AI in tax advisory: time back in the firm, not a conflict of principles

In short: In tax advisory, the value of AI does not come from spectacular full automation but from the disappearance of waiting time, follow-up queries and double entry. A quarter of German firms now use generative AI regularly — a year earlier it was one in eleven. The bottleneck is not the technology but the question of which steps in a firm’s workflow actually need a human.

Where does tax advisory really stand in 2026?

The profession is growing, its talent pipeline is not. As of 1 January 2026, 105,953 members were registered with the German chambers of tax advisors, including 89,549 tax advisors — up 0.6% year over year. Over the same period, the number of trainee contracts for tax clerks fell for the third consecutive year to 17,081 (2024: 17,301; 2023: 17,355). The average age in the profession remains high at 53.7 years (source: Bundessteuerberaterkammer, Berufsstatistik 2025, reference date 1 January 2026).

That is the real starting position: workload grows, headcount does not. At the same time, the value mix is shifting. According to STAX 2024, the representative firm survey run by the federal chamber with 5,815 participants, bookkeeping fell from 25.9% of firm revenue (2018) to 21.4%, and annual financial statements from 28.4% to 24.8%. The same study found that the higher a firm’s level of digitalization, the more positive its revenue development.

Classic volume work carries less weight than it used to — and that is exactly where AI lands first.

How many firms already work with AI today?

One in four — and the curve is steep. In the DATEV Seismograf 2025, 25% of the firms surveyed said they use generative AI regularly. A year earlier the figure was 9%. The term itself has arrived almost everywhere: 92% of firms know it, but only 14% rate their own competence as good (source: DATEV Seismograf, survey run 2–15 June 2025, 504 responding firms out of 7,500 invited).

That gap between awareness and competence is the most interesting number in the whole survey. It means the advantage in 2026 does not come from having heard about AI but from having actually rebuilt one process around it. For comparison, look at the client side: 41% of German companies actively use AI, but only 21% have a formal AI strategy (source: Bitkom AI Study 2026, n = 604 companies with 20+ employees). On both sides of the engagement there is more movement than structure.

What does a day in an AI-supported firm actually look like?

It looks unspectacular — and that is the point. The difference is not a new interface but the absence of waiting loops.

Morning, document intake. Instead of a shared folder full of mixed PDFs and phone photos, receipts arrive pre-sorted: recognized, coded, assigned to the right client and period. Anything unambiguous passes through. Anything unclear — a missing service period, an implausible VAT rate, a duplicate entry — is pulled out as an exception with a specific question attached. Staff no longer review everything, only the exceptions.

Late morning, client communication. Nobody types the standard follow-up email anymore. The draft is generated from the exception itself and references the specific document. A human reads it, corrects it, sends it. Missing paperwork is chased continuously rather than collected into one end-of-month complaint, which visibly reduces the pile-up before the VAT return.

Midday, technical research. A VAT treatment question is no longer hunted across three databases; it is asked. The system returns a proposal with citations. The professional checks the citations, not the phrasing.

Afternoon, drafting. Appeal justifications, client letters, file notes: the first draft comes from the system, the qualified head decides. This is also where the DATEV survey locates the most-used applications — drafting and research assistance, not autonomous case handling.

What does not happen in that day: the AI makes no tax decision. It clears the path to one.

Why is e-invoicing the actual starting gun?

Because it supplies the data foundation without which any automation stays patchwork. Since 1 January 2025, every German company must be able to receive and process e-invoices from other German companies. From 1 January 2027, the obligation to issue them applies to companies with prior-year revenue above 800,000 euros, and from 1 January 2028 to everyone else (source: § 14 German VAT Act as amended by the Wachstumschancengesetz; German Federal Ministry of Finance FAQ on mandatory e-invoicing). What counts is a structured, machine-readable data set under EN 16931 — XRechnung or ZUGFeRD. A PDF explicitly does not qualify.

For firms this means the share of incoming data that already arrives structured will rise unavoidably over the next two years. Align your processes now and the automation is effectively delivered to you. Wait, and you will rebuild under time pressure in 2028. For once, the regulatory calendar is an ally — it makes the investment plannable.

Where does AI stop inside a tax firm?

At confidentiality and at responsibility — and neither is negotiable. In its FAQ catalog “AI in tax advisory” (as of 27 January 2026), the federal chamber of tax advisors made it explicit: using AI tools neither lifts nor weakens the professional duty of confidentiality under § 57 StBerG and § 203 of the German Criminal Code. Confidential client data may only be passed to AI providers if those providers are strictly bound to confidentiality in law and by contract. Ultimate responsibility for accuracy and completeness stays with the firm.

Then there is the technical risk that serious vendors should not play down: language models fill gaps with plausible-sounding but false statements. In a field where an invented citation becomes a liability case, a human final review before anything leaves the building is not a comfort feature but part of the process. And the AI literacy obligation under Article 4 of the EU AI Act has applied since 2 February 2025 — if you deploy AI, you must qualify your team for it.

Honesty also requires this: the circulating promises of twenty hours saved per week do not hold up. What is realistic is a strong effect on repetitive, rule-based steps and a smaller one on anything requiring judgment. Anyone trying to convince a firm with a blanket percentage has not looked at the process.

How does a firm start small?

With a process, not a platform. A pragmatic 90-day entry looks like this:

• Pick one bottleneck that actually hurts. Usually that is document intake or chasing missing paperwork — high repetition, clear rules, a measurable backlog.
• Measure the baseline before changing anything. How many exceptions per client per month? How long from document to booking? Without that number there is no proof later.
• Settle data protection and professional law up front, not afterwards. Data processing agreement, the provider’s confidentiality obligation, processing location, no model training on client data. In D-A-CH this is not a brake but the argument you make to your clients.
• Pilot with a small group, not the whole firm. Two or three clients, four to six weeks, a fixed final review, documented errors.
• Decide after the pilot — including against rollout. A pilot that proves a process unsuitable has still paid for itself.

This is exactly where weooo works: not dropping another tool into an organically grown firm, but shaping the workflow so AI takes hold where it measurably saves time — and stops where professional law and responsibility begin. Data protection compliance is the precondition here, not the bonus round. How such a digital employee is scoped in general is covered in Digital employees: what AI agents really deliver in mid-sized companies.

Conclusion

Tax advisory is not an industry where AI replaces qualified professionals — it is one where AI takes over the supporting work for which staff are increasingly hard to find. The numbers show a profession that grows while its trainee intake has declined for three straight years, and volume work that carries a shrinking share of revenue. A quarter of firms have started, but only 14% trust their own competence. Rebuild one process properly now, with confidentiality and final review designed in, and 2028 will not be a scramble — it will be an advantage.

Want to know which process is the right first one in your firm? Talk to us — we look at the workflow before we talk about tools.

Sources

• Bundessteuerberaterkammer, Berufsstatistik 2025 (reference date 1 January 2026, published April 2026) — 105,953 chamber members, of which 89,549 tax advisors (+0.6%); average age 53.7 years; 17,081 trainee contracts (2024: 17,301; 2023: 17,355); 53,932 practices, 67.1% of them sole practices.
• Bundessteuerberaterkammer, STAX 2024 (representative profession survey, 5,815 participants, 25.1% response rate) — bookkeeping share of revenue 25.9% (2018) → 21.4%; annual financial statements 28.4% → 24.8%; average revenue of professional practice companies €1,265,000 (2018: €922,000), sole practices €305,000 (2018: €257,000); positive correlation between level of digitalization and revenue development.
• DATEV Seismograf 2025 (online survey 2–15 June 2025; 504 responding firms out of 7,500 invited) — regular use of generative AI 25% (prior year: 9%); term known to 92%; 14% rate own competence as good.
• Bundessteuerberaterkammer, FAQ catalog “AI in tax advisory”, as of 27 January 2026 — confidentiality under § 57 StBerG / § 203 StGB remains untouched; confidential client data may only be shared with providers strictly bound to confidentiality in law and contract; ultimate responsibility and human final review stay with the firm.
• § 14 German VAT Act as amended by the Wachstumschancengesetz; German Federal Ministry of Finance FAQ on mandatory e-invoicing — receiving obligation since 1 January 2025; issuing obligation from 1 January 2027 (prior-year revenue > €800,000) and from 1 January 2028 for all others; EN 16931 format (XRechnung/ZUGFeRD), PDF insufficient; exemptions for small-value invoices up to €250 and B2C.
• Regulation (EU) 2024/1689 (AI Act), Art. 4 — AI literacy obligation for staff since 2 February 2025; further provisions from 2 August 2026.
• Bitkom, AI Study 2026 (n = 604 companies with 20+ employees) — 41% actively use AI, 21% have a formal AI strategy.