A tax firm sells knowledge and diligence, but a considerable part of the day is spent on a third thing: documentation. The annual review with the client needs minutes, the phone advice on forming a company needs to be captured, the agreed deadline entered into the system, the confirmation email written. With an acute talent shortage, this is the work firms want least and have most.
At the same time, few industries are rightly as cautious about AI: client data is professional secret. This article shows what an AI agent concretely takes over in a firm, and which data protection requirements a tool must meet for it.
Why tax advisors are the perfect and the hardest case for AI
Perfect, because the firm's raw material is speech: consultations, calls, coordination with clients, tax authorities and colleagues. That is exactly what AI processes best today. Hard, because tax advisors are bound by professional secrecy (in Germany, Section 203 of the Criminal Code): whoever entrusts client data to a service provider must select them carefully and bind them contractually to confidentiality. Add the GDPR, and after Schrems II, US cloud tools that transfer data to the US are legally delicate to unusable for client data.
The consequence is not "no AI", but: AI with a German data location and clear contracts. Sally is operated by Aliru GmbH from Mannheim and processes exclusively in Germany, without third-country transfer. Details are on the page about GDPR and security.
The biggest lever: conversations document themselves
The core of an AI agent in the firm is unspectacular and precisely for that reason valuable: every conversation automatically leaves a result.

Annual and financial statement reviews: the minutes write themselves
The annual review is the densest conversation of the engagement year: results discussion, structuring considerations, open points for next year, private topics of the client that might become tax-relevant. Sally attends the meeting, via video in Google Meet, Zoom, Microsoft Teams and Webex, or in person through the app. Afterwards you have the transcript, a summary and the discussed points, speaker-separated and searchable. The advisor leads the conversation instead of taking notes in parallel, and the client can receive a clean summary instead of nothing.
Initial consultations and calls: liability precaution on the side
The advice on choosing a legal form, the recommendation on the investment, the note about the deadline: in a firm, documentation is not just tidiness, it is protection. What was recommended, and with which caveat, is provable in the documented conversation, verbatim and dated. If the question "was I advised about this?" comes up years later, a source exists. The short client phone call in between, today the worst-documented conversation type in any firm, is captured via the app just like the big meeting.
Tasks and deadlines: side remarks become tracked items
Sally recognises tasks in the conversation automatically, including owners and dates, more on the page about task recognition by AI. "Submit documents by 15 August" becomes a tracked task instead of a margin note, and on both sides: the client's promise to deliver the receipts is captured too. Especially in firms where missed deadlines are directly liability-relevant, that is worth more than any reminder app.
From the conversation into the firm's processes
Documentation alone would be another silo. The agent idea is that results flow to where the firm works.
Firm tools and DATEV: no double entry
Sally connects to 8,000+ tools, including DATEV, calendar, task and email systems. Conversation results and deadlines land inside existing processes, not next to them. That is the difference between yet another tool that wants to be maintained and an assistant that feeds the systems you already have. Nobody retypes what has already been said.
Client communication: confirmations with substance
The confirmation email after the meeting is drafted from the real content, with the actually discussed points and deadlines instead of text blocks. For the client, that is tangible service quality: the same day, they get in writing what was agreed and what they need to deliver by when. For the firm, it is the fastest way to stay within the 24-hour rule; why that matters is covered in our article on the follow-up email after a meeting.
Client history: the firm's memory
All documented conversations form a searchable knowledge base. "What did we recommend to client Krüger about the holding structure in 2024?" is a seconds-long search query, even if the colleague from back then has left the firm. During staff changes, holiday cover and the handover of engagements, that is the difference between "read the file and hope" and "read what was actually discussed". For firms in a generational transition, this history is worth real money.
And the direction is clear: Sally is evolving into an assistant that takes instructions directly, from task entries on demand to email drafts built from the conversation context. The firm stays in control: drafts are reviewed and approved, not sent blindly.
Introducing it legally clean: the three points
Client consent
Secret recordings are prohibited (in Germany, Section 201 of the Criminal Code); documentation happens only with agreement. One sentence at the start of the conversation solves it in the meeting, and firms that want it systematic add a note about conversation documentation to the engagement letter. Practice shows many clients explicitly welcome it, because what was discussed and recommended is recorded for them too.
Data processing agreement and confidentiality
The provider is brought in properly: a data processing agreement under Article 28 GDPR, a confidentiality obligation matching professional secrecy rules, disclosed sub-processors. Exactly these points should be checked with any AI vendor before the first client conversation is documented. With Sally, they are standard.
Data location and training exclusion
Processing exclusively in Germany, no US transfer, no use of client data to train AI models. The last point is often overlooked: with some US tools, the terms of service grant the vendor training rights, a knockout criterion for professionals bound by secrecy.
Staying honest: limits
An AI agent does not judge tax matters and does not sign annual financial statements. Professional responsibility stays with the qualified advisor; every automatically created summary and every email draft is a working state for review, not a final product. The realistic gain lies in the writing and transfer work around conversations, and in a conversation-heavy firm day, that gain is substantial.
Conclusion
For tax firms, the AI entry point with the highest benefit and the lowest risk is conversation documentation: automatic, GDPR-compliant, secrecy-proof, with tasks, deadlines and client history as by-products. Which tools help beyond that is covered in our overview of AI tools for tax advisors.
The best test is your next client meeting: test Sally free for 30 days, hosted and developed in Germany.




