Use cases by industry4 use cases

AI automation for accounting and bookkeeping firms

An accounting firm's month is decided by what clients send, and when. Twelve people look after 180 small-business clients; much of the work is asking for missing receipts and bank statements, sorting what arrives by email, WhatsApp and the portal, and posting it to DATEV Unternehmen online, Xero or QuickBooks. Hiring does not fix it: in Germany, a STAX survey found that firms fill only about 70% of their open positions, and sole practices about 40%.

Capture is no longer the hard part. DATEV, Dext and Hubdoc read receipts well, and many ledgers receive e-invoices natively. What no product does for you is the chasing, per client and per period, in the client's language, which is why accounting document chasing is the build I would start with. It runs on checklist logic only your firm knows: which client forgets the second bank account, which one sends screenshots instead of statements.

E-invoicing changes the intake side. German businesses have had to be able to receive structured e-invoices since 1 January 2025, Belgium moved B2B invoicing to Peppol in January 2026, Poland's KSeF went live in February and April 2026, and France's reform started on 1 September 2026. XML removes the OCR but not the validation, matching and coding, which e-invoice intake and supplier invoice processing cover.

What the firm actually spends its hours on

  • Checklists that know each client

    Expected documents per client and period are learned from history: two bank accounts, a monthly Amazon statement, quarterly rent. The gap list is exact, so reminders name the missing items instead of asking for everything.

  • Bank statements with holes

    A statement set is complete only when the balances run continuously from one period to the next. Checking that sequence in code catches the missing month before the reviewer does.

  • Booking proposals with reasons

    Account and tax codes are proposed with a reason and the precedent from earlier periods. Only genuinely unclear items reach the client question list, such as a EUR 480 payment to Amazon with no receipt.

  • Hybrid invoices that disagree

    A ZUGFeRD file carries both a PDF and XML. When they differ, the XML is the part that counts, so the system validates it and books from it, and keeps the original under GoBD archiving rules.

  • Quarterly filings in the UK

    Making Tax Digital for Income Tax brought quarterly submissions for many sole traders and landlords from April 2026, turning once-a-year clients into four chasing cycles. The same checklist engine runs the quarterly rhythm.

The lines software should not cross in a firm

  • A person approves every posting

    The model proposes bookings and a reviewer approves them. The fully autonomous AI accountant is still a sales pitch, and the professional liability stays with the firm.

  • Reminders, never pressure

    Tone and frequency are capped per client, and the one client who never answers is escalated to the accountant rather than sent an ever sharper message.

  • WhatsApp with consent only

    Chasing on WhatsApp needs the client's opt-in and a business account. The message names the missing documents and links to the upload page; balances and amounts stay out of the chat.

  • US close software is a different market

    Basis, Numeric and Rillet, which raised $100M at a $1B valuation in August 2026, target finance teams on US ledgers. For a European firm on DATEV, Exact or Pennylane, the gap is integration, not a new platform.

Software these builds usually connect to

The systems do not get replaced. The build sits across them, reads from them through their APIs, and writes results back.

DATEV Kanzlei-Rechnungswesen / DATEV Unternehmen online / Addison / Xero / QuickBooks Online / Dext / Hubdoc / Candis / Karbon / TaxDome / Pennylane / Exact Online

Frequently asked questions

Can AI do bookkeeping in QuickBooks or Xero?

It can do much of the preparation: read receipts and invoices, match them to bank transactions, propose account and tax codes with reasons, and list the items only the client can explain. Xero and QuickBooks Online both have full APIs and official MCP servers, so proposals can be written back as drafts. A bookkeeper still approves postings; the time saved is in preparation, not in removing the review.

Is there an AI agent for DATEV?

DATEV has its own capture and assistant features, and they are worth checking first. There is no official MCP server for it, so custom work usually sits beside DATEV: chasing documents, validating e-invoices, preparing booking proposals and handing data over through the interfaces DATEV supports. German-language firm workflows are exactly where most off-the-shelf AI tools stop.

What does e-invoicing change for accounting firms in 2026 and 2027?

Inbound invoices increasingly arrive as XML, by email, Peppol or national platforms, so the work moves from reading PDFs to validating formats, rendering them readably for approvers, matching, coding and archiving the original. In Germany, issuing becomes mandatory from 2027 for businesses above EUR 800,000 turnover and from 2028 for all, so clients on old billing software will ask their firm what to do.

How do we stop chasing clients for receipts by hand?

Keep an expected-document list per client and period, compare it daily with what is already in the ledger and the inbox, and send one reminder that lists exactly what is missing, by email or WhatsApp, in the client's language. Make uploading a single link, cap the frequency, and escalate non-responders to the accountant. Photographed receipts go through PDF and image extraction with a review queue.

Also browse: Finance and accounting · DATEV · Xero · QuickBooks · all use cases