The score that predicts your future, not your past

Personal Money 3 min 80 sources

A credit score isn't a reward for good behaviour — it's a lender's guess at how reliably you'll repay money you haven't borrowed yet. That one fact explains why its rules feel backwards.

Key takeaways

A credit score isn't a reward for past good behaviour — it's a lender's prediction of whether you'll repay money you borrow next.
The rules feel backwards because of that: paying off a loan, closing a card, or never borrowing can all lower a number built to predict borrowing.
It's a model you can't see, run by companies you didn't choose, that gates your rates, your rent, sometimes your job — and it's wrong often enough that you can dispute it for free.

Most people picture a credit score as a report card: pay your bills, do the responsible thing, and the number goes up. It mostly works that way — until the day you do something plainly sensible, like clearing a loan, and the number drops [20].

That surprise is the clue to what the score actually is. It is not a record of how good you’ve been with money. It is a prediction — a lender’s estimate of how likely you are to repay credit you take on next [29]. The whole thing is built to answer one question: if we lend this person money, will they pay it back? Everything that feels backwards about it follows from that.

What the number is, and what counts as “good”

The most-used score, FICO, runs from 300 to 850 [11]. Lenders read bands, not exact points: roughly 670 to 739 is “good,” 740 and up is very good to excellent, and below the mid-600s starts to limit what you’re offered and at what rate [8][13]. The score is sold to lenders, landlords, and sometimes insurers and employers as a fast read on risk [4]. You are scored constantly, by a model you can’t see, run by companies you didn’t choose.

What moves it

Five things make up a standard FICO score, and they don’t count equally. Payment history — whether you pay on time — is the biggest piece, about 35% [11]. How much of your available credit you’re using (“utilization”) is next, around 30% [2][6]. The length of your credit history is about 15%; your mix of credit types about 10%; and brand-new credit applications about 10% [11].

Utilization is the one that trips people up. It’s the share of your credit limits you’re actually using. A common rule of thumb is to keep it under 30% — owe less than £300 on a £1,000 limit [6]. Use a lot of your available credit and the model reads you as stretched, even if you pay in full every month [2]. So someone who never borrows can score worse than someone who borrows a little and repays it — because the model has nothing to predict from, and what it can see looks maxed out.

Why doing the sensible thing can hurt the number

Pay off your only loan and close the account, and you can lose history length and credit mix at once — two of the five factors gone [20][9]. The Guardian columnist who watched her score fall after clearing her mortgage hit exactly this: the model lost a long, perfectly-paid account it had been using to vouch for her [20]. Nothing about her got riskier. The machine just had less to go on.

Same with applications. Each formal credit check — a “hard inquiry” — can shave a few points and lingers on your report for up to two years, because applying for a lot of credit at once is a risk signal [16][17]. Shopping for one mortgage or car loan inside a short window usually counts as a single inquiry, so comparing rates isn’t punished [21]. But the logic is the lender’s, not yours.

The part that’s genuinely uncertain

Scores vary by country, by which of the three main bureaus is reporting, and by which scoring model a given lender uses — your number is really several slightly different numbers [13]. Errors are common: a wrong account, a payment marked late that wasn’t. You have a right to dispute these directly with the bureau, free, and they must investigate [3][25]. The model is powerful, but it is not always right about you.

02 · Lesson · why it matters

The grade that was never about your past

When a number is built to predict what you'll do next, doing the obviously-right thing can move it the wrong way — because you misread what it was measuring.

A drop that makes no sense

A woman pays off her mortgage. Years of payments, never once late. She owns her home outright. Then her credit score falls — from well above average to below it [20]. Nothing about her got riskier. She has more money and less debt than the day before. Yet the number that decides her future borrowing went down.

This feels like a glitch. It isn’t. It’s the system working exactly as designed, and the surprise comes entirely from believing it was measuring something it never was.

A report card looks backwards. This looks forwards.

We’re trained on report cards. You did the work, you get the grade. The grade is a record of the past — a reward, earned and kept. Almost everything we’re scored on early in life works this way.

A credit score does not. It is a prediction [29]. The only question it exists to answer is: if a lender hands this person money tomorrow, will it come back? Every input is chosen because it helps guess that one thing [11]. The number isn’t a verdict on your character. It’s a bet on your behaviour — and bets are about the future.

Once you see that, the “glitches” dissolve. Paying off the mortgage didn’t make the woman worse. It removed a long, spotless account the model had been leaning on to predict her — so the prediction got shakier, and a shakier prediction scores lower [20]. The person who never borrows at all scores worse than the person who borrows a little and repays it, for the same reason: the model can’t predict from nothing [2]. Use most of your available credit and it reads you as stretched, even if you clear the balance monthly [6]. None of these is a punishment. Each is the machine saying I have less reason to be confident you’ll repay.

The measure you optimise for is rarely the thing you wanted

Here is where it gets uncomfortable. Because the score gates real things — the rate on a loan, sometimes a flat, sometimes a job — people learn to feed it [4]. Keep a card open you don’t use. Carry a small balance you don’t need. Don’t pay the loan off early. These can genuinely lift the number.

But notice what’s happened. You’re no longer managing your money for your life. You’re managing it for the model’s guess about your life. The thing being optimised — the score — has quietly replaced the thing you actually cared about, which was being secure with money. Sometimes those point the same way. Sometimes, as with the early mortgage payoff, they pull apart, and you find yourself choosing between a higher number and a freer life.

You can see the grade. You can’t see the grader.

The deepest part is the one that’s easy to miss. You can check your score. You cannot see the model that makes it. The exact weightings shift, the three main bureaus hold different records, and different lenders run different versions — so your “score” is really several slightly different numbers, none of which you control [13]. A wrong account or a payment marked late that wasn’t can drag it down, and you may not notice until you’re refused [3].

So you are being continuously read by a system built by lenders, for lenders, to serve their question, not yours — and you are standing on the outside of it, holding a single output, unable to see the machine that produced it. You can dispute an error, and you should — that right is free and real [25]. But you can’t audit the whole. Almost nobody who is scored can.

That’s the part to carry. The number on your statement isn’t your worth, and it isn’t your past. It’s one institution’s guess about your future, made for its own ends, with most of the reasoning hidden from the person it’s about. Knowing that won’t change the score. It changes how tightly you hold it.

03 · Lab · your turn

Feed the Machine

Make money decisions and watch a hidden scoring model react, feeling when the responsible thing and the rewarded thing pull apart.

04 · Hope · carry this

The machine is opaque, but it is not closed against you — you can read your own file for free, fix what's wrong, and watch the number quietly rebuild on the one input nobody can fake: time, and your own steadiness.