We lied about the five minutes. A customer did 160 invoices in 5.6.

On the BillyBox How it works page we say BillyBox gets you through 80 invoices in under 5 minutes. We have said it for months, in two languages. This month a customer went through 160 documents in 5.6 minutes — and that was his worst month. We were not overselling. We were underselling, by roughly half.

He runs two small food businesses through one mailbox — a grocery store and a deli restaurant. Eight months of unfiled paperwork had built up in it: 1,102 documents, about 138 a month. He worked through all of it in 27 minutes of actual work, on his phone, and exported all eight months seventy-four minutes after he first signed up.

What eight months actually looks like

He went newest month first and worked straight back, one clean pass per month. Here is the whole session:

March is the row that broke our claim, and it is the one we would put on a poster. It was also his slowest by a distance. The other seven months all came in under four minutes — between 2.2 and 3.7, averaging 3.1. His fastest was January: 133 documents in 2 minutes 12 seconds, which works out at almost exactly one second per document. And he got quicker as he went — January was the last month he touched, and it ran at about 60% of the time per document of the first one he started on.

Put our own claim next to it. Eighty invoices in five minutes is 3.75 seconds a document. His worst month beat that, at 2.1. His best beat it nearly fourfold. That is the part we got wrong: not the promise, the modesty.

What he actually used it for

Finding the invoices was the easy half. The hard half was sorting them. His two businesses and his personal life share one inbox, as they do for almost everyone who runs something small, and the job he had been putting off for eight months was not “where are my invoices” but “which of these is the company’s and which is mine.” Of the 1,102 documents, he marked about half business and just under half personal, and ignored a handful that were neither. Roughly 520 documents that would otherwise have sat in a business export — or been left out of one — each got a one-second decision from the person who actually knows the answer.

The pile was the usual mix for someone running a shop and a restaurant: supplier bills, card-terminal statements, utilities, personal spend — and, because the restaurant is on Wolt, a steady stream of delivery-platform paperwork. Wolt payout emails carry several PDFs each — a commission invoice, a payout report, a sales report — and BillyBox gathered every one of them along with everything else. It works very well for that, and there is nothing to set up: it reads every attachment on every email, not just the first.

How 1,102 decisions fit in 27 minutes

There is no trick to it, which is the point. Each document needs exactly one decision, and BillyBox asks for it with exactly one gesture.

• The queue is already sorted. By the time you open it, every document has been collected, its vendor, amount and date read, and a suggestion attached — business, personal, or ignore. You are confirming or overriding, not filling anything in.
• On a phone it is a swipe deck. One card per document. Swipe right for business, left for personal, down to ignore. Your thumb never leaves the screen and your eyes never leave the vendor name. That is how a month of paperwork gets done standing up, in the time it takes a fryer to heat.
• On a desktop it is three keys. B, P, I. No mouse, no menus, no folder tree. If you can touch-type you can clear a month in the time it takes to read it.
• Repetition gets faster, not slower. The tenth invoice from the same vendor looks like the ninth. That is why his per-document time fell by 40% across a single sitting — the interface gets out of the way and your hands learn the rhythm.
• Then one export. The business half goes to your accountant as one organised folder per month. The personal half stays yours. Nothing else to do.

At that point the limit is not the software. It is how fast you can read a vendor name and decide, which for most people is about a second. Which is exactly what the January row says.

Why the last second is yours

You might reasonably ask why there is a decision at all. If BillyBox can already read the vendor, the amount and the date, and already suggest an answer, why not just file everything and send the folder? We could. We chose not to, and it is the most deliberate choice in the product.

The plain reason first: a document records who was billed and what it cost — not why the money was spent. And the why is what your books actually need, because that is what tax rules test: was this for the business? The laptop billed to you personally that lives behind the counter. The phone contract in the company’s name that is really your own line. The restaurant bill that was a supplier meeting, and the identical one from the week before that was dinner. On paper those pairs look the same; in your books they must not be. The answer is not printed anywhere on the document — it exists only in the head of the person who spent the money.

Your accountant, meanwhile, only ever sees the paper. If nobody who knows the purpose marks each document, the paper’s face value becomes the books — and the face value is wrong in a pattern, not at random: every business expense billed to your own name silently drops out, every private cost that landed on the company stays in. One second per document, from the one person who knows, is the cheapest possible fix for that. You mark what it was for; your accountant applies the rules.

That one second is the only moment in the month when you actually look at what you are paying for. The commission you forgot was that high. The subscription you stopped using in March. The card-terminal fee that crept up. Not a report about them — the documents themselves, one at a time, passing under your thumb. Fully automate that and you get a perfect folder for your accountant and no idea what is in it. Keep it, and you stay in touch with your own business for the price of a second a document.

And there is a kind of peace in it that is hard to explain until you have felt it. Eight months of paperwork you have been avoiding, cleared, and you know exactly what went into every folder because you touched every item. The queue is empty because you emptied it. Nothing is hanging over you that you have not looked at. That is a calmer place to be than a robot telling you it has been handled, and it is why the interface is built to make the decision fast rather than to make it disappear.

Why speed was the whole point

A word on why that number is on our How it works page at all. I built BillyBox for my own one-person company, and I built it because I do not want to deal with accounting. I want to hand my accountant everything they need, once a month, and go back to work. That is the entire ambition: not to do the bookkeeping, but to make the handover take minutes instead of an evening.

I tried the tools that already existed first, and gave up on each of them — not because they were bad at what they did, but because I could not get through a month’s invoices in them without it turning into a project. Too many steps between “here is my inbox” and “here is the folder for my accountant”, and interfaces that assumed someone would be sitting at a desk doing this as their job. I am not that person, and neither is anyone running a shop. So the design goal from the first day was the one you can measure with a stopwatch: how fast can a person who hates this get it done and get back to work. That is why there is a minutes claim on the page, and why we were glad to be wrong about it in this direction.

All of it on a phone

Worth saying plainly, because it is the part people assume is impossible: none of this happened at a desk. He signed up on his phone, connected his mailbox on his phone, and made all 1,102 decisions with his thumb. Bookkeeping loses to the lunch rush because it needs a laptop, a quiet half hour and a mood. A second a document while the fryer heats up is a different proposition.

How we measured this

Every number here comes from one BillyBox account, measured on 18 August 2026. The documents are what BillyBox collected from eight months of one mailbox (January–August 2026). The timings are derived from the timestamp on each classification decision, with any gap longer than 60 seconds treated as a break and excluded — so “27 minutes” is time spent deciding, not elapsed time at the screen, which was 65 minutes including eleven pauses. Document counts per month are exact. Per-document rates are documents divided by active seconds.

It is one customer. A messier mailbox, or one with ten times the volume, will not look like this. We are publishing it because it is the first time we have been able to put a stopwatch on the whole job end to end, and the result surprised us. He is not named here and will not be; nothing in this piece is more specific than the kind of business he runs. If he recognises himself: the numbers impressed us, and thank you.

Frequently asked questions

This piece first appeared on the BillyBox blog — read the original. BillyBox collects invoices and receipts from your inbox automatically; more about it here.