Internal Concept — Regnology × Orcha

Context

What Regnology Asked For

Regnology is a global regulatory technology company with a high-volume AP operation across multiple entities. They want Orcha as an intelligence layer alongside their existing Medius + D365 stack.

1. Quality Checks

Validate invoice fields on inflow, flag issues for AP review or return to vendor. Company-wide duplicate detection. VAT rate and amount correctness.

2. Fraud Detection

Flag changes in invoice field values — especially bank account changes — for further checking before payment.

3. New Vendor Creation

Automate the manual process of capturing master data for vendor creation across Medius and Dynamics 365.

→

4. Approval Assistance

Help approvers handle high-volume contractor invoices (rates, hours). Improved routing based on historic patterns and sign-off matrix.

Key Constraint

Orcha runs alongside Medius, not replacing it. Medius remains the system of record for AP workflow and approvals. Orcha adds an intelligence and quality layer. The AP team wants a dashboard to review Orcha’s findings.

Two Modules

Module	Scope	Goal
AP Intelligence	Quality Checks + Fraud Detection + New Vendor Creation	Measurable quality improvement within weeks. AP team gets a dashboard, Medius gets enriched data.
Intelligent Approval	+ Approval Assistance + Smart Routing + Contractual Checks	Extend from quality layer to decision-support layer for high-volume contractor invoices.

Module 1

AP Intelligence

Orcha processes invoices in parallel to Medius and surfaces results in a dedicated AP review dashboard.

Invoice Processing Flow

flowchart LR A["📧 Invoice\narrives"] --> B["AI Extraction\n& OCR"] B --> C["Quality\nChecks"] C --> D["Fraud\nDetection"] D --> E["Vendor\nLookup"] E --> F{"All checks\npass?"} F -- "✓ Clean" --> G["✅ Medius proceeds\nOrcha writes quality\nscore to Medius fields"] F -- "⚠ Issues" --> H["🟡 AP Dashboard\nAP team reviews,\nresolves, or returns"] F -- "🚨 Fraud" --> I["🔴 Teams alert\nImmediate notification,\nmanual investigation"] F -- "👤 Unknown vendor" --> J["🟣 Vendor creation queue\nExtract, verify,\ncreate in D365"] style B fill:#0078d4,color:#fff,stroke:#005a9e style C fill:#0078d4,color:#fff,stroke:#005a9e style D fill:#0078d4,color:#fff,stroke:#005a9e style E fill:#0078d4,color:#fff,stroke:#005a9e style G fill:#dcfce7,stroke:#16a34a style H fill:#fef3c7,stroke:#f59e0b style I fill:#fee2e2,stroke:#dc2626 style J fill:#f5f3ff,stroke:#8b5cf6

Capabilities

Quality Checks

All mandatory fields present (invoice number, date, amounts, tax, vendor details)
§14 UStG compliance for German invoices
Line item × quantity = net; net + VAT = gross
VAT rate correctness per item category
Company-wide cross-entity duplicate detection
Configurable auto-return emails citing specific issues

Fraud Detection

IBAN changed since last invoice → immediate flag
IBAN country mismatch vs. vendor’s registered country
Cross-referenced against D365 master data
Amount anomalies vs. historical average
Invoice splitting pattern detection
VAT ID validation (VIES) & sanction list screening (EU, OFAC)

New Vendor Creation Flow

flowchart LR A{"Unknown vendor\ndetected"} --> B["Extract master data\nName, address, IBAN,\nVAT ID, contact"] B --> C["Verify automatically\nVIES, register check,\nsanction screening"] C --> D["AP team reviews\nPre-filled form in\nOrcha dashboard"] D -- "Approve" --> E["Create in D365\nMedius picks up via\nD365 connector"] D -- "Reject" --> F["Notify with reason"] style B fill:#0078d4,color:#fff,stroke:#005a9e style C fill:#0078d4,color:#fff,stroke:#005a9e style D fill:#fef3c7,stroke:#f59e0b style E fill:#dcfce7,stroke:#16a34a

AP Review Dashboard

Flagged Invoices Queue

All invoices with quality or fraud flags
Severity indicator, filter by type/vendor/entity
One-click: approve override, return to vendor, escalate

Vendor Creation Queue

Pre-filled vendor forms with verification results inline
Approve → auto-creates in D365
Reject → notifies with reason

Medius Enrichment

Orcha writes results into Medius custom fields so the AP team sees Orcha’s assessment without switching tools.

Data	Medius Custom Field
Quality score (0–100)	Numeric field on invoice
Fraud flags (none / warning / critical)	Status field + hold if critical
Validation summary	Text field or comment
Duplicate indicator	Flag with link to original
Vendor verification status	Status field (verified / pending / flagged)

Principle

The AP team can work from the Orcha dashboard, from Medius, or both. Orcha enriches Medius but never blocks or changes its workflow.

Success Metrics

>95%

Quality catch rate

<5 min

New vendor creation
(vs. ~10 min today)

100%

Bank detail changes
flagged before payment

Duplicate invoices paid

Module 2

Intelligent Approval

Builds on AP Intelligence with decision support for approvers. Medius remains the approval system — Orcha provides the intelligence layer.

Why This Builds on AP Intelligence

AP Intelligence establishes the data foundation that Intelligent Approval needs: vendor profiles, historical invoice patterns, quality baselines, and validated D365/Medius integrations. The approval layer builds on top — it doesn’t start from scratch.

Regnology’s Approval Pain Points

From interviews (Kees Pruim) and the email exchange:

Current Challenges

Finding the right approver for every invoice takes significant effort
High volume of monthly contractor invoices in Professional Services and R&D
Manual checking of rates and hours worked

What They Want

Assistance for approvers — not full automation, not a Medius replacement
Improved routing based on historic patterns and sign-off matrix
Future: check against contractual terms

A. Contractor Invoice Validation

The highest-value use case: Professional Services and R&D teams receive monthly contractor invoices that each need manual checking.

flowchart LR A["Contractor\ninvoice"] --> B["Extract hours\n& rates"] B --> C["Pull contract\nfrom D365"] C --> D["Compare &\nscore"] D --> E{"Match?"} E -- "✓ Match" --> F["Recommend approve\nRate €120/hr matches contract.\n160 hours within range."] E -- "⚠ Mismatch" --> G["Flag with details\nRate €135/hr exceeds contract\n€120/hr (+12.5%)"] E -- "📦 Batchable" --> H["Bundle similar\n8 contractor invoices,\nall rate-matched. Approve all?"] style B fill:#0078d4,color:#fff,stroke:#005a9e style C fill:#f5f3ff,stroke:#8b5cf6 style D fill:#0078d4,color:#fff,stroke:#005a9e style F fill:#dcfce7,stroke:#16a34a style G fill:#fef3c7,stroke:#f59e0b style H fill:#eef2ff,stroke:#6366f1

B. Smart Routing Engine

A unified routing engine that decides — for every invoice that isn’t auto-approved — who sees it, when, how urgently, and whether it can be bundled.

⏱

How Urgently?

Skonto/early-payment: fast-track with “Save €X by approving before [date]”
Dunning letters: immediate escalation with urgency flag
Strategic suppliers: critical vendors get priority
Penalty clauses: flag approaching contractual deadlines
Period-end cutoffs: accelerate if it affects current month’s P&L

👤

Who Sees It?

Learned routing: analyse who actually approves what, rank by historical patterns
Parallel paths: run independent approvers in parallel, not sequentially
Auto-delegation: absent approver (M365 calendar/OOO) auto-delegates to backup
Revert on return: delegation automatically reverts when approver is back
Auto-escalation: unresponsive approver → escalate to next level

📦

Can It Be Bundled?

Low-risk, recurring invoices from the same vendor → group into batches
“12 invoices from Office Supplies GmbH, total €3,400, all PO-matched — approve all?”
Reduces approval fatigue for operational spend
Individual override still possible

C. Proactive Context Assembly

Before routing to an approver, Orcha assembles a complete decision package so they can decide without switching systems.

Financial Context

Current budget utilisation
Predicted month-end impact
Historical spend with vendor

Document Context

Related PO, GRN, contract
Prior invoices from same vendor
Open disputes or credits

Risk Context

Quality & fraud check results
Vendor trust score
Contract compliance status

D. Teams Approval Card (Mockup)

Invoice Approval — Acme Consulting GmbH

Invoice #2026-0847 · Due: 15 Apr 2026 · €19,200.00

Recommended: Approve

Rate Check

✓ €120/hr matches contract

Hours

✓ 160 hrs (avg: 155 hrs)

Budget Impact

78% → 83% utilised

Last 6 invoices: €18,600 · €19,200 · €18,000 · €19,200 · €17,400 · €19,200 — consistent pattern.

Key Principle

Approvers act in Teams. Medius gets the result. The approver never needs to open Medius or the Orcha dashboard for routine decisions.

Technical

Integration & Build Assessment

How Orcha connects to Regnology’s systems without disrupting them, and what we still need to build.

Integration Map

flowchart TD EMAIL["📧 Email Inbox\nExchange / Microsoft 365\nOAuth — same source feeds both systems"] EMAIL --> ORCHA & MEDIUS subgraph ORCHA_SYS ["Orcha"] ORCHA["AI Pipeline\nAP Intelligence: extraction, validation, fraud, vendor\nIntelligent Approval: + decision support, routing"] DASH["AP Dashboard"] VENDOR["Vendor Queue"] FRAUD["Fraud Monitor"] end subgraph MEDIUS_SYS ["Medius (unchanged)"] MEDIUS["AP Workflow &\nApproval System of Record"] end ORCHA -- "flags &\ncustom fields" --> MEDIUS ORCHA --> D365["Dynamics 365\nRead: vendor master, contracts, rates, POs, budgets\nWrite: new vendors, approval status"] ORCHA --> TEAMS["Microsoft Teams\nAP Intelligence: fraud alerts, vendor notifications\nIntelligent Approval: + approval cards, escalation"] style ORCHA_SYS fill:#eff6ff,stroke:#0078d4 style MEDIUS_SYS fill:#f5f3ff,stroke:#8b5cf6 style D365 fill:#f0fdf4,stroke:#16a34a style TEAMS fill:#eef2ff,stroke:#6366f1

Integration Touchpoints

System	Direction	Data	Module
Email (M365)	← Read	Invoice emails + attachments via OAuth/Graph API	AP Intelligence
Medius	→ Write	Quality scores, fraud flags, validation summary, duplicate indicator, vendor status	AP Intelligence
D365	← Read	Vendor master data (AP Intelligence); contract terms, rates, POs, budgets (Intelligent Approval)	AP Intelligence Intelligent Approval
D365	→ Write	New vendor creation (AP Intelligence); approval status updates (Intelligent Approval)	AP Intelligence Intelligent Approval
Teams	↔ Both	Notifications (AP Intelligence); interactive approval cards with action callbacks (Intelligent Approval)	AP Intelligence Intelligent Approval

Technical Requirements from Regnology

For AP Intelligence

OAuth consent for email inbox (M365)
Medius API credentials (custom fields write)
D365 API access (vendor master read/write)
Teams webhook or app registration
Sample invoices for initial model training

For Intelligent Approval

D365 Project module access (contractor rates, hours)
D365 contract and budget data access
Existing approval matrix export
Historical approval data for routing model
Teams app registration (bot framework)

AP Intelligence — Build Status

Capability	Status	Notes
AI extraction & OCR	✅ Built	Production-ready. Claude Sonnet + Google Document AI.
VAT & arithmetic validation	✅ Built	VAT rate checking, §14 UStG, arithmetic validation in production.
Duplicate detection	✅ Built	In production. Needs extension for cross-entity detection.
Fraud detection (IBAN, anomalies)	✅ Built	Bank detail changes, anomaly detection in production.
Supplier verification (VIES)	✅ Built	VIES validation, register checks, risk scoring operational.
Supplier return communication	✅ Built	Auto-return emails with configurable templates.
AP review dashboard	🔨 Adapt	Admin dashboard exists. Needs Regnology-specific AP review views.
Medius integration	🔧 Build	New. Medius API for custom field writes. Medium effort.
D365 vendor creation	🔧 Build	New. D365 API for vendor master data. Standard REST API.
Teams notifications	🔨 Adapt	Slack integration exists. Teams equivalent needs building.

Intelligent Approval — Build Status

Capability	Status	Notes
Contractor rate/hours validation	🔧 Build	New. D365 project module integration + comparison logic.
Smart routing suggestions	🔧 Build	New. Historical approval data import, pattern learning, matrix integration. Co-design piece.
Proactive context assembly	🔨 Adapt	Partial — vendor history exists. Budget and contract data need D365 integration.
Teams interactive cards	🔧 Build	New. Bot framework + adaptive cards + action webhook handling.
Escalation & delegation	🔧 Build	New. M365 calendar integration for OOO, configurable escalation rules.
Contractual terms checking	🔨 Adapt	Contract management module exists. Needs extension for rate/SLA comparison.

Summary

AP Intelligence is ~70% built. Core AI extraction, validation, and fraud detection are production-ready. Main build items: Medius and D365 integrations, dashboard adaptation. Intelligent Approval requires more new development — routing intelligence and Teams interactive cards are where the co-design partnership adds the most value.

Internal Discussion

Open Decisions

Questions that affect scope, pricing, and positioning. The meta question: do we push for deeper integration (justifiable setup fees, stronger lock-in) or keep it lean (faster to ship, lower barrier)?

Decision 1: New Vendor Creation — How Deep?

Option	What It Means	Effort
A. Dashboard only	Pre-filled form, AP copies data into D365 manually. Value = accuracy, not speed.	Low
B. Export template	Orcha generates D365-compatible import file. AP uploads via standard import.	Medium
C. Direct API write	Approve in Orcha → vendor created in D365 via API → Medius picks up. This is what they asked for.	High

Decision 2: Orcha’s Position in the Invoice Flow

Option	What It Means	Effort
A. Fully parallel	Both systems read from inbox independently. Orcha is purely additive. If Orcha has an issue, Medius keeps working.	Low
B. In between	Orcha sits between inbox and Medius as a quality gate. Can block bad invoices. But becomes critical path.	High

Decision 3: Write Back to Medius?

Option	What It Means	Effort
A. Dashboard only	Orcha results only in Orcha. AP checks two screens. No Medius API needed.	None
B. Custom fields	Orcha writes scores/flags into Medius. AP sees everything in one place. Medius API integration required.	Medium

Decision 4: Approval Scope

Option	What It Means	Effort
A. Advisory only	Orcha provides recommendations and context (Teams/dashboard). Approve/reject stays in Medius. Matches what they asked for.	Medium
B. Teams-native + sync	Approvers act via Teams cards. Orcha writes decision back to Medius. Richer experience, two systems involved.	High

Decision 5: Contractor Validation — Data Access

We Need to Understand

Where are contracted rates stored? (D365 Project? Separate system? Spreadsheets?)
Are timesheets/hours tracked in D365 or externally?
What does “checking rates and hours” look like today?

Impact

If data is in D365: standard API, medium effort
If in spreadsheets: we need a reference data ingestion path
Determines whether this is part of Module 2 or a later phase

Additional Questions

Topic	Question
Multi-entity scope	How many entities in Module 1 vs. full rollout? Cross-entity duplicate detection from day 1?
D365 version	Business Central or Finance & Operations? Different APIs.
Data & privacy	EU data residency requirements? DPA/sub-processor standards? LLM data processing consent?
Vendor return coordination	If Orcha returns an invoice to a vendor, what happens to it in Medius? Need a clear handoff protocol.

Next Steps

Internal

Alignment on Decisions 1–5
Draft pricing model based on integration depth
Prepare approval demo/mockup for follow-up meeting

With Regnology

Clarify contractor data location
Confirm Medius API capabilities and D365 version
Scope initial rollout to specific entity/entities

Intelligent AP Processing for Regnology