AI for accounting firms: client reports where the AI never writes a figure

Ask a chat assistant to turn a client’s trial balance into a monthly report and it will produce something that looks finished. The trouble starts in review: a total that does not add, a ledger counted twice, a margin on the wrong base. The model wrote those figures, so every one has to be checked, and next month it does the whole job again from nothing.

AI tools for accountants are worth using when they take the reading and the wording off the team and leave the figures to arithmetic that is the same every time.

Every figure in the report is computed by a deterministic engine from the client’s mapped trial balance, checked to tie back to it, and opens onto the ledgers behind it. The model receives a sheet’s structure, a small redacted sample and ledger names for an action you started — never a whole file, and never a party’s name.

No member of our staff can open a client’s file. Every time the server opens one is on a record you can see, and deleting a client’s company destroys the key its data is sealed under. More on security and data handling.

The first month for a client is where the AI earns its keep: recognising the exports and proposing the mapping you confirm. From then on the mapping is a record. When the next month’s trial balance has the same structure, the refresh is arithmetic and makes no AI calls at all — only new or renamed ledgers come back to you to place.

For CPA firms running client accounting services, the package is the same every month: a P&L against last year and budget, a balance sheet, cash flow, KPIs, A/R and A/P aging where the registers are loaded, and a page of commentary. Here it is built from each client’s raw exports, and the workbook can carry a balance sheet and income statement in US GAAP presentation. Your firm’s name and logo can sit beside the client’s when you present the board and on the workbook’s cover.

It is paid from prepaid credits per action, with a monthly fee for each active company and no per-seat or per-client subscription.

Practices producing monthly management accounts from Xero, QuickBooks, Sage or MYOB load the trial balance and the aged debtors and creditors reports they already run. The statements can follow the UK Companies Act formats or IAS 1, figures are shown in the client’s own currency, and the commentary can be written in the client’s language.

For CA firms, the monthly MIS for each client starts from Tally’s raw reports — the trial balance ledger by ledger, and the bills receivable and payable for ageing. Tally’s group structure is read first, figures can be shown in lakhs and crores, and the workbook can add a Schedule III balance sheet and profit and loss. AI tools for chartered accountants are only as useful as their figures are traceable, and every figure here opens onto its ledgers. AI for Tally, in detail.

• Who computes the figures — the model or arithmetic? If a language model writes the numbers, every report needs checking line by line, every month.
• Can every figure be opened onto the ledgers it was added up from? A reviewer signs off what they can trace.
• Who can open the client's file? Ask whether staff at the vendor can, and whether every opening is recorded somewhere you can see it.
• What does the model receive? A whole file, or the structure, a redacted sample and ledger names? Are party names replaced before it sees them?
• Is client data used to train a model?
• What happens next month? A tool that sends the whole job to the model again charges you again for work that did not change.

Every client is a company in one account, each with its own mapping, history and encryption key. The companies page shows where each client’s month stands, and month end for several clients takes a batch of files at once: each is matched to its company by name, and nothing is uploaded until you have confirmed every match. How a practice uses it day to day is on client reporting software for accounting firms.