Moritz Diedrich

Date: 11.11.2025 (2025-11-11)

Moritz Diedrich (11.11.2025) • https://www.linkedin.com/in/moritz-diedrich/ • Head of Finance at Sastrify (58ppl on LI) • Wir entwickeln proprietäre KI-Modelle für Controlling & Accounting, die auch für komplexe Fälle trainiert sind - etwa mehrere Purchase Orders auf einer Rechnung, Price-Variance-Tolerances bei Lieferanten etc. Ziel ist eine ~60% Automatisierung in AP/AR und auch 50% weniger Aufwand beim Closing & Reporting - ohne neues SaaS-Tool, direkt integriert in bestehende Systeme (auch ohne API) und angepasst an aktuelle Prozesse. • Per Du

Notes: • Notes: • Built Sastrify up - on peak 200 employees • Hypergrowth startup 2021/2022 • Focus on profitability • He leads the finance team • B2B SW company • Cologne based • He likes talking about it • He gets a lot of LinkedIn messages • He thinks to found in the finance direction • Automation: • Segment B2B SaaS up to 200 employees -> finance is a big topic • More and more automation • Closing automation -> Check out Numeric • Next step: complete automation • Bookkeeping is very manual • Abgrenzung, Rückstellung etc. • There are a lot of positions on Stepstone -> almost the most common job that is searched for • WTP based on human labour • A lot of competition • Best tool FP&A: Excel • It is very flexible -> kleinstmöglicher Nenner that covers all • Check out Pigment, Agicap (also problems - not enough automated), Hayde, Lucanet • Best tool Accounting: Numeric • No tool that replaces Datev • They use Chargebee • Others are Chargify, Stripe
• They use Spendesk and use Datev in Accounting • GDPDU export -> analyse with AI • -> reporting on top of SuSa • Challenge: finance has very accurate -> until processes are not 99% correct someone needs to check (95% is not enough) • Human is also not perfect but its a mental hurdle • There are edge cases that are complex • FP&A is very strongly besetzt • Still a lot of space but a lot of companies -> accounting is not that crowded • Team: • He as lead • 1 very operational -> accounting > all inhouse • 1 for FP&A and operational • Working student • FP&A is quite automated -> e.g. investor reporting • Data model pitch is not really clear • He thinks integration and data cleaning is the challenge • Different data from CRM vs. Chargebee -> edge cases • Very cool guy -> keep him involved • He thinks Moss und Candis are out of the game -> outdated • Spendesk and Candis are not in the competition anymore • They are too slow / too big • Spendesk is very slow releasing features • He thinks about founding > could imagine to found -> he asked me if we need someone to join • Biggest objections / challenges in his daily life? • • Most valuable flows for him? • • Context: team setup? internal and external? • • Which tasks are the biggest pain? With magic wand -> what would he solve? • Classic accounting -> he would replace one person with me • Datev, Spendesk, Abstimmung betweens teams • Reporting building • Works almost with the press of a button -> many sheets • Budget • Forecast • Actuals • Revenue on customer analytics • A lot of KPIs • Abweichungsanalysen • Sanity checks: sometimes different ARR • Download ARR per customer and compare - e.g. first year discount • CRM (Hubspot data vs. Accounting data) -> discrepancies • Which tasks would he focus on? • • Do they already have automations? Which ones? How well do they work? • • Which companies would he target? • • General feedback to pitch? •

– – – Summary done (see Appendix below) – – –

Appendix: Interview Summary 11.11.2025 • https://chatgpt.com/c/69137323-5afc-832c-bb17-35144f294611 • See transcript below

A. Procurement, Purchasing, & Supplier Management A1. Existence of buying specialists / external purchasing agents (n=1)  A2. Wareneingang confirmation required (n=2)  A3. Procurement processes are complex / redundant systems across legal entities (n=2)  A4. Tendering processes in public institutions (n=1)  A5. Some invoices must be downloaded manually from supplier portals (n=1)  A6. Supplier negotiation uses data like volume changes, payment terms, average payment date (n=4)  A7. Some companies use supplier credit cards for teams (n=1)  A8. Procurement potential identified (esp. retail, e-commerce, production) (n=4)  A9. Per-delivery documentation: Lieferschein, invoice, PO, etc. (n=2)  A10. Contract analysis including penalties is done manually and is difficult (n=1)  A11. Benchmarks considered unreliable; industry-specific playbooks used instead (n=1)  A12. ERP-to-ERP invoice transfer exists but only for large suppliers (n=1) B. Accounts Payable (AP) & Accounts Receivable (AR) B1. AP/AR highly manual (n=15+)  B2. Missing invoices for expenses (20–50%) (n=2)  B3. Manual reconciliation / problematic reconciliation (n=12+)  B4. Multiple invoices per PO, multiple POs per invoice (n=4)  B5. Fraud detection relevance (duplicate payments, self-approval, bank account change) (n=5)  B6. Invoice matching problems (especially goods received vs. invoice) (n=6)  B7. AR automation exists in some companies and works well (n=3)  B8. Companies with standard AR formats automate easier; AP is harder (heterogeneous formats) (n=3)  B9. Some companies scan invoices manually; some still paper-based (n=3)  B10. VAT / Umsatzsteuervoranmeldung manual or partially manual (n=4)  B11. Sconto importance varies: very important in some markets, absent in UK (n=3)  B12. AP/AR tools often fragmented by country (local requirements) (n=2)  B13. Tax advisor workload on AP/AR is high; data quality from clients is inconsistent (n=4) C. Accounting, Pre-Accounting & Tax C1. Accruals / Abgrenzungen are complex and often manual (n=10+)  C2. Reconciliation effort high; monthly reconciliation desired but rare (n=6)  C3. VAT rules vary by country; monthly VAT important (n=4)  C4. Lack of talent in tax advisory; high fluctuation, high sick leave (n=4)  C5. Tools like DATEV, LexOffice, MOS, Candis, etc. widely used but limited (n=10+)  C6. Integrations into DATEV are painful; GDPdU files important source (n=5)  C7. Tax advisors depend heavily on client data; reminders often ignored (n=2)  C8. Manual work due to missing/corrupted data from subsidiaries (n=3)  C9. Regulatory differences across countries (invoice fields, rules, formats) (n=4)  C10. Rückstellungen at month-end painful (n=2)  C11. Paper-based accounting still exists (n=2) D. FP&A, Controlling, Budgeting D1. Controlling is highly manual and fragmented (n=10+)  D2. Tools often insufficient; Excel heavily used for FP&A (n=12+)  D3. Budget vs. actuals comparison important; often flagged incorrectly (n=3)  D4. Profitability analysis complex (per product, per client, per geography) (n=4)  D5. Need for scenario simulation tools (n=2)  D6. Distribution of fixed costs challenging (Kostenstellen/Kostenträger) (n=3)  D7. Consolidation of subsidiaries is a major pain (n=3)  D8. Cash flow planning and forecasting are high-value (n=5)  D9. FP&A tools crowded; Excel remains fallback (n=3)  D10. Manual formatting for reporting/PowerPoints (n=2) E. Systems, Integrations, Data Models & Technical Landscape E1. Legacy systems everywhere; many ERPs per company (n=6)  E2. Multi-system redundancy common (n=3)  E3. Integrations are difficult and time-consuming (n=8)  E4. Bank integrations often fail or unreliable (n=3)  E5. Need for a unified data model (objects + events) (n=2)  E6. Datev/SAP integrations often painful (n=6)  E7. RPA rarely used in practice due to instability (n=4)  E8. Browser automation valuable but OS-dependent (n=2)  E9. Some companies use SFTP/CSV heavily (n=3)  E10. Some consider directly reading databases where no API exists (n=1) F. AI & Automation F1. AI adoption still low; POCs exist but production rare (n=8)  F2. Need for explainability and transparency (n=2)  F3. AI must adapt to changing processes (n=2)  F4. Self-healing RPA would increase willingness to adopt (n=2)  F5. Training agents with SOPs improves stability (n=1)  F6. Concern: smart fraudsters evolve with AI (n=1)  F7. Accounting automation level expected: >95% accuracy needed (n=1)  F8. Need for anomaly detection (bank account changes, unusual amounts) (n=3)  F9. Need for AI that scales with company size and complexity (n=2) G. Country Differences G1. UK: QuickBooks strong; Sconto not used (n=3)  G2. Finland: strong API ecosystem (n=1)  G3. China: no IBAN; transfers more complex (n=1)  G4. USA: QuickBooks widely used (n=1) H. Organizational Dynamics & Talent H1. Many accountants/AP staff under pressure and burnout (n=2)  H2. High turnover, low skill consistency (n=3)  H3. Dependence on 1 person knowing automation (bus factor) (n=1)  H4. Companies often rely on temporary workers for repetitive tasks (n=1)  H5. Many teams still manually chase people for invoices/receipts (n=3) I. VC, Market & Commercial Insights I1. VCs see messy accounting in portfolio companies (n=1)  I2. Potential to sell via VC portfolios (n=1)  I3. Pre-seed ARR expectations 50–60k; seed 300–500k (n=1)  I4. People expect price pressure once competitors exist (n=1)

Appendix: Interview Summary 24.10.2025 Note: 3 FRI interviews not included yet.

Legacy Software & System Fragmentation Mentioned often • Legacy systems are widespread and critical — especially in real estate, manufacturing, finance, and construction. They are deeply embedded in operations and thus not easily replaceable (Luca, Marcel, Pedro, John, Siegfried). • System silos and heterogeneity create massive friction. Subsidiaries or country-level setups often run different ERPs or disconnected tools (Luca, John, Simon, Siegfried). • On-premise, non-API systems are still common, though some sectors (banking, newer mid-market firms) are slowly modernizing (Marcel, Maximilian Rüppel). Mentioned a few times • Service or enterprise buses (e.g. Sophico Orchestra, ESB) act as bridges across systems in manufacturing. They can handle near-real-time data but not millisecond-critical data (Mike). • Companies resist replacing legacy systems because they form the operational backbone. Instead, they layer automation or RPA on top (Luca, Pedro). Mentioned once • Some industries (e.g. parts of healthcare) are less legacy-bound, but regulated ones (banking, insurance, government) are heavily so (Dominik). 
Data Quality & Integration Gaps Mentioned often • Data is poorly maintained or incomplete — employees often skip entering it altogether (Luca, John). • Integration challenges between systems (e.g., DocuSign ↔ OpenText, ERP ↔ PIM, energy software) are recurring pain points (Shyam, Maximilian Rüppel, Carl). • APIs are inconsistent or insufficient, especially in accounting (DATEV, HubSpot) — forcing RPA or browser automations (Pedro, Carl). Mentioned a few times • Firewalls and IT restrictions in industrial setups limit automation or data flow (Mike, Shyam). • Manual Excel-based workarounds remain common for data consolidation and reporting (John, Shyam). 
Automation & AI Opportunities Mentioned often • RPA and workflow automation are widely used to bridge legacy gaps — e.g. make.com, n8n, Power Automate, macros, etc. (Pedro, Carl, Maximilian Rüppel, Shyam, John). • Automation potential in finance/accounting is consistently highlighted — reconciliation, invoice matching, billing, reporting (John, Carl, Maximilian Rüppel, Lasse). • ROI and simplicity are key — companies want quick wins and low setup effort before scaling automation (Simon, Thomas, Maximilian Rüppel). Mentioned a few times • AI agents are seen as too early or impractical in high-speed or highly regulated environments (Mike, Thomas). • Standardized templates for common automations could accelerate adoption and reduce shadow processes (Thomas). • Building a reliable data layer first is often necessary before automation can work (Maximilian Rüppel). Mentioned once • Browser automation seen as a promising new way to access legacy systems, but reliability and error handling are challenges (Carl). 
Organizational & Cultural Constraints Mentioned often • IT departments act as blockers — security, validation, and regulatory processes make integration and automation difficult (Shyam, Thomas, Siegfried). • Manual work persists due to lack of in-house expertise; business users depend on external consultants for technical setups (Marcel, Maximilian Rüppel). Mentioned a few times • Shadow processes emerge when official systems don’t meet user needs, causing technical debt (Thomas). • Regulatory validation (especially in healthcare/medical devices) slows software changes (Shyam, Dominik). • Subsidiaries operate semi-independently, leading to duplication and inconsistent processes (John, Siegfried). 
Industry-Specific Observations Mentioned often • Construction: Extremely fragmented; local vs. international complexity; chaotic ERP and BI landscape; high potential in procurement and accounting automation (John, Simon, Siegfried). • Manufacturing: Real-time production control is millisecond-sensitive; less room for LLM-type agents but viable for higher-level orchestration (Mike). • Finance & Banking: Legacy is improving; low-code and outsourcing of process setup are trending (Marcel, Thomas). • Healthcare / Regulated industries: Bureaucratic overhead and software validation are major slowdowns (Shyam, Dominik). Mentioned a few times • Procurement: manual price requests, inconsistent negotiations, potential for supplier platforms or booking.com-like marketplaces (Simon, Siegfried). • Document management & delivery notes: persistent manual scanning and re-entry (Simon). • Time tracking & compliance scheduling: repetitive, regulation-heavy, high automation potential (Simon). 
Strategic & Product Insights Mentioned often • Start with narrow, ROI-proof use cases — small automations that demonstrate clear value (Simon, Thomas, Maximilian Rüppel). • Standardization over deep customization — mid-market firms don’t have resources for bespoke automation (Lasse). • Data privacy and on-premise preferences are strong, especially in Europe and regulated sectors (Siegfried, Shyam). Mentioned a few times • Outsourced implementation support and “automation as a service” models are valued because firms lack in-house technical skills (Marcel). • Potential in holding structures — automating shared services across portfolio companies (Lasse). • ERP advisory and integration firms are expensive but fill the skills gap (Maximilian Rüppel). 
Emerging Concepts & Tools Mentioned • Tools: n8n, make.com, Power Automate, Commander, Sophico Orchestra, Shopify, PIMcore, HubSpot, OpenText, DATEV, Construct, EDI. • Ideas surfaced: • Booking.com-style supplier marketplace (Siegfried). • Enterprise service bus / integration fabric for non-millisecond data (Mike). • Browser automation for legacy UI interaction (Carl). • Predefined process templates to speed adoption (Thomas). Appendix: Interview Summary 18.10.2025 Summary Legend (frequency tags) • Frequent = came up many times across interviews • Repeated = mentioned by multiple interviewees • Single = mentioned once  Core Themes Across Interviews (neutral statements, frequency-tagged) A. Accounting & Finance Operations • DATEV constraints: Weak API, hard to extract data for analytics/EoM; perceived monopoly; very low pricing. (Frequent) • Excel as the backbone: Finance/controlling heavily relies on Microsoft Excel/Sheets; copy-paste between systems → Excel → PowerPoint/email. (Frequent) • Manual invoice work: Matching invoices to transactions; categorizing expenses; PO/contract comparisons; accruals handling; double-payment checks. (Frequent) • Investor reporting load: Monthly investor reporting costs significant time (e.g., ~8h/month at 40 ppl startup). (Repeated) • Budgeting/forecasting: Need variance explanations (actuals vs budget), upload/compare forecasts. (Repeated) • SME invoicing gaps: Few good tools for sending invoices to customers (SMBs). (Repeated) • Travel expenses: Manual, multi-step approval and payout processes. (Repeated) • KYC: Document collection heavy. (Single) B. Automation Landscape & Org Setup • Power Automate trend: Widespread/ growing use; RPA capabilities exist but considered immature/poor by some. (Frequent) • n8n / make.com usage: Adopted for glue workflows; integrations (incl. DATEV via n8n). Complexity can balloon (e.g., 200 nodes). (Repeated) • Decentralized automation: Individuals/teams build their own automations; lack of central automation owner. (Repeated) • RPA fatigue & fragility: Processes break when UIs change; traditional RPA perceived as brittle; some say interest is declining while organizations “wait for AI.” (Repeated) • Deterministic first: Move back to deterministic workflows; LLM used to call tools rather than drive entire process. (Repeated) • Throughput blockers: Human approvals pause automations; serial bot execution; testing is hard; some roll out without full tests. (Repeated) C. Data & Integration Constraints • Poor/incomplete APIs: Across ERPs/finance tools; CSV/SFTP still common; ERP integrations costly (3–4 months). (Frequent) • Fragmentation: Many disparate systems; local tools per country/entity due to taxes/pricing/policy → data silos. (Frequent) • Access hurdles: MFA/2FA/USB keys; VPN to internal apps; credentials/setup friction. (Repeated) • Vendor lock-in concerns: Especially with Palantir; desire to avoid lock-in. (Repeated) • EDI: Still used for supplier links. (Single) D. Desired Capabilities (verbatim needs, non-evaluative) • “Digital finance manager”: Replace a working student; AI preps everything so reviewer spends ~30 minutes. (Repeated) • Variance diagnosis: Automatic explanations for actual vs budget gaps. (Repeated) • Automated approvals: E.g., auto-approve when invoice matches PO. (Repeated) • Supplier spend queries: Quick query “what did we pay supplier X last month/period?” (Repeated) • Slack-first surfaces: Run processes in existing UIs (Slack) vs introducing new apps. (Single) • Quality & observability: Failure reporting with screenshots/video; human review over “self-healing.” (Repeated) E. Industry/Function Pointers (non-interpreted) • High activity/need: Banking, insurance, logistics/shipping, construction, airlines, public sector, hotels/hospitality, rail. (Repeated) • Approval flows matter: Vacation, onboarding/offboarding, finance approvals. (Repeated) • Specialized planning engines: Airline crew scheduling etc. hard to operate well; users struggle to use the system despite capability. (Single) F. Market/Cost Signals (reported by interviewees) • UiPath cost: Can reach ~€50k/yr for licenses in meaningful deployments. (Single) • External accountants: Spend ~€2k–€10k/month reported in examples. (Repeated) • WTP examples: AHEAD would pay €1–2k/month to free CFO time; investor reporting time seen as ~€10k/yr in internal salary value. (Repeated)

Process Inventory Process Pain / Current Pattern Tools/Context Mentioned Frequency Invoice matching & posting Manual match to transactions; line-item categorization; PO/contract comparisons; double-payment checks DATEV, external accountants, n8n/make, Excel Frequent Investor reporting (monthly) Manual incl. exports to Excel/Slides Excel/PowerPoint; data pulls from finance tools Repeated Budgeting/Forecasting Upload forecasts; compare actuals vs budget; want variance explanations Desire: “AGICAP + forecast upload”; Excel Repeated Approvals (finance & HR) Multi-step approvals (PO, invoices, vacation, onboarding/offboarding) Power Automate; custom tools Repeated Expense claims Ensure correct invoice uploads; categorization Excel/email; accounting tools Repeated Travel expenses Excel forms → assistant → HR → payout; very manual Excel; HR/payroll Repeated Supplier spend lookups Quick queries by supplier and period Desire for simple query tool Repeated Onboarding/Offboarding Seen as promising automation target Power Automate; HR systems Repeated SME invoicing (AR) Few good tools for issuing customer invoices Microsoft suite/manual Repeated KYC Heavy document collection/verification Compliance systems Single Reconciliation (GL) Full ledger reconciliation; validations both directions Accounting stack + contracts/POs Repeated Tendering / procurement Manual tracking; approvals; negotiation support ERP; spreadsheets Repeated

Constraints & Frictions Constraint Notes Frequency APIs incomplete/weak ERPs/finance tools lack comprehensive endpoints Frequent Auth & access 2FA/3FA, USB keys, VPN to internal apps Repeated Org fragmentation Local tools per country/entity; siloed decisions Frequent Automation fragility UI changes break bots; serial execution; testing pain Repeated Change mgmt & UX Users default to Excel; training gaps; fear of replacement Repeated IT governance Authorization blocks; policy hurdles; HIPAA/GDPR compliance for local tools Repeated Vendor lock-in risk Desire to avoid long-term lock-in (e.g., Palantir) Repeated Complexity creep Workflow graphs become unmanageable (e.g., 200 nodes) Repeated

Willingness-to-Pay / Economics Signal Example Frequency Monthly budget €1–2k/month to free CFO time (Jan) Single External accounting spend €2k–€10k/month cited across companies Repeated Time cost proxy ~~8h/month investor reporting (~~€10k/yr internal value) Single RPA license scale UiPath ~€550k/yr for larger deployments Single

Opportunities (Digital Assistants) Digital Assistant Human Equivalent Core Scope Why It Matters Order Assistant (Workist-type) Order-entry clerk / sales back office Reads incoming orders (PDF/email), validates, posts to ERP Front door of automation; high manual volume, proven demand Procurement Assistant Procurement manager Analyzes supplier data, identifies renegotiation and savings Turns spend analytics into actionable insights Purchasing Assistant Purchasing clerk Generates POs, tracks approvals, confirms deliveries Connects order intake with accounting; bridges silos Accounting Assistant AP/AR clerk Matches invoices to POs, reconciles ledger, ensures compliance, DATEV export Heaviest manual workload; easiest ROI story Finance Assistant Controller / FP&A analyst Builds budgets, analyzes actuals, creates investor reports Replaces recurring analysis and reporting tasks

Notes: • Each “assistant” mirrors a human role but operates across legacy tools, powered by a shared automation core (UI-as-API + LLM reasoning). It is not another SaaS solution. • Together, they form a Digital Workforce for Finance Operations.

Accounting Assistant - Concept Summary Purpose The Accounting Assistant is a virtual accounting employee that automates invoice processing, PO matching, and DATEV booking - directly inside Microsoft Teams, DATEV, SAP, and shared folders.  It doesn’t introduce any new UI. Users interact with it exactly as they would with a human colleague.

How It Works • Intake: Reads incoming invoices from an accounting inbox or shared folder. • Extraction: Uses OCR + LLM reasoning to extract key data (supplier, amount, tax, items). • Validation: Retrieves POs from SAP and supplier contracts from a folder; checks for mismatches. • Approval: Posts message in Teams:  “Invoice #842 from Müller GmbH matches PO #421. Approve for DATEV booking?” • Manager clicks Approve ✅ or Reject ❌. • Execution: After approval, books the entry in DATEV via UI automation (Legacy-Link Layer). • Audit: Posts confirmation and proof (video/screenshot) in #finance-automation-logs channel.  Key UX Principles • No new tool: All communication happens via Teams messages and channels. • Approvals in place: Native Teams buttons for quick approval/rejection. • Observability: Every action logged and auditable; summaries posted to a dedicated log channel. • Memory: Remembers past invoices, suppliers, and patterns for better categorization. • Error handling: Posts structured error messages in Teams for quick review and retry.  Automation Core (Shared Infrastructure) • Legacy-Link Layer: Automates browser and desktop UIs (DATEV, SAP) when APIs are missing. • Hybrid Logic: Deterministic steps for reliability, LLM reasoning for flexibility and anomaly checks. • Central Audit Log: All actions traceable with metadata and replayable proof. • Composable Design: Same infrastructure powers other assistants if added later. • Security Model: Each assistant has its own company account and access rights — onboarded/offboarded like a real employee.  Why It Works • No user training: it behaves like a colleague inside Teams. • Fully transparent: every booking, approval, and exception visible. • Compatible with IT and compliance: controlled, auditable, and safe. • Measurable ROI: replaces hours of manual booking and reconciliation work.

Challenges (to be extended) • MFA • Regular logins, password changes etc. • Usage of desktop software • Local vs. private cloud vs. our cloud • Getting customer IT on board