How We Rank and Verify Every Account

A star rating on this site is one of exactly two things, and the tooltip next to every rating tells you which. The first is a rating somebody else published out of five — today that means the public average score of a bank's own mobile app, transcribed with its source link and the date we checked it, exactly like any other figure here. We do not adjust it. The second is a Bank Compass score, which we calculate ourselves.

A Bank Compass score is arithmetic on figures that are already on the page. Nothing goes into it that you can't check yourself in the same card: for a credit card it is the annual fee, the regular APR and whether the issuer publishes an intro APR offer; for a deposit account, the APY, the monthly fee and the opening deposit; for an insurer it is dominated by the AM Best financial strength grade. Each criterion is weighted, scored from 0 to 1, and the weighted average is mapped onto a one-to-five scale. The tooltip beside any rating lists exactly which criteria went into it.

What each criterion is measured against matters more than its weight, so here it is. A cost — a fee, a commission, an account minimum — is measured against zero, because free is the real best and nothing has to be invented. That has a consequence worth stating plainly: when every provider charges $0, they all get full marks on that criterion rather than being ranked against each other, because there is genuinely nothing between them.

An interest rate can't use zero, since no loan is free, so an APR is measured against 36% — the cap the Military Lending Act puts on consumer credit to active-duty service members, and the line most consumer advocates treat as the edge of predatory lending. Where somebody else already publishes a scale, we use theirs: five for an app store rating, A++ down to D for AM Best. Only where none of those apply does a criterion get measured against the range of the other products we track.

What happens when we have not verified one of those criteria is worth spelling out, because it changed in August 2026. A missing criterion used to be dropped from the score entirely, and that quietly rewarded the product: a savings account whose APY we had not confirmed was scored on its fee alone and came out near the top, when the APY is the whole point of the account. A gap should never be an advantage. So a criterion we cannot verify is now scored at the average of the products we compare it with — not a made-up middling value, but the actual calculated average of that peer group on that same criterion. It neither rewards nor punishes the gap.

That is an estimate, and we label it as one rather than burying it. The tooltip beside a rating separates the criteria scored on the product's own published figures from the ones estimated at the category average, and tells you what percentage of the weight came from real figures. When most of a score is estimated it is marked 'Est.' next to the stars, and when we have verified nothing at all for that product it is marked 'Category avg' — the number is literally the average of its peers, and you should read it as one.

One consequence we will state plainly: an average is better than what roughly half a group scores, so a product missing a figure can end up above one whose verified figure is poor. The fix for that is verifying the figure, not adjusting the formula, and the gaps are the work we prioritise.

There are two figures we will never estimate, because getting them wrong costs you money rather than a fraction of a percent: an insurer's financial strength grade, and the price of credit.

If AM Best publishes no rating for an insurer — which is the case for two of the companies we list — that company gets no score at all, rather than inheriting the average of the A++ and A+ companies around it and appearing to outrank them on the one measure of whether a claim gets paid. That absence is AM Best's, not a judgement of ours about the company.

The same applies to a lender that does not publish an APR: a cash-advance app whose real cost is a subscription and an express fee cannot be scored as though it charged what a personal loan charges. Six products on this site carry no stars for these reasons, and in each case the card still shows everything we did verify.

Two things the score deliberately does not do. It does not rate mortgage rates against each other, because every lender quotes its sample on different assumptions — loan size, down payment, state, discount points already bought — so those numbers are not comparable and ranking them would invent a comparison that does not exist. There, the score measures how much the lender publishes up front. And it never rates how good a rewards programme or a platform 'feels', because that is an opinion nobody has measured.

A star rating cannot be bought, and no company is told its score before it is published. There is no affiliate, referral or paid-placement arrangement behind any figure here, no company reviews these pages before they go live, and a provider asking us to change a score gets the same answer as anyone else: bring the published figure and we will re-verify the field. If we get one wrong, we correct it and the verification date changes with it.

Star ratings are also never marked up as structured data for search engines: a site scoring the things it compares is not a customer review, and we won't present it as one.