Loading…
Loading…
Most AI invoice automation is a black box — the model decides, you can't explain why. Archiet generates a governed invoice approval agent: the approval flow is BPMN (auditable), the decision logic is a DMN policy table your finance team can edit without code changes, and the AI only reads documents — it never decides the outcome.
Invoice approval has three properties that make it perfect for governed AI: the inputs are structured (amount, PO number, vendor, department), the decision logic is explicit (threshold-based, policy-governed), and the outcome must be auditable (finance auditors and AP clerks need to explain every decision). Standard AI agents fail this test — LLMs make probabilistic decisions that cannot be reliably explained or audited. Governed AI agents pass it: the BPMN process defines the flow, the DMN policy table defines the routing logic, and the LLM only extracts structured data from the invoice document.
The generated invoice approval agent has three layers. Layer 1 — AI data extraction: the LLM reads the invoice document (PDF or image) and extracts structured fields: invoice amount, vendor name, PO reference, invoice date, line items, and VAT. The LLM does not make any approval decision. Layer 2 — DMN policy evaluation: a deterministic rule engine evaluates the extracted fields against your policy table. Example rules: if amount < £500 and PO matched, auto-approve; if amount > £10,000, escalate to CFO; if vendor is not on approved list, hold for review. Layer 3 — BPMN routing: the decision output routes the invoice to the correct recipient (auto-approve queue, AP manager inbox, CFO escalation) via the BPMN process.
The DMN (Decision Model and Notation) policy table is a JSON file that your finance team can edit without touching code. It contains rows like: {amount_lt: 500, po_matched: true, department: 'any', result: 'auto_approve'}. Business analysts use the included operator console to add, edit, and remove rules. Changes take effect immediately — no code deployment required. Every rule change is logged with the editor's name and timestamp. This is the governance layer: the AI cannot approve an invoice that the policy table rejects, and the policy table cannot be changed without an audit record.
The generated application includes: a Python/FastAPI invoice approval agent runtime with the three-layer architecture, BPMN 2.0 process XML (opens in bpmn.io or Camunda Modeler), DMN policy table in JSON format with sample rules matching your thresholds, a React/Next.js operator console where AP teams submit invoices and finance managers edit policy rules, a full audit log model recording every decision with the matched rule, EU AI Act Article 9 risk management documentation, and a Postman collection for testing the approval API. The agent is wired into the rest of your generated application — not a standalone service that needs manual integration.
Under the EU AI Act, AI systems used in financial decisions may be classified as high-risk (Annex III). Governed agents are designed to satisfy the high-risk requirements: Article 9 (risk management system) — the BPMN process and DMN policy constitute the documented risk management framework. Article 13 (transparency) — every decision includes the matched policy rule, not an LLM confidence score. Article 14 (human oversight) — escalation paths and override capabilities are built into the BPMN process. The generated AI governance pack includes the risk classification assessment, model card, and data-lineage diagram required under Annex IV.
Rossum, Hypatos, and Basware use machine learning models to classify and route invoices — but the model's decision cannot be fully explained or audited. Archiet generates a governed agent where the routing decision is made by a DMN policy table your team owns, not an ML model. The AI (LLM) only extracts fields from the document; the policy evaluates them deterministically. Every decision is explainable: 'Amount £8,500 > threshold £5,000 → escalated to Finance Director per Rule 4.'
Yes. The DMN policy table is a JSON file editable via the operator console. Finance managers can add rules, adjust thresholds, and change escalation paths without touching code. Changes take effect on the next invoice submitted. Every change is logged with the editor, timestamp, and previous/new rule state.
PDF invoices (digital and scanned), PNG/JPEG invoice images, structured XML (e.g., UBL 2.1 or EN 16931 e-invoices), and plain text. The extraction model returns structured JSON regardless of input format. Confidence scores are included — low-confidence extractions are flagged for human review before policy evaluation.
Yes. The DMN policy table can include a po_matched and gr_matched flag. The extraction layer checks the invoice's PO reference against your PO database API and sets these flags before policy evaluation. The generated integration layer includes a configurable PO lookup endpoint that your ERP or procurement system populates.
Under one day for a standard deployment. The generated ZIP includes the full application stack — run docker compose up and the agent is running locally in 15 minutes. Connect your invoice ingestion source (email inbox, ERP webhook, or manual upload), set the policy table thresholds, and the agent is processing invoices. Production deployment to Azure, AWS, or your private cloud takes one additional day for infrastructure provisioning.
Enterprise procurement teams need approval automation that can be explained to auditors, edited by business analysts, and trusted in regulated industries. Archiet generates a governed procurement agent: BPMN workflow, DMN policy table, bounded LLM for document reading, and a complete audit trail from day one.
Insurance claims triage requires speed, accuracy, and full auditability. Archiet generates a governed claims triage agent: FNOL intake, AI-powered data extraction from claim documents, deterministic DMN severity routing that actuaries control, and a complete audit trail for regulatory examination.
Loan approval decisions need to be explainable to regulators and customers. Archiet generates a governed loan adjudication agent: the BPMN workflow defines the process, the DMN credit policy table defines the approval criteria, and the AI only reads documents — it never decides the outcome.
free plan. No credit card required. Generate your first compliant architecture blueprint in under 10 minutes.