Audit-ready books,
without a bookkeeper.
Scalebook takes over everything between the document arriving and the books being correct. Staff photograph a receipt, AI decides the entry, your ledger stays in Xero — and your auditor still signs, and is still responsible.
Every Hong Kong company must be audited.
Almost none of them has a bookkeeper.
Under the Companies Ordinance, every non-dormant Hong Kong company must file accounts audited by a CPA. There is no small-company exemption from the audit itself — only from some disclosures. The obligation is universal; the capability is not.
The work does not disappear. It lands on the founder, on a part-time bookkeeper, or on a scramble every March — a year of documents scattered across colleagues' phones, inboxes and drawers, assembled at the last minute by someone trying to remember what a payment six months ago was for.
Trade & Industry Department, March 2026
AFRC, 2022 data
Currently mandatory only for large MNE groups
Five stages. You are involved in one of them.
The only place a human is genuinely needed is where the system is unsure. The rest of the time it should be invisible.
-
Submit
Staff photograph a receipt, or forward an invoice to an address. No login, no category to pick. For a shop assistant the entire interaction is about twenty seconds.
-
Understand
A vision model extracts every field — supplier, date, currency, line items, total — with a confidence score on each one. Traditional and Simplified Chinese, thermal-paper receipts and mainland VAT fapiao are all in scope.
-
Decide
Coded from your own company's history. Not a general model trained on everyone's data, but retrieval over your own past entries: "you have coded DHL this way 47 times."
-
Post
Written into Xero with full evidence attached. Below the confidence threshold — or failing any policy check — it stops and waits for a person, and says plainly where it stopped.
-
Prove
Every decision logged immutably: which model, which version, what confidence, who approved, what changed. An auditor can trace any entry back to its source document.
Three people, three completely different products
The most important design constraint: two of the three never open an app.
Shop and warehouse staff
Buy something, photograph the receipt, done. No login, no account code, no form. Below the threshold, that is the entire interaction.
Whoever approves
One digest a day — "4 things need you, HK$2,840" — approve or ask. Never a notification per document; an approval system that pings per item gets muted within a week.
You
If something needs deciding it reaches you; if not, the system stays quiet. Roughly ten minutes a week on exceptions, half an hour a month to close, two hours a year for the audit.
How we measure success
Not clicks, not hours saved. How many days a month you have to think about bookkeeping at all. Today that is eight or ten. The target is one.
Three things we don't do — as important as what we do
We don't build a ledger — yet
In phase one, Xero keeps the statutory books, the tax rates and the reports. You don't migrate, and your accountant doesn't relearn anything. Their trust in Xero is an asset we borrow for now.
We don't replace the accountant
In Hong Kong the audit signature cannot be automated away. Any product promising "no accountant needed" is selling something it cannot deliver. We replace the monthly data entry, not the signature.
We don't post unattended by default
Automatic posting requires both model confidence and a set of deterministic checks. Everything is reviewed at first — and even once loosened, 10% of auto-posted entries are sampled permanently.
Two companies in this category raised over US$100 million each on "AI bookkeeping, no accountant needed" and then collapsed. We think that was a promise problem, not a technology problem.
The layer first. The ledger after.
Accounting software built for human bookkeepers will eventually be replaced by software built for machines. We believe that, and we intend to build the replacement. But that is the destination, not the entrance — and you never have to migrate to get there.
-
Phase one · The evidence layer (now)
The ledger stays in Xero. We own the documents, the decisions and the evidence, and we write the coded entries into it. The benefit is immediate: no rebuilding the books every March.
-
Phase two · The shadow ledger
We reconstruct a mirror behind the scenes and reconcile it against Xero's trial balance every month — it has to agree to the cent. Two sets of books run in parallel; you do nothing and notice nothing.
-
Phase three · The statutory ledger
Once the mirror has agreed for twelve consecutive months and carries better evidence than Xero does, switching is a decision rather than a migration — and the decision is yours, not ours.
Why say this out loud
Because it is where we are going, and hiding it would not serve you. But we will never ask anyone to change what they do today for a phase three that does not exist yet — phase one has to stand on its own, or the other two are meaningless.
Three readers, three paths
You are the bookkeeper
A cross-border team, multi-currency revenue, platform settlements, and a rebuild of the books every March — this product is shaped around exactly that company.
Same team, more clients
Your cost is labour and your margin lever is utilisation. This is not about cutting staff — it is about the same people carrying more engagements. Priced per client entity per month.
What would you need to accept it?
No standard-setter has blessed an AI accounting workflow, so the burden of proof is ours. We publish the evidence format and we want to hear whether it is enough.
In development. We prove it on our own books first.
Scalebook is built by an operator with the problem. The first version runs on two real Hong Kong companies: a leather goods brand selling into nine markets with a Shenzhen team, and a water filtration retailer with a shop. Both have multi-currency revenue, platform settlements, Chinese-language documents and a statutory audit — and neither has a full-time accountant.
The first sentence we need is this: the auditor completed the audit with materially fewer queries than last year, and put it in writing. Until that sentence exists, there is nothing to sell.
If you run an SME or an accounting practice in Hong Kong and you're willing to try this at an early stage, we'd like to talk. Design partners pay nothing; what we ask for is real documents, an hour a week of opinion, and an honest answer.
We reply personally. No newsletter, no drip sequence — your address is used to talk to you about this and nothing else.