Where time gets lost in accounting
In many accounting teams, most of the work is not really accounting at all. Invoices arrive as PDFs by email, on paper in the post or as photos taken on a phone. Someone opens them, reads off the supplier, invoice number, amounts and tax rates, looks for the matching purchase order, decides which account and cost centre the invoice belongs to, and types everything into another system. That takes time every single day and is prone to errors, especially when things need to move quickly.
These very steps follow a recognisable pattern. That is why accounting is one of the departments where AI automation delivers a verifiable benefit fastest. This article shows which tasks are suitable, what a typical workflow looks like and where the limits lie.
Which tasks are suitable
Not every task in accounting lends itself equally well to automation. Good candidates occur frequently, follow rules and are easy to check. Typical examples are:
- Incoming invoices: capturing and reading invoices from the mailbox, scans and uploads
- Pre-coding: suggesting account, cost centre and tax code based on your previous postings
- Matching against purchase order and delivery note, and flagging discrepancies
- Passing documents with a posting suggestion to DATEV or your accounting software
- Preparing dunning: identifying open items and drafting payment reminders
- Checking travel expenses: matching receipts to claims and checking them against your policies
What all these tasks have in common is that the AI prepares and a person decides wherever it matters. That is not a compromise; it is the core of a system your team can trust.
How an automated invoice intake works
Take a supplier invoice that arrives as a PDF in your accounting mailbox. A workflow looks roughly like this:
Read and check
The system recognises the supplier, invoice number, date, line items, amounts and tax rates, even when every invoice is laid out differently. It then matches the invoice against the purchase order and delivery note. If quantities or prices do not match, the discrepancy is flagged instead of being posted without anyone noticing.
Pre-code and approve
Based on your previous postings, the system derives a suggestion for account and cost centre. Below a threshold that you set, the invoice goes straight through. Above it, the invoice goes to a person for approval, together with the reasoning. Only then do the document and posting suggestion land in DATEV or your accounting system, along with an audit log.
The article on automating incoming invoices describes this workflow in more detail and also covers typical special cases.
Working with your tax adviser
Many small and mid-sized businesses do not do all their own bookkeeping but work with a tax adviser. Automation should make that collaboration easier, not bypass it. So speak to your tax adviser early and agree together on:
- Which documents are handed over and in what form, for example via DATEV Unternehmen online, the document exchange platform used by most German tax advisers
- Which postings you prepare and which the firm takes care of
- How special cases are handled, such as down payments, credit notes or intra-EU supplies
- Who checks which suggestions before they are finally posted
For the tax adviser, a cleanly captured document with a traceable suggestion is usually far more welcome than a pile of unsorted PDFs. Agreeing things properly at the start saves follow-up questions every month.
GoBD, audit logs and traceability
In Germany, accounting is subject to the principles for the proper keeping and retention of books and records in electronic form, the German bookkeeping rules known as GoBD. For automation, this mainly means that documents must be stored completely, unalterably and traceably, and it must be clear how each document was processed.
A well-built system records every automated step with a timestamp and source. That way, even months later, you can see why an invoice was pre-coded to a particular account and who approved it. You should adapt your process documentation together with your tax adviser. How the rules apply in any individual case is a matter for qualified professionals; this article does not replace tax advice.
What stays with people
AI can read documents, compare them and make suggestions. But it bears no responsibility for the annual accounts, and it should not make decisions with tax or legal consequences. These include, for example, how to treat unusual transactions, approving payments above your threshold, or dunning decisions involving important customers.
In practice, the work shifts from typing to checking. Instead of capturing every document themselves, your team looks at the cases where the system is uncertain or has found a discrepancy.
How to get started
Start with a single workflow, usually incoming invoices. Collect sample invoices from the last few weeks, including the difficult ones, and decide above which amount approval is required. Check the suggestions particularly carefully in the first few weeks and refine the rules. Only once this workflow runs reliably is it worth taking the next step, for example dunning or travel expenses.
Conclusion
Accounting is an obvious starting point for AI automation: lots of similar documents, clear rules and results that are easy to check. What matters is close coordination with your tax adviser, a complete audit log and thresholds above which a person gives approval.
At Scalient Group, a single process such as incoming invoices starts with the Starter package at €2,490 for setup and €490 per month, excluding VAT. Whether that pays off for your volume of documents is something we clarify in the initial call.