Skip to content

AI in accounting: payment reminders, open items, work instructions

By the KI können editorial team9 min read

Written with AI and checked against the sources by a second AI. As of November 16, 2026. More in the notes on AI.

In accounting, every figure counts, and every entry needs a receipt. That’s why an AI doesn’t calculate for you and doesn’t decide which account something is posted to. It helps with everything around that: with reminders that are friendly and clear, with formulas for analyses, with work instructions, and with the question of where you need to look something up.

In the accounting office, a woman takes a binder from the open cabinet, at the desk beside her a colleague checks receipts with a calculator.

Key points

  • The AI phrases reminders, queries to suppliers and work instructions from your facts.
  • On posting questions, the AI provides keywords to look up, not a binding answer.
  • You build formulas for open items with sample rows and try them out in a copy.
  • An e-invoice is a structured data record, not a PDF from an AI chat.
  • Bank details, salaries and customer data go at most into an AI covered by a contract, never into a private account.

What an AI prepares in accounting

Where the AI helps in accounting
TaskYour ingredientsYour check
Payment reminder and dunning letterAmount, due date, stage, toneAmounts, dates, every consequence that is named
Analyzing open itemsColumns and sample rows, cutoff dateRecalculate two rows by hand
Clarifying a posting questionThe facts without namesEvery statement against the law, the guidelines or your tax advisor
Writing a work instructionYour description of the processWhether the process really runs that way
Explaining figuresThe analysis as a tableWhether every figure named is in it
Query to a supplierWhat is missing from the invoiceInvoice number, amount, deadline

In the trial lesson you put your first instruction together and get feedback. No account, in a few minutes. Start the trial lesson

Payment reminders and dunning letters: friendly and unambiguous

Andre works in accounts receivable at a beverage wholesaler. Every week he writes reminders to restaurants, clubs and companies. The first one should be friendly, because most invoices are simply forgotten. The second one should say clearly what happens next.

Sample instruction
Write a second reminder to a business customer (we’re on formal terms).
Outstanding: invoice [no.] of [date] for [amount], due since [date]. First reminder on [date], no response.
New payment deadline: [date].
Consequence if they don’t pay: we will only deliver against prepayment. Don’t mention any other consequences.
Offer that they can call with questions about the invoice, phone number as [phone].
Tone: matter-of-fact, without accusation, 100 words at most.

Andre fills in amounts and dates from the list of open items, not from memory. The consequence in the instruction has been agreed with sales. Otherwise, an AI likes to suggest phrases such as “we reserve the right to take legal action”, which he neither means nor is allowed to decide on.

What the law says about default

Under Section 286 of the German Civil Code (Bürgerliches Gesetzbuch, BGB), anyone who doesn’t pay after a reminder sent after the due date is in default (Verzug). At the latest, this applies 30 days after the due date and receipt of the invoice; for consumers, however, only if the invoice expressly points this out. From then on, under Section 288 of the Civil Code, the business can charge default interest and, from customers who aren’t consumers, also a lump sum. Whether and when you do so is decided by management.

Sorting open items by age

Isla is an accountant at a furniture store. Every month, management wants to see how much money has been outstanding for 30, 60 or more than 90 days. The accounting software can export the list, but not grouped like that. Isla has a formula built for her, with invented sample rows instead of real customers.

Sample instruction
I use Excel in English.
Columns: A document number, B customer (initials), C due date, D open amount in euros. Data from row 2. The cutoff date is in G1 and should stay fixed when copying down ($G$1).
In E I need the days since the due date (negative if not yet due), and in F the group: “not due” (0 days or less), “1–30”, “31–60”, “61–90”, “over 90”.
Below that, a small table with the total per group.
Give me the formulas for row 2 and explain each part in one sentence.
Here are five invented sample rows for testing: [rows]

Isla first pastes the formulas into a copy. Then she checks the rows right at the boundaries: due exactly 30 days ago, 31 days ago, due today. If they’re all right, she compares the grand total with the balance in the software. More on formulas in the guide AI and Excel.

Posting questions: keywords, not an answer

Mei is a certified accountant at an engineering firm. With an unusual transaction, such as a down payment on a machine that won’t be delivered until next year, she doesn’t ask the AI “How do I post this?”. An answer like that sounds certain and can still be wrong, complete with an invented account number. She asks for the points she needs to clarify.

Sample instruction
I’m an accountant in Germany and I’m clarifying a transaction: [facts in three sentences, without names and amounts].
Don’t give me a posting or an account number.
Instead, list:
1. Which questions I need to clarify about the facts.
2. Which provisions or terms I should look up, with the law and section, as far as you know them for certain.
3. Where you are unsure; mark that explicitly.

Mei looks up every section named in the original: in the laws, the administrative guidelines or the commentary the firm uses. If any doubt remains, the question goes to the tax advisor. How to spot invented details is described in the guide Checking AI answers.

Writing down processes as work instructions

Jamal is the accountant at a scaffolding company and so far the only one who knows how incoming invoices are checked, approved and filed. If he is out, everything stops. On top of that, the tax authorities expect a business to describe its accounting processes, in what is known as a procedural documentation (Verfahrensdokumentation). Jamal describes the process to the AI in his own words and has it put them in order.

Sample instruction
Below I describe how incoming invoices are handled at our company.
Turn it into a work instruction with the sections: purpose, responsibility, process in numbered steps, filing, cover.
Use only what I describe. Where a step is unclear or a responsibility is missing, write [clarify: …].
At the end, ask me five questions about gaps a stand-in would run into on the first day.
Description:
[text]

The AI’s questions uncover what was only in Jamal’s head: who approves invoices above a certain amount, and what happens to invoices without a purchase order number. Before the text applies, management reads it, and a colleague tries it out once. The finished instruction is itself one of the documents that are kept for ten years.

Translating figures for management

Not every boss enjoys reading a monthly management report (betriebswirtschaftliche Auswertung). An AI can turn the table into a short text if you tell it to name only the figures that are there and not to interpret anything it doesn’t know.

Sample instruction
Here is the monthly report as a table, with the previous month and the same month last year.
Write five sentences for management: what went up, what went down, what stands out.
Name only figures that are in the table, and only calculate percentage changes if I say so.
You don’t know the reasons for changes; write [reason: …] instead.
[table]

E-invoices: what an AI doesn’t replace

Between businesses in Germany, the e-invoice (E-Rechnung) is becoming mandatory. It means an invoice in a structured electronic format, such as XRechnung or ZUGFeRD, that software can process without retyping. ZUGFeRD is a PDF with an embedded data record. A plain PDF without this data record is not an e-invoice, not even one an AI has laid out nicely.

Timeline of e-invoicing between domestic businesses (as of November 2026)
Since or fromWhat applies
January 1, 2025All businesses must be able to receive e-invoices.
until the end of 2026Issuing on paper or, with consent, as a PDF remains allowed.
2027This only applies with a total turnover of at most 800,000 euros in the previous year; EDI procedures with consent remain possible.
from 2028E-invoices for everyone; invoices up to 250 euros may still come on paper or as a plain PDF.

E-invoices are created and read with the accounting or invoicing software. Using a sample invoice without real customer data, an AI can explain to you what a field in an XRechnung means or why a supplier is sending a query. It doesn’t build the data record.

Data, receipts and retention

Accounting holds a lot of data that doesn’t belong in a private AI chat:

  • Bank statements, IBANs and direct debit mandates
  • Salaries and everything from payroll
  • Names and addresses of customers and suppliers
  • Terms, purchase prices and margins

Some accounting programs now read receipts with AI. Whether you use that is decided by the business, with a data processing agreement (Auftragsverarbeitung) and in consultation with your tax advisor.

And a chat is not an archive. Under Section 147 of the German Fiscal Code (Abgabenordnung), a business keeps books and annual financial statements for ten years, accounting receipts for eight and business letters for six years. Whatever counts for the accounts belongs where these periods are observed.

Check before you take anything over

  1. Every amount, every date and every number comes from the software or the receipt, not from the chat.
  2. Check every formula on edge cases and against the grand total, in a copy first.
  3. Look up every section and every posting rule in the original.
  4. Replace every [clarify: …], every [reason: …] and all placeholders.
  5. Have new templates and instructions read by a second person.

Learning on your own processes

Anyone who works in accounting checks carefully anyway. Working with AI needs exactly this attitude: a clear instruction, and the result compared with the original. That is the basis of the learning app “KI können”; you practice with your own AI or your company’s, on cases from many professions. The trial lesson is the place to start, without signing up.

Sources

Information as of November 16, 2026. Laws and offers can change; this article is not legal advice.