45 ways to cut 368 neobanks — and the two counts we won't merge
The directory has always been able to filter by licence type, custody model, KYC posture, card network, region and audience. What it couldn't do was give you a page. A filter state lives in a query string, and a query string is not something you can link in an argument, cite in a paper, or send to someone who asked which neobanks actually hold their own banking licence.
So the filters are now 45 pages at /browse/. Each one answers a single question outright, states its own count, explains what the category means, and lists every member with custody, licence and geography. They're generated from data.json, so they can't disagree with the directory.
That part is plumbing. The interesting part was deciding what not to put on them.
The counts that refuse to be added together
Search "neobanks in Germany" and you'll get lists that quietly merge two completely different facts: companies based in Germany, and products available to someone in Germany. N26 is both. Revolut is only the second. Wise is available in Germany and headquartered in London. Merging them produces a number that means nothing, and it's the number everyone publishes.
In this dataset those two facts have very different reliability, and that's the whole reason to keep them apart:
| Fact | Known for | Why |
|---|---|---|
| Where the company is headquartered | 334 of 368 | It's in filings and on the About page. Cheap to verify, rarely disputed. |
| Where the product is available | 167 of 368 | Needs checking market by market, and most neobanks describe their footprint as "select regions" or "60+ countries" without naming one. |
So /countries/germany/ counts the 8 companies based there, then lists separately the ones verified as available there without being based there — and says outright that a neobank missing from the second list has not been ruled out, it has not been checked. Absence of evidence, labelled as such.
The tempting alternative was to let the availability field set the headline number. It would have looked more complete and been wrong on all 17 country pages, understating every one by an unknown margin. A 45%-complete field should never be allowed to drive a total.
Where a ranking would have been a guess
Every one of these pages sorts by reported users, which immediately exposed a problem: only 22 of 368 neobanks disclose a user figure at all. For the big regulatory cuts that's enough to name the giants honestly. For most of the niche cuts it isn't. Sorting 30 SMB neobanks by a field that 28 of them leave blank produces an alphabetical list — and calling the top of an alphabetical list "the biggest" is a fabrication with a plausible shape.
So each page checks whether it can support a ranking before offering one. Where it can't, it says so: "Not answerable from public data: only 2 of these 14 disclose a user figure at all. The list below is therefore not a ranking, and anyone presenting one for this category is guessing."
That sentence appears on a lot of pages. It should appear on a lot more pages across this industry than it does.
What the cuts actually show
Slicing the same 368 products six different ways surfaced things the directory's default view buries.
60 neobanks issue no card at all
The card-free list is the one I'd send to anyone who thinks "neobank" means "app with a debit card." It splits cleanly in two: 39 are traditional licensed institutions — Marcus, OakNorth, Atom Bank, Tandem — where deposits and lending are the business and a card would be a distraction. Another 17 are self-custodial wallets that leave spending to a partner or an exchange. A card is a distribution choice, and its absence usually tells you where the revenue comes from.
The regulatory split is the least fashionable and most important cut
127 hold their own banking licence. 77 run on somebody else's. 48 are e-money institutions that may issue cards but not lend your deposits. 51 are software that never holds your money, which means there is nothing to insure and nothing to freeze. Four structures, four completely different answers to "what happens to my balance if this company dies," and almost no app tells you which one it is.
No-KYC is real, tiny, and structurally explainable
13 products work with no identity check — and every one of them is self-custodial, because that's the only structure where it's possible. There's no account to open and no balance held for you, so there's nobody to verify you to. A further 32 are card-only: the wallet needs nothing, the card needs ID, because the card's licensed issuer carries the obligation. The line isn't ideological, it's where the regulated entity starts.
16 neobanks run on domestic card rails
RuPay in India, Verve in Nigeria, Mada in Saudi Arabia. These 16 trade international acceptance for cheaper interchange, usually under regulatory pressure to keep everyday payments on local infrastructure. It's a small list today and a live policy trend.
Cuts that didn't get a page
A page needs at least six members to exist here. That excluded eight audience niches covering 15 products — neobanks for women, for climate, for seniors, for the Black community, for gig workers, for people with disabilities. They're in the dataset and on their profiles; they don't get a hub, because a page listing two companies isn't a resource, it's a stub competing with its own members for attention.
The threshold is arbitrary. Having one is not: the alternative is thin pages that exist to catch a search query rather than to answer it, which is how a useful dataset turns into content sludge.
Use them
Start at /browse/, or jump straight in: by licence, by KYC, by card, by region, by country, by who it's for. Every neobank's profile now links the cuts it belongs to, so you can walk sideways from any product to its structural peers.
And if you'd rather compute your own cuts, that's the actual point — the raw file is MIT-licensed, and it's mirrored on Hugging Face and Kaggle as a flat CSV. Every count on every one of these 45 pages is reproducible in one groupby. Check them.
The availability gap is the open one. If you can confirm, with a source, that a product on this list serves a market we haven't recorded, that's a genuinely useful contribution: open a data-fix issue. Filling that field is what turns 17 country pages from half-answers into complete ones.
Counts reflect the open dataset as of July 2026 and are regenerated from data.json, so the pages and the directory cannot disagree. Figures are compiled from public filings, registers and product documentation. Nothing here is financial advice. Sources on each profile page.