AI infrastructure for tax advisories in Germany.
AI infrastructure for tax advisories in Germany automates recurring tasks such as voucher capture, matching, and filing, without breaching tax advisors' professional conduct rules or GoBD-compliant retention. NexPatch connects systems such as DATEV and your existing document management, so AI agents pre-sort vouchers and practice staff handle the professional review - instead of capturing data by hand.
Two standards every solution for tax practices must meet
AI in a tax practice does not operate in a legal vacuum. Two requirements are binding for every solution, regardless of which specific task is being automated. NexPatch checks both with the practice before a project starts, rather than accounting for them only after a solution is already live.
Professional conduct rules for tax advisors govern, among other things, the duty of confidentiality and personal responsibility for tax advice. An AI solution may therefore only support, pre-sort, and propose - the professional review and sign-off stays with the tax advisor. NexPatch builds automation along that line: AI agents take on recurring, rule-bound steps, while the final professional decision stays with a person. That applies in particular to the duty of confidentiality: client data is processed exclusively within systems the practice controls, with no handoff to uninvolved third parties.
Vouchers and booking entries must be stored, unaltered and traceable, for the full statutory retention period. An AI solution that captures and matches vouchers automatically therefore has to log every processing step and produce a result that meets GoBD-compliant retention requirements, rather than filing vouchers in a separate, unaudited system. If an AI agent changes a match - say, because a voucher was initially misclassified - the original assignment stays traceable in the log rather than being overwritten.
Typical workflow: from voucher intake to booking
In practice, automated voucher capture runs through several traceable steps. A voucher arrives - as a scan, an email attachment, or an upload from the client. An AI agent reads the voucher, matches it to a client, account, and booking circle, and files it in the document management system. Vouchers with a clear, recurring pattern are proposed directly for booking; vouchers with anomalies or unusual details are flagged for manual review. Only after the practice signs off does a booking actually enter DATEV. That final step stays with a person in every case, regardless of how confidently the AI agent assessed the voucher.
Two systems NexPatch connects to
DATEV is the central line-of-business application for bookkeeping and annual accounts in most tax practices. NexPatch connects AI agents to DATEV so that pre-sorted, matched vouchers flow directly into the practice's familiar working environment, instead of landing in a separate system that would need its own upkeep. Staff keep working in DATEV; the AI support adds to that environment rather than replacing it.
The practice's own document management system handles GoBD-compliant filing. Incoming vouchers - by email, scan, or client upload - are captured, classified, and filed automatically in a way that meets retention requirements and stays findable for later review. Follow-up questions to clients, for example about missing or unclear vouchers, can be documented traceably through the same structure, instead of getting lost across separate email threads.
Use case: automating voucher capture
Automating voucher capture is the most common entry point for AI in a tax practice. The following is NexPatch's own model calculation with disclosed assumptions, not a verified customer figure.
| Assumption | Value |
|---|---|
| Voucher volume per month at the example practice | 3,000 vouchers |
| Time per voucher for manual capture and matching | 3 minutes |
| Calculated total time per month without automation | about 150 hours |
| Share of vouchers with a clear, recurring pattern per model assumption | about half |
| Calculated time saved for that share, per month | about 75 hours |
For that share, AI handles pre-sorting and matching, while the practice reviews and signs off. The actual saving depends on the voucher mix, the client base, and how far the practice has already digitized, and is reviewed individually for each practice before it becomes a firm planning figure for the practice team.
How a project starts at a practice
An initial internal pilot using real vouchers from the practice stands within 72 hours, and productive rollout into day-to-day operations follows in 4 to 10 weeks, depending on the system landscape. Which AI agents come into play is shown on the AI agents and automation overview. How NexPatch secures the ongoing operation of solutions like this is described on the operations page. An example of how a comparable project comes together in manufacturing is on our manufacturing reference, and background on automated document processing is available in our knowledge base.