file slightly differently [5][15].

The score is built from the information in your credit report — a record, kept by reporting companies, of how you’ve borrowed and repaid [4]. Five categories feed the FICO model, and they don’t count equally. Payment history — whether you’ve paid past bills on time — is the biggest, at about 35% of the score [14]. Amounts owed, which is mostly how much of your available credit you’re using, is next at about 30% [8]. The remaining three are smaller: length of credit history (about 15%), credit mix — the different types of credit you hold (about 10%), and new credit, meaning recent applications (about 10%) [21][10].

That weighting carries a plain lesson. Pay on time and keep your balances low relative to your limits, and you’ve handled roughly two-thirds of what the score cares about [14][8].

The second factor — amounts owed — runs on a number called the credit utilization ratio: the share of your available credit you’re actually using [1]. If you have a £2,000 limit and a £600 balance, your utilisation is 30%. Lower is generally read as healthier; high utilisation signals you may be stretched [3]. It’s calculated across your cards, so a single maxed-out card can drag the whole picture.

Two things people get wrong cost them. First, checking your own score does not lower it. When you look at your own report it’s a “soft inquiry,” which never affects the score [12]. Only a “hard inquiry” — when a lender pulls your file after you apply for credit — can ding it, because scoring models watch how often and how recently you apply [12]. Second, a single late payment is not a seven-year sentence. The late mark can stay on your report for years, but its drag on the score fades as it ages and you keep paying on time [14].

You can see all of this for yourself. You’re entitled to free copies of your credit reports from the major reporting companies, and checking them is the soft-inquiry kind that costs you nothing [16][19]. The score is not a verdict on you as a person — it’s a prediction built from a handful of behaviours, most of which you control.

02 · Lesson · why it matters

A credit score isn't a grade — it's a bet on your next twelve months

The number doesn't judge your past; it predicts your future, and a few behaviours carry almost all the prediction.

The number is answering one question

A credit score feels like a report card. It isn’t. A report card grades what you did. A credit score does something stranger: it predicts what you’ll do next [6].

Every input exists for one reason — to forecast whether you’ll repay on time in the months ahead [1]. The lender doesn’t care about your character. They care about a probability. The whole machine is built to turn your borrowing history into one number that answers: if we lend to this person, how likely are they to pay us back? [5]

Once you see it as a prediction, the rules stop seeming arbitrary.

Why two factors carry most of the weight

The five things that build a FICO score don’t count equally, and the imbalance is the whole point. Payment history is about 35%. Amounts owed is about 30%. The other three — length of history, credit mix, new credit — split the rest [21][10].

That’s not a random split. It’s a ranking of how strongly each thing predicts repayment. The best forecast of whether you’ll pay next month is whether you paid last month [14]. So payment history dominates. The second-best signal is how stretched you already are — someone using most of their available credit is closer to the edge than someone using a little [3]. So amounts owed comes next.

The model weights each input by its predictive power. The factors you’d expect to matter — do you pay, are you overextended — are exactly the ones that do.

The pattern: the system rewards the behaviour it’s trying to measure

Here’s the part worth carrying. A prediction system isn’t neutral. The moment people know what it measures, they change what they do — and a good system is designed so that gaming it and improving are the same move.

You can’t fake a long history of on-time payments. The only way to score well on the 35% is to actually pay on time, month after month [14]. The only way to score well on the 30% is to genuinely not be overextended — keep your balances low against your limits [8]. There’s no trick that satisfies the score without also making you a safer borrower in real life.

This is why “credit score hacks” mostly disappoint. The score is hard to fool because it measures the thing it claims to measure. Improve the underlying behaviour and the number follows; try to shortcut the number and you find there’s no shortcut.

Where the prediction breaks from intuition

Two things trip people up, and both make sense once you remember the system is forecasting, not grading.

Checking your own score doesn’t lower it. Your own look is a “soft inquiry” — it carries no information about future borrowing, so the model ignores it [12]. A lender’s pull after you apply is a “hard inquiry,” and that does carry a signal: applying for credit is a sign you might take on debt, and applying a lot in a short window is a sign you might be desperate for it [12]. The system reacts to the second and shrugs at the first, because only one predicts anything.

A single late payment also isn’t a seven-year wound. The mark can sit on your report for years, but its weight in the prediction fades as it ages [14]. A forecast of your next twelve months cares far more about what you did last month than about one slip three years ago. Recent behaviour predicts; ancient behaviour barely does.

What you actually control

See the score as a bet and your leverage gets clear. You can’t argue with a probability. But you set the inputs the probability is built from.

Pay on time, and the strongest predictor moves your way. Keep balances low against your limits, and the second-strongest does too [8][14]. Don’t apply for credit you don’t need, and you stop feeding the model the one signal that says “this person may be reaching” [12]. Everything else — length of history, mix of credit — accrues slowly with time and can’t be rushed [21].

The number isn’t a measure of your worth. It’s a forecast of one narrow thing, built from a few behaviours, most of them yours to shape. Understand what it’s predicting, and you stop chasing the score and start owning the inputs.

03 · Lab · your turn

Move the Inputs

Rehearse a borrower's behaviour and watch the repayment prediction move, feeling which factors actually carry the score.