Skip to content

AI Accounting Software for E-Commerce with DATEV: What Works

Updated on

AI accounting software works for an e-commerce company using DATEV only if it does two things: it reconciles orders, payments, fees, returns and VAT before anything leaves the system, and it delivers the result as a DATEV booking batch in the EXTF format that your Steuerberater imports. The AI belongs on top of that reconciled ledger, handling setup, questions, changes and the cases rules cannot resolve. CONA, a platform for automated revenue reconciliation and DATEV export for e-commerce brands selling through Shopify, Amazon, OTTO, Kaufland and other marketplaces, works exactly this way, with its AI features currently in Beta.

If you run finance for a brand selling in Germany, you have probably noticed the gap. The bookkeeping tools popular in English-language e-commerce communities turn payouts into journals for Xero or QuickBooks, while your Steuerberater works in DATEV. This guide explains why, what "DATEV compatible" has to mean, and how to evaluate AI features without taking a label on trust.

Why do most e-commerce bookkeeping tools not post to DATEV?

Because they were built for cloud ledgers. The e-commerce bookkeeping tools we checked summarise marketplace and payment provider payouts into journal entries or summary invoices for Xero, QuickBooks, Sage or NetSuite. None of the three we checked names DATEV on its integration pages. Whatever automation or AI a tool advertises, it acts on the ledger it posts into, and for a German business that ledger is usually DATEV.

The table below uses only the vendors' own pages, retrieved on 18 November 2026. We assess integrations here, not AI features.

ToolLedgers named by the vendorDATEVSource
A2XQuickBooks Online and Xero; NetSuite; Sage (integrations FAQ: "Does A2X integrate with Sage?" "Yes."); plus "a useful custom export method for businesses using ERP systems"No statement on the pages checkedA2X integrations
Link My BooksXero and QuickBooks Online ("Hook up your Xero or QuickBooks Online account in seconds")No statement on the pages checkedLink My Books
SynderQuickBooks Online and Desktop, Xero, Sage Intacct, NetSuite, Intuit Enterprise SuiteNo statement on the pages checkedSynder integrations

There is a second, quieter gap: EU VAT. Link My Books is unusually candid about it in its help centre, stating that it "is not a multi-country EU VAT engine" and that "it does not split OSS from Local VAT on a per-country basis" (Link My Books help centre). That is useful honesty, and it tells you what a German setup has to add on top.

None of this makes these tools bad, just built for a different ledger. With DATEV books, a Xero-shaped journal still has to be re-mapped to DATEV accounts and tax keys before your Steuerberater can import it.

What does DATEV compatibility actually mean?

DATEV compatibility means the tool hands over finished booking lines in the DATEV format, not a spreadsheet someone still has to interpret. DATEV describes the format as "a CSV file with a fixed file name prefix (EXTF_....CSV) and a fixed record structure" (DATEV developer portal). For e-commerce, four details decide whether the batch imports cleanly and survives an audit: tax keys, OSS, document references and Festschreibung.

A DATEV-Format interface description form for a third-party tool, published on datev.de, lists the mandatory fields of a booking line as amount, debit or credit indicator, account, contra account and document date, with fields such as Belegfeld 1, BU-Schlüssel, Leistungsdatum and Festschreibung on top (DATEV-Format interface description). Here is what each of the parts that matter most for e-commerce means in practice:

DATEV elementWhat it isWhat goes wrong without it
EXTF BuchungsstapelThe booking batch file with header line and fixed column structureThe import is rejected, or someone rebuilds the file by hand
Steuerschlüssel (BU-Schlüssel)The tax key on a booking line that tells DATEV which VAT treatment applies, set according to your Steuerberater's chart of accountsVAT lands on the wrong rate or not at all, and the VAT return has to be corrected
OSS splitCross-border B2C sales booked per destination country with that country's rateThe OSS return cannot be filled per Member State from the books
Belegreferenz (Belegfeld 1)The document reference on each lineA collective booking cannot be traced back to the orders behind it
FestschreibungLocking the batch so it can no longer be changedBooking data stays editable after it should be final

OSS deserves its own sentence. Cross-border B2C distance sales inside the EU stay taxed in your home country only while their EU-wide total does "not exceed EUR 10 000 (without VAT) in the current and the preceding calendar year" (European Commission, OSS); above that, the destination country's VAT applies, and the OSS return "contains the details of supplies made to customers in each Member State of consumption" (European Commission, declare and pay). Your booking batch has to carry that split, or someone recalculates it at filing time.

Festschreibung is also about timing. DATEV summarises the GoBD rule as booking batches being locked "bis zum Ablauf des Folgemonats", in practice "spätestens zum UStVA-Termin" (DATEV on GoBD). Reconciliation has to be done before that date. For a deeper look at the export itself, see what makes a DATEV export reliable.

Where can AI go wrong in accounting?

AI goes wrong in accounting when it has no fixed ground to stand on. A language model that reads an exported file can produce plausible totals that match no account, change a mapping without leaving a trace, or blur who is responsible for the final booking. German bookkeeping law is strict on exactly these points, so the safeguards have to sit in the software, not in the model.

Three requirements frame what is acceptable.

  1. Traceability. Under § 145 AO, business transactions "müssen sich in ihrer Entstehung und Abwicklung verfolgen lassen". An AI answer that cannot be traced to the underlying transactions fails that test.
  2. No silent changes. § 146 Abs. 4 AO states: "Eine Buchung oder eine Aufzeichnung darf nicht in einer Weise verändert werden, dass der ursprüngliche Inhalt nicht mehr feststellbar ist." A tool that lets AI change account mappings without a before and after record creates an audit problem. The detailed rules are in the GoBD, the BMF letter last amended on 14 July 2025.
  3. Consent and control over AI access. The Model Context Protocol specification says users "must explicitly consent to and understand all data access and operations" and "must retain control over what data is shared and what actions are taken". It is equally clear that "MCP itself cannot enforce these security principles at the protocol level". The implementation has to.

The consequence: rules should handle the bulk of matching, because the same input must always produce the same booking. AI earns its place where rules are expensive: setup, explanations, changes on instruction and the residue matching leaves behind.

How should you evaluate AI accounting tools for DATEV?

Evaluate the foundation first and the AI second. Ask where the figures live, whether every step is logged down to the order, how the result reaches DATEV and what stays with you and your Steuerberater. Then test the AI on your own data: ask it why last month's payout differs from revenue and check whether the answer points to real accounts.

CriterionWhat to checkHow CONA handles it
DATEV outputFinished EXTF booking batch, not a generic CSVExports an EXTF Buchungsstapel to your Steuerberater's specifications
Reconciliation before exportOrders, payments, fees, returns and VAT matched before anything leavesReconciled on live data before export, not in a month-end run
VAT and OSSSplit per destination country, OSS kept separateVAT and OSS reconciled before export
TraceabilityEvery collective booking walks back to single ordersActivity log traces every collective booking down to the individual order
Where the AI worksOn a ledger with accounts and balances, or on a fileAI (Beta) works on CONA's own general ledger
What the AI may doSetup, changes, questions, edge cases, with a record of each stepCopilot, MCP server and Routines, all in Beta, every step in the activity log
What stays with youDATEV import and FestschreibungStay with the business and its Steuerberater
Your own AI clientCan you reach your books from Claude or ChatGPTCONA MCP server (Beta)

The broader market for e-commerce accounting automation, including the DATEV-native group, is compared in accounting automation software for e-commerce.

What does CONA's AI do on top of the DATEV export?

CONA keeps reconciled figures in its own general ledger, and its AI works directly on that ledger in three forms, all in Beta. Together they can do in CONA basically what a person can do in CONA: set up, change, ask, reconcile and prepare corrections. Rule-based matching still handles the bulk deterministically.

CONA Copilot (Beta) is the AI assistant inside CONA. It makes setup fast by guiding onboarding, connecting sources and setting up booking rules according to your Steuerberater's specifications. It makes changes fast: you adjust booking rules, account mappings and settings by instruction in plain language instead of clicking through configuration. And it answers questions about the current state of your accounting against your own data, for example why a payout differs from revenue, what is still open, which account a difference belongs on or how a booking was derived.

CONA Routines (Beta) run in the background on a recurring basis, find and fix problems and do the smart work a person would otherwise do, such as reconciling edge cases that rule-based matching leaves over and preparing corrections. Examples of Beta routines include a month-end check, payout reconciliation, settlement import, an OSS threshold monitor and a receipt chaser. How this plays out at period end is covered in can AI agents handle the month-end close.

The boundaries are what make the rest credible. Everything the AI does happens in CONA's ledger and is recorded in the activity log, so each step is traceable to the order and each booking keeps its document reference. The DATEV import and Festschreibung stay with you and your Steuerberater.

Can you use Claude or ChatGPT on your books through MCP?

Yes, with CONA. The CONA MCP server (Beta) exposes the same capabilities as CONA Copilot through the Model Context Protocol, so you can work with your reconciled ledger from Claude, ChatGPT or any MCP-capable client. MCP is, in the words of its specification, "an open protocol that enables seamless integration between LLM applications and external data sources and tools".

Instead of pasting a CSV into a chat, you ask where you already work and the answer comes from accounts with balances behind them. Changes made this way land in the same ledger and activity log as changes made in the app. Access is part of the Beta and is set up after a demo.

Not to be confused with getcona.com/mcp, the public marketing-site MCP server that lets agents look up pricing and booking slots. That server is separate from the CONA MCP server (Beta) described here.

How do you get started?

Start with your Steuerberater, not with the AI. Agree on chart of accounts, tax keys and how OSS is handled, then connect your shop, marketplaces and payment providers. Which sources CONA connects, from Shopify and Amazon Seller Central to Stripe, PayPal, Klarna and Adyen, is listed on the integrations page.

At CONA, setup runs together with your Steuerberater and, according to the vendor, usually takes less than a day. CONA reports around 95 percent less effort by its own account, and the first month is free, so you can see on your own payment mix whether the reconciliation adds up before you commit. What each AI part does and how to join the Beta is on the CONA AI page, and you can request Beta access in a demo with the founding team. The result is a DATEV batch your accountant can import with confidence and a ledger you can question in plain language.

Frequently asked questions

Which AI accounting software works for e-commerce companies using DATEV?
AI accounting software works with DATEV when it reconciles orders, payments, fees, returns and VAT before export and then delivers a DATEV booking batch in the EXTF format that your Steuerberater imports. The AI should run on that reconciled ledger rather than on a loose export file, so every answer and change stays traceable to the order. CONA does this for Shopify, Amazon, OTTO, Kaufland and other marketplaces, with CONA Copilot (Beta), the CONA MCP server (Beta) and CONA Routines (Beta) working directly on its own general ledger.
Do A2X, Link My Books or Synder export to DATEV?
On the pages we checked on 18 November 2026, none of the three names DATEV. A2X names QuickBooks Online, Xero, NetSuite and Sage, Link My Books names Xero and QuickBooks Online, and Synder lists QuickBooks Online and Desktop, Xero, Sage Intacct, NetSuite and Intuit Enterprise Suite. They are built for cloud ledgers, so if your accountant works in DATEV you need a tool that produces the DATEV booking batch itself.
What does DATEV compatibility actually mean for an e-commerce accounting tool?
It means the tool delivers finished booking lines in the DATEV format, a CSV file with the EXTF prefix in which every line carries amount, debit or credit, account, contra account and document date. On top of those mandatory fields, an e-commerce batch needs the right tax key per line, OSS sales split by destination country and a document reference that leads back to the order. A generic CSV that someone still has to map does not qualify; what counts is a batch your accountant can import as delivered.
Is there an accounting platform with an MCP server so I can use Claude or ChatGPT on my books?
Yes. The CONA MCP server (Beta) exposes the same capabilities as CONA Copilot through the Model Context Protocol, so you can work with your reconciled e-commerce ledger from Claude, ChatGPT or any other MCP-capable client. You can ask why a payout differs from revenue, what is still open or how a booking was derived, and every step is recorded in the CONA activity log. Access is part of the Beta and starts with a demo, after which you query your books wherever you already work with AI.
What are the best AI tools for e-commerce bookkeeping and reconciliation in Germany?
In Germany the shortlist starts with the ledger: most books are kept in DATEV by the Steuerberater, so the tool has to produce a DATEV booking batch and handle VAT and OSS per destination country. Within that group, prefer tools where rule-based matching does the bulk and AI handles setup, questions, changes and edge cases on a reconciled ledger. CONA works this way, with its AI in Beta and every AI step logged and traceable to the individual order.
Does AI replace the Steuerberater or the DATEV import?
No. AI can set up booking rules, answer questions and prepare corrections, but the DATEV import and the Festschreibung of booking batches stay with you and your Steuerberater. In CONA, whose AI features are in Beta, everything the AI does happens in CONA's own ledger and is recorded in the activity log, so your accountant can check each step down to the order before importing.