AI-Native Tax Advisory Firm: Software Task Allocation Concept

Date: 2026-04-03 From: Orcha

Context

We're planning an AI-native Steuerkanzlei (mid-size, 150-500 mandates) that uses three pillars:

• DATEV — complex, regulated tasks (tax declarations, payroll initially, Jahresabschluss, ELSTER filing)
• Orcha — automation of high-volume repetitive processes (document ingestion, pre-accounting, reconciliation, reporting) — with long-term goal to take over payroll
• Claude — ad-hoc consulting, analysis, client communication support

The goal: automate 60-70% of daily Kanzlei work (FiBu + Lohn) through Orcha, keep DATEV for what it's legally required for or does best, and use Claude as the intelligent layer for advisory and exception handling.

Firm profile: Mid-size Kanzlei with multiple Steuerberater, Steuerfachangestellte, and Buchhalter. Delegation workflows, team-based mandate assignment, approval chains for complex matters.

1. Finanzbuchführung (Financial Accounting) — ~40% of Kanzlei work

What Orcha needs to build for FiBu:

• Document ingestion pipeline (live)
• AI-powered Kontierung with SKR03/04 (live)
• DATEV export via Maesn (live)
• Tax compliance validation (live)
• Duplicate detection (live)
• DATEV REWE pull — bidirectional sync to see what's been booked
• Bank reconciliation view — match Orcha invoices against DATEV bank transactions
• UStVA pre-flight check — validate VAT coding before DATEV generates the return
• Automated dunning — OPOS-based payment reminders
• BWA enrichment — pull BWA data from DATEV, add AI commentary via Claude

2. Lohnbuchhaltung (Payroll) — ~20% of Kanzlei work

Phase A: DATEV handles payroll (now → 2027)

Phase B: Orcha takes over payroll (2027+)

Long-term vision: build payroll processing into Orcha to eliminate DATEV dependency. This requires:

• Gross-to-net calculation engine (Steuerklassen, SV-Beiträge, Kirchensteuer, Soli)
• ELSTER integration for Lohnsteueranmeldung
• SV-Meldeportal integration (DEÜV)
• ELStAM interface (electronic tax deduction features)
• eAU integration (electronic work incapacity certificates)
• Beitragsnachweis generation for Krankenkassen
• Certified payroll software status (ITSG certification)

Risk: Payroll certification (ITSG) and SV-Meldeportal integration are significant regulatory hurdles. This is a 12-18 month project minimum.

What Orcha needs to build for Lohn:

• Employee document ingestion — extract data from employment contracts, tax cards, SV certificates
• Travel expense module — receipt capture → Verpflegungsmehraufwand calculation → DATEV export
• Payroll data preparation — digitize monthly movement data (hours, absences, bonuses) for DATEV import
• Payroll engine (Phase B) — full gross-to-net, SV, LSt, ELSTER, DEÜV

3. Jahresabschluss (Annual Financial Statements) — seasonal, ~10%

What Orcha needs to build for Jahresabschluss:

• Closing checklist automation — track year-end tasks per mandate, auto-remind
• Rückstellungen assistant — Claude-powered suggestions based on prior year and current data
• Anhang generator — Claude drafts notes, pulls data from DATEV via Orcha

4. Steuererklärungen (Tax Declarations) — seasonal, ~10%

What Orcha needs to build for Steuererklärungen:

• Declaration data package — auto-compile all relevant documents per mandate per tax type
• Bescheid comparison — OCR tax assessments, compare to filed returns, flag deviations
• Deadline tracking — Fristenkalender with automatic reminders per mandate

5. Steuerliche Beratung (Tax Advisory) — ~5%

What Orcha needs to build:

• Mandate knowledge base — aggregate all financial data, documents, and history per mandate for Claude context
• Scenario modeling — tax burden comparison across entity structures (GmbH vs. GmbH & Co. KG vs. Holding)

6. Kommunikation mit Finanzamt — ~5%

What Orcha needs to build:

• Finanzamt correspondence tracking — log all communication, deadlines, responses
• Audit support module — quick retrieval of any document/transaction the auditor requests
• Letter generation — Claude-powered templates for standard Finanzamt correspondence

7. Mandantenbetreuung (Client Management) — ~5%

What Orcha needs to build:

• Client portal — mandates see their documents, BWA, open items, deadlines
• StBVV billing module — Gegenstandswert-based fee calculation
• Meeting prep automation — Claude generates briefing from mandate data before each meeting

8. Compliance & Meldepflichten

9. AI Workflow Orchestration Layer (the "AI-Native" differentiator)

This is what makes the Kanzlei truly AI-native — not just using AI tools, but having AI orchestrate the daily work.

9a. Inbox Monitoring & Auto-Triage

Orcha monitors all incoming channels (email, DATEV DUO uploads, client portal) and automatically:

• Classifies each item (invoice, Steuerbescheid, Finanzamt letter, client question, Mandantenbeschluss, etc.)
• Creates tasks with the right assignee, priority, and deadline
• Drafts replies for routine client questions (e.g., "When is my Steuererklärung due?" → Claude drafts answer from Fristenkalender, employee confirms/sends)
• Escalates unusual items (e.g., Betriebsprüfungsanordnung → immediate alert to Steuerberater)

flowchart LR
    subgraph Inbound Channels
        E[Email Inbox]
        D[DATEV DUO]
        P[Client Portal]
    end

    E --> T[Orcha AI Triage]
    D --> T
    P --> T

    T -->|Invoice| I[Ingestion Pipeline]
    T -->|Steuerbescheid| B[Bescheid Check → Task]
    T -->|Client Question| Q[Claude Draft Reply → Confirm]
    T -->|Finanzamt Letter| F[Deadline Extract → Alert]
    T -->|Unknown| U[Flag for Human Review]

    I --> K[Kontierung → DATEV]
    B --> R[Compare vs. Filed Return]
    Q --> S[Employee Reviews & Sends]

9b. Morning Briefing & Priority To-Do List

Every morning (scheduled job), Orcha + Claude generate a personalized priority list per employee:

1. Scan all open tasks across mandates assigned to this employee
2. Check deadlines — what's due today, this week, approaching statutory limits
3. Identify AI-completable tasks — mark which tasks Claude can do autonomously or with minimal review
4. Prioritize — statutory deadlines > client-facing > internal > nice-to-have
5. Generate briefing — short summary with action items, delivered via email or dashboard

Example morning briefing:

Guten Morgen, Frau Schmidt — Ihr Tagesplan:

🔴 DRINGEND
  • UStVA Mandant Müller GmbH — Frist heute, Daten vollständig,
    → Claude hat Entwurf erstellt, bitte prüfen und freigeben
  • Einspruch Mandant Weber — Frist morgen
    → Claude hat Entwurfschreiben vorbereitet

🟡 HEUTE FÄLLIG
  • 12 neue Belege Mandant Fischer KG — Kontierung von Orcha vorgeschlagen
    → 10/12 automatisch korrekt, 2 brauchen Ihre Entscheidung
  • Lohnabrechnung Mandant Becker — Bewegungsdaten eingegangen
    → Orcha hat Daten extrahiert, bitte in DATEV LODAS übernehmen

🟢 DIESE WOCHE
  • BWA-Besprechung Mandant Schulz — Donnerstag 14:00
    → Claude hat Briefing mit Trend-Analyse vorbereitet
  • Jahresabschluss Mandant Klein — noch 3 offene Posten
    → Claude kann Rückstellungsvorschläge erstellen, soll ich starten?

✅ AI ERLEDIGT (zur Kenntnisnahme)
  • 47 Belege automatisch kontiert und an DATEV übermittelt
  • 3 Mandanten-Anfragen beantwortet (Entwürfe im Postausgang)
  • Fristenkalender aktualisiert nach 2 eingegangenen Steuerbescheiden

9c. Deep Claude Integration (Mandate-Aware AI Assistant)

Claude has full context on each mandate through Orcha's MCP tools. An advisor can have conversations like:

Example 1: Monthly closing check

"Für Mandant Müller GmbH: Was fehlt noch für den Monatsabschluss März? Zeig mir den Status."

Claude queries Orcha → checks documents received, DATEV booking status, open items, missing Belege → generates:

• Checklist of what's done vs. missing
• List of unbooked documents with suggested Kontierung
• Open items that need clarification
• Visualization of the closing status

Example 2: Client-specific analysis

"Vergleich die Personalkosten von Mandant Weber Q1 2025 vs Q1 2026 und erkläre die Abweichungen."

Claude pulls BWA data from Orcha (via DATEV REWE sync) → analyzes → generates narrative with chart.

Example 3: Ad-hoc advisory

"Mandant Fischer überlegt, einen Firmenwagen anzuschaffen. Was ist steuerlich günstiger — 1%-Regelung oder Fahrtenbuch? Er fährt ca. 30.000 km/Jahr, davon 60% geschäftlich."

Claude runs both scenarios with actual numbers → presents comparison table → recommends.

What Orcha needs to build for AI Orchestration:

• Inbox monitor — watch email/DUO/portal, classify incoming items, create tasks + draft replies
• Task engine — mandate-aware task management with assignees, priorities, deadlines, AI-completability flags
• Morning briefing generator — scheduled Claude job per employee, delivered via email/dashboard
• Mandate context API — aggregate all data per mandate (docs, bookings, deadlines, history) as Claude MCP context
• Reply drafting — Claude generates draft responses for routine inquiries, queued for human confirmation
• Closing status tracker — per mandate per period, what's done/missing/blocked

Architecture Overview

flowchart TB
    subgraph Clients["Mandanten"]
        CL1[Email]
        CL2[Client Portal]
        CL3[DATEV DUO Upload]
    end

    subgraph Orcha["Orcha — AI Automation Layer"]
        TRIAGE[AI Triage & Classification]
        INGEST[Document Ingestion<br/>OCR + Claude Extraction]
        KONT[AI Kontierung<br/>SKR03/04 + Cost Centers]
        VALID[Tax Compliance<br/>Validation]
        RECON[Reconciliation<br/>& Matching]
        TASK[Task Engine<br/>& Priority Queue]
        BRIEF[Morning Briefing<br/>Generator]
        REPLY[Draft Reply<br/>Engine]
        DMS[Document<br/>Management]
        OPOS[Open Items<br/>& Dunning]
        FRISTEN[Deadline<br/>Tracking]
        PORTAL[Client<br/>Portal]
    end

    subgraph Claude["Claude — Intelligence Layer"]
        COPILOT[AI Copilot<br/>Ad-hoc Queries]
        BWA_AI[BWA Analysis<br/>& Commentary]
        ADVISORY[Tax Advisory<br/>Support]
        LETTERS[Letter & Document<br/>Drafting]
        SCENARIO[Scenario<br/>Modeling]
    end

    subgraph DATEV["DATEV — Regulated Processing"]
        DUO[Unternehmen Online<br/>Document Exchange]
        REWE[Kanzlei-Rechnungswesen<br/>Final Books]
        LODAS[LODAS / L&G<br/>Payroll]
        STEUER[Steuern<br/>Tax Declarations]
        ELSTER[ELSTER<br/>Filing]
        VDB[Vollmachtsdatenbank]
    end

    subgraph Employees["Kanzlei-Mitarbeiter"]
        STB[Steuerberater]
        SFA[Steuerfachangestellte]
        BH[Buchhalter]
    end

    CL1 --> TRIAGE
    CL2 --> TRIAGE
    CL3 --> TRIAGE

    TRIAGE --> INGEST
    TRIAGE --> TASK
    TRIAGE --> REPLY

    INGEST --> KONT --> VALID
    VALID --> DUO
    DUO --> REWE

    REWE -->|Journal entries back| RECON
    REWE -->|BWA data| BWA_AI

    TASK --> BRIEF
    BRIEF --> SFA
    BRIEF --> BH
    BRIEF --> STB

    COPILOT --> ADVISORY
    COPILOT --> BWA_AI
    COPILOT --> LETTERS

    STB --> COPILOT
    SFA --> TASK
    BH --> TASK

    REWE --> STEUER --> ELSTER
    LODAS --> ELSTER

    style Orcha fill:#e8f5e9,stroke:#2e7d32
    style Claude fill:#e3f2fd,stroke:#1565c0
    style DATEV fill:#fff3e0,stroke:#e65100

flowchart LR
    subgraph Daily["Täglicher Workflow"]
        direction TB
        A[08:00 Morning Briefing<br/>per Employee] --> B[Priorisierte<br/>Aufgabenliste]
        B --> C{AI-erledigbar?}
        C -->|Ja| D[Claude erledigt<br/>→ zur Prüfung]
        C -->|Nein| E[Mitarbeiter<br/>bearbeitet]
        D --> F[Mitarbeiter<br/>prüft & bestätigt]
        E --> G[Ergebnis in<br/>Orcha/DATEV]
        F --> G
    end

    subgraph Continuous["Laufend"]
        direction TB
        H[Inbox Monitor] --> I[Neue Belege<br/>→ Ingestion]
        H --> J[Mandanten-Fragen<br/>→ Draft Reply]
        H --> K[Steuerbescheide<br/>→ Bescheid-Check]
        H --> L[Fristen<br/>→ Kalender]
    end

Summary: Software Distribution

Estimated automation rates:

• FiBu: 70-80% automated (Orcha + DATEV Automatisierungsservices)
• Lohn: 80-90% automated (DATEV does most, Orcha handles document prep)
• Jahresabschluss: 30-40% automated (still needs heavy human review)
• Steuererklärungen: 20-30% automated (data gathering automated, judgment not)
• Beratung: 10-20% automated (Claude assists research/drafting, human decides)
• Overall: ~60% of total Kanzlei hours automated

Priority Build Order for Orcha

Phase 1 — Foundation (already live)

• Document ingestion (email, upload, API)
• AI extraction (Claude Sonnet)
• Pre-accounting (Kontierung, cost centers)
• DATEV export via Maesn
• Tax compliance validation
• Master data management
• AI copilot (Claude MCP)

Phase 2 — Core Kanzlei (Q2 2026)

• DATEV REWE bidirectional sync
• Bank reconciliation view
• Deadline/Fristen management
• Approval workflows
• Task engine (mandate-aware, with assignees & priorities)
• Inbox monitor & AI triage
• Client portal (basic)

Phase 3 — AI-Native Workflows (Q3-Q4 2026)

• Morning briefing generator (per employee, scheduled)
• Draft reply engine (routine client questions)
• Closing status tracker (per mandate per period)
• Mandate context API (full Claude MCP integration)
• Travel expense module
• StBVV billing
• Bescheid comparison tool
• Meeting prep / briefing generation

Phase 4 — Differentiation (2027)

• Scenario modeling (entity structure comparison)
• Audit support module
• Payroll document preparation (DATEV import)
• Cash flow forecasting
• Client self-service portal

Phase 5 — DATEV Independence (2027-2028)

• Payroll engine (gross-to-net, SV, LSt)
• ITSG certification
• ELSTER direct integration (bypass DATEV)
• SV-Meldeportal / DEÜV integration
• ELStAM interface
• Own Rechnungswesen engine (replace DATEV KaReWe for FiBu)

Verification

• Review with a practicing Steuerberater to validate task completeness
• Cross-reference with StBVV fee schedule to ensure all billable services are covered
• Validate DATEV integration assumptions against Maesn API capabilities
• Test Claude advisory quality on sample tax scenarios