AI in accounting, without client records leaving your network.
A practice holds other people's financial records, which makes most AI a difficult conversation with your own risk and professional indemnity requirements. The architecture matters more than the feature list, so that is where this starts.
A chilled meals producer, in their finance function
Around two million pounds in turnover, processing 180 to 230 supplier invoices a month.
Invoices arrived in several email formats. Proof of delivery lived in a warehouse folder or on paper. Price discrepancies surfaced after the payment run rather than before it.
Invoices are matched against what was actually received, with discrepancies raised before payment rather than after. The system never creates a payment, alters a bank detail or writes to a financial record.
A delivered Orchyn build. Clients are anonymised at their request, so the sector, scale and measured outcome are real and the name is withheld.
Where the time usually goes
Client records arriving in every format
Bank statements, invoices and receipts turn up as PDFs, photographs, spreadsheets and paper, and somebody has to normalise all of it before any work starts.
Month end assembled by hand
The picture is rebuilt from exports because no single system holds it. The reconstruction is the job, and it happens every month.
Deadlines tracked in parallel
Filing dates live in practice software, a spreadsheet and somebody's head. The risk is not that any one is wrong, it is that nobody sees all three at once.
Chasing clients for the same records
A meaningful share of practice time is spent asking people for documents they have already been asked for.
Where the system stops
It does not give an opinion, sign anything, file anything, or decide a treatment. It does not create a payment, alter a bank detail or write to a financial record. It organises, extracts, matches and surfaces exceptions so a qualified person spends their time on judgement rather than assembly.
This is the house position across every Orchyn build: AI assisted, never AI executed. A model is used where judgement over messy input is needed. Every figure is computed by deterministic code against a named source, and a person holds authority over anything with commercial or legal consequence.
Common questions
01Have you delivered a system for an accountancy practice?
No, and we would rather say so than let you assume otherwise. Our delivered work is in food production, wholesale, logistics, construction, property management, packaging manufacture, private healthcare and professional training. Several of those engagements solved finance operations problems directly, including supplier invoice matching and month end consolidation, which is the closest transferable work. If you would rather use a supplier who has already done exactly this inside a practice, that is a fair position and we will say so on the call.
02Where would client financial data be processed?
Inside your own network, on your own servers. In a standard Orchyn build nothing is sent to us, to a model provider or to a data centre in another country. For a practice holding other people's financial records, that is a structural answer rather than a contractual one: there is no transfer to disclose because there is no transfer.
03Would it prepare accounts or give an opinion?
No. It organises, extracts, matches and surfaces exceptions. Preparing accounts, forming an opinion, deciding a treatment and signing anything are all decisions for a qualified person. Where an action carries professional or regulatory consequence, the system prepares and evidences it and a named person makes it.
04Is our client data used to train a model?
No. Nothing about your practice or your clients is pooled, resold or used to train anything, whether ours or a third party's. Because the system is not connected to an external model service, there is no route by which it could be.