Services
Solutions
Use cases
Industries
Tools
Pricing
Book a call
Industry — Tax advisors & law firms

AI for tax advisors and law firms

§ 203 StGB isn't a footnote — it's a setup prerequisite. Five concrete cases where tax advisors and law firms put to good use, with an honest dividing line between preparation and assessment.

The daily reality AI has to fit into

Morning client call. Receipts in the inbox, in DATEV receipt transfer, in mail attachments. A tax audit query, three incoming notices with deadlines, a reminder to a client who never delivers all the documents. Quarter end is approaching, the next deadline pack is due. And above it all, the duty that not a single one of these documents may end up in the wrong hands.

Tax advisors and law firms mostly work with documents and deadlines — both areas where can genuinely take the load off when integrated cleanly. What doesn't replace: the professional assessment. What makes reliable: preparation, classification, review, first extract. That dividing line is the foundation of every setup in this industry.

Professional secrecy is no checkbox set at the end, but the prerequisite that shapes every architecture decision. In most cases it leads to pipelines or providers with explicit §203-compliant DPAs — cloud standard usually isn't enough.

Five places where AI in the firm really makes sense

For each case: the situation today, where plugs in, a pointer to the matching setup tiers, and an honest trade-off.

01. Receipt OCR with pre-coding

Situation today

Receipts land in the mailbox in the morning, by email at noon, in DATEV receipt transfer in the evening. An assistant sorts for an hour, then case workers take over. With two hundred clients and twenty receipt types, this is above all a constant switch between client accounts.

Where AI helps

extracts amount, date, VAT ID and IBAN, classifies the receipt type and suggests a pre-coding based on the client history. Handover to DATEV/ADDISON/simba for approval — not for auto-booking.

What it can't do

Coding sovereignty stays with the tax advisor. Low confidence automatically lifts the receipt into the review folder — anyone setting the threshold too loose automates errors into the quarter.

02. Clause and deadline review of notices and contracts

Situation today

A notice arrives, somewhere between pages three and four sits an objection deadline, in the next attachment a liability clause with unusual wording. With client takeovers, contracts going back ten years often land on the desk — the first read eats concentration that is then missing in the assessment.

Where AI helps

extracts deadlines, flags unusual clauses against your own reference patterns and hands over a structured advance report. The case worker reviews the report, not the document from scratch.

What it can't do

Legal assessment remains professional work. A flagged clause is a hint, not a verdict — four-eyes review is mandatory here.

03. Firm knowledge base with client history

Situation today

Own templates, earlier opinions, in-house standards for GoBD-compliant process documentation, your own lines of argument in audit disputes — all in the partners' heads or in private Outlook folders. When an employee leaves, part of this knowledge leaves too.

Where AI helps

Your own knowledge base with process SOPs, client history notes, standard clauses and argument patterns — semantically searchable and exclusively on your own infrastructure.

What it can't do

Client-related content falls under professional secrecy. isn't a matter of convenience here, it's a setup prerequisite.

04. Client FAQ bot for standard requests

Situation today

“Which documents do you need for the EÜR?”, “When does the deadline for my objection expire?”, “Who is my point of contact?” — standard questions the secretariat answers thirty times a week, often with identical replies.

Where AI helps

A bot on the client portal or firm website answers standard questions from the curated firm knowledge base, with source attribution and immediate escalation to a human as soon as it gets individual.

What it can't do

The bot doesn't answer on client-specific data (balances, notices, individual deadlines). That dividing line must be set with the data protection officer before go-live.

05. Deadlines from notices into the follow-up system

Situation today

Objection deadlines, audit appointments, action deadlines, extension requests — scattered across notice PDFs, email attachments and handwritten notes. A missed deadline can quickly cost six figures.

Where AI helps

detects deadlines including the legal basis in the incoming document, places them into the follow-up system and triggers a confirmation task for the case worker. It's documented who saw what when.

What it can't do

The first detection is a suggestion. The final deadline belongs in the firm's deadline book — with human confirmation, not from a log.

What in firms isn't (yet) working

Four promises that turn disproportionately expensive in firms, because an error here doesn't just cost efficiency, but mandates and liability:

AI as a substitute for tax advisor or lawyer

hallucinate on sections of law, calculations and references to BMF letters. Even for a “simple” EÜR, the error rate is high enough that reviewing becomes more expensive than manual preparation. is preparation, not authoring.

Fully automatic handling of objections and audit responses

Lines of argument in disputes need context, client knowledge and a feel for negotiation. can deliver building blocks, the letter belongs in human hands — otherwise the client lands in the next dispute.

Auto-answering individual client questions

As soon as the question goes beyond standard processes (“what does this notice mean for my GmbH?”), a human must answer. Otherwise liability questions arise that dwarf any efficiency gain.

Client data in public cloud LLMs without EU DPA

§ 203 StGB isn't handled with a DPA checkbox. Cloud models without EU hosting and without an explicit professional-secrecy-compliant agreement are practically excluded in firm practice.

What absolutely has to be thought through for AI in the firm

Four legal pillars against which every setup in tax and law is measured:

§ 203 StGB — professional secrecy

Client-related content is subject to criminally enforceable professional secrecy. setup begins with the question: what is each component even allowed to see? In most cases this leads to or a strict DPA with a professional-secrecy clause.

BStBK or BRAK professional rules

Marketing, client retention, file keeping — the professional rules regulate more than many vendors realise. A client FAQ bot has to fit BStBK rules, a firm knowledge base too.

GoBD and process documentation

Immutability, retention periods, traceability — automated receipt workflows have to map the GoBD requirements. Audit trail and versioning are not a bonus here, but a prerequisite for the next audit.

AI literacy under Art. 4 EU AI Act

Since February 2026, firms as employers are required to train staff in the proper use of — especially in tax and law, where hallucinations can directly trigger liability.

Tools that often already run in firms

doesn't replace these systems — it plugs in. Where the interface typically sits:

Accounting and receipts

DATEV (DMS, receipt transfer, Eigenorganisation), ADDISON, simba, BMD, agenda — typical plug-in point is the DATEV interface or direct client receipt transfer

Law-firm software

RA-MICRO, advoware, NoRA, AnNoText, LegalVisio — pipelines plug in via email gateway, document intake or follow-up interface

Document management

DATEV DMS, ELO, d.velop, M-Files, Habel — prerequisite for clean receipt workflows, often already in use

Communication

beA (the special German electronic lawyer mailbox), KIM in law firms with eHBA connection, DATEV client portal, Adobe Sign / DocuSign for signatures

How firms typically get started

Anyone starting without experience has two clear candidates — and one where many fail when they pick it as the first step.

Typical entry point 1 — receipt OCR with pre-coding

Immediately measurable relief in case work, clean connection to DATEV or ADDISON. Risk low, because every suggestion passes through human approval.

Typical entry point 2 — deadline detection in the follow-up system

High risk lever with immediate impact. Combined with human confirmation, the deadline book becomes more reliable than pure manual work — and the audit trail documents it.

Don't start with the client bot

The temptation to push standard questions straight onto a bot is large. Without clear escalation rules and a maintained knowledge base, the bot disappoints clients and secretariat in the first week — and the firm's reputation goes with it.

Funding for firms

SMB consulting funding also applies to tax advisor and law firms — the subsidy covers the conception and introduction phase, not ongoing software licences. For investment components (server hardware, licences), state-level digital-bonus programmes or Mittelstand-4.0 initiatives come into play depending on the federal state. For §203-relevant setups, the investment in infrastructure is typically not eligible, but the consulting for it is.

→ Details on BAFA funding
FAQ

Frequently asked questions about AI in tax and law firms

Client data is professional secret — external processing requires explicit consent or strict data processing agreements. We primarily use models on premises or EU hosting with a DPA. Cloud for client matters only after a vetted data classification.
AI-captured documents must be archived unaltered, complete, and verifiable — like any other voucher. We design the capture step so that the original document, extraction, and manual approval are stored versioned. GoBD-compliant interface into DATEV or Addison.
Yes. DATEV offers REST APIs (DATEV Connect, DATEV DMS) and CSV export. We build adapters that feed approved results as booking suggestions or incoming documents. You decide what actually gets booked — does nothing unprompted.
For typical (invoice capture, client FAQ, incoming invoice check) usually 6 to 12 months. With funding sooner. We recommend starting with that have measurable hour savings rather than full automation from day one.
No and yes. It does not replace professional judgment — that stays with the consultant. But generative can produce plausible-sounding wrong statements. Therefore: never let advise clients standalone, always use it as a research and preparation tool with professional final review.

Ready for the next step?

Free intro call, no strings attached. In 30 minutes you'll know whether and how AI can help your business.

Book a callCheck eligibility
Book a callBAFA funding