koncile: AI-Powered OCR and Procurement Analytics

7 min read

On This Page

Koncile is a Paris-based intelligent document processing (IDP) platform combining optical character recognition (OCR) with large language models (LLMs) for general document extraction and procurement spend analytics.

Overview

Founded in 2022 by Jules R., Tristan Thommen, and Raphael Balbous, koncile targets a structural gap in procurement: less than 20% of data in procurement documents is currently exploited, leaving contracts, purchase orders, and supplier invoices as largely untapped sources of spend intelligence. The company's stated ambition is to make 100% of that information actionable.

Koncile operates two distinct products. Koncile Extract handles general document processing at €49/month for 200 pages, targeting teams that need accurate extraction from complex or degraded inputs. Koncile Analytics addresses procurement spend management at $199/month, layering supplier analysis and cost optimization on top of the extraction pipeline. This dual-product structure reflects a deliberate move from horizontal OCR into a defensible vertical niche before broader go-to-market investment.

In January 2026, koncile was named Startup of the Year at the 10th ADRA'Up awards by ADRA, France's leading procurement network, whose members include LVMH and TotalEnergies. Each startup can participate only once, making this a one-time milestone. The award validates the procurement analytics thesis with an audience that evaluates spend management tools professionally.

In March 2026, the company launched an MCP (Model Context Protocol) OCR server that lets AI agents including Claude and Cursor extract documents without custom code, cutting integration time from 3 to 5 days down to approximately 15 minutes. This positions koncile as infrastructure for agentic document workflows, not just a standalone extraction tool.

The company received early backing from Agoranov and Entrepreneurs First. It competes directly against Amazon Textract, Mindee, Nanonets, and Klippa.

How koncile processes documents

Koncile's architecture follows a three-stage pipeline. Pre-processing corrects skew, normalizes resolution, and prepares scanned or photographed documents for recognition. Advanced OCR then converts the prepared image to text, handling printed content, handwriting, tables, and non-standard layouts. Finally, LLM analysis interprets the extracted text in context, applying domain understanding to structure output and assign confidence scores to each field.

This hybrid approach addresses the core limitation of traditional OCR: text recognition without comprehension. A standard OCR engine returns raw strings. Koncile's LLM layer understands that a number following a currency symbol on an invoice line is a unit price, not a reference code. The result is structured, validated output rather than a text dump requiring downstream parsing.

The platform publishes specific accuracy benchmarks. According to koncile's technical documentation: "For simple, unique field recognition, such as the total amount on an invoice, a seller's name, or an account holder's name, a 99% success rate is achievable." For complex fields, the same source states: "For invoice line items with numerous columns, 95-96% can be reached provided the Invoice OCR engine can parse and structure each line with precision." These are vendor-reported figures without independent verification.

The hybrid model involves a speed trade-off. Traditional machine learning OCR processes documents in 1 to 4 seconds. LLM-based vision OCR takes 5 to 10 seconds per document. Koncile applies both depending on document complexity, using field-level confidence scoring to route extractions requiring human review. Handwriting recognition (HTR) uses LLM-based and deep learning models within the same pipeline, enabling processing of handwritten medical prescriptions and mixed-format forms without a separate workflow.

Output formats include JSON, XML, and CSV. Native integrations cover Google Drive, Slack, and ERP systems. The REST API and SDK support custom integration paths. Deployment options now include both cloud SaaS and on-premises hosting for organizations with strict data residency requirements.

MCP OCR server and agentic workflows

The March 2026 MCP OCR server launch represents koncile's most significant technical direction signal to date. The server exposes 24 tools covering the full document lifecycle: file upload, task status polling, structured data retrieval, template management, and folder operations. AI agents connect to the server and extract documents via natural language instructions, without JSON knowledge or API configuration.

Koncile offers both a hosted deployment at mcp.koncile.ai and self-hosted options via pip install or Docker for teams with data residency compliance requirements. The announced roadmap includes batch processing for large volumes, real-time webhook notifications, and accounting software integration via chained MCP tool calls.

The practical implication is that finance and procurement teams using AI assistants can trigger document extraction as part of a broader workflow, rather than routing documents through a separate extraction tool. This reduces the technical barrier for non-developer users and positions koncile as a component in agentic finance automation rather than a point solution.

Use cases

Invoice and financial document processing

Accounting teams use Koncile Extract to automate invoice processing with multi-currency support and structured output for direct accounting system integration. The €49/month entry tier targets small teams and proof-of-concept deployments. The platform's competitive comparison positions it as the only tool offering a no-code interface combined with a complete API and structured multi-format export, though this claim is self-reported and not independently verified.

Procurement analytics and spend management

Koncile Analytics extracts and analyzes procurement data from contracts, purchase orders, and supplier invoices for cost optimization and vendor management. The ADRA'Up award centers on this use case: converting procurement documents into actionable spend data to detect savings opportunities and monitor supplier performance at scale. With ADRA members including major French enterprises, the award signals credibility with the procurement community koncile targets.

Healthcare document management

Healthcare organizations process medical forms and handwritten prescriptions using the platform's HTR pipeline while maintaining GDPR compliance for sensitive patient data. Confidence scoring provides per-field reliability metrics, supporting human review workflows where regulatory requirements demand auditability.

Complex and multi-domain document processing

Organizations encountering accuracy failures with traditional OCR tools use koncile for challenging inputs: scanned forms with degraded quality, mixed-format files combining printed and handwritten content, and documents with non-standard layouts. The LLM layer provides contextual correction that rule-based post-processing cannot replicate. Koncile targets logistics, transport paperwork, bank statement extraction, finance, HR, and health sectors as named verticals.

Technical specifications

Feature Specification
Core technology Hybrid OCR plus LLMs (3-step pipeline)
Product lines Extract (general), Analytics (procurement)
Recognition capabilities Printed text, handwriting (HTR), tables, complex layouts
Supported formats PDF, images, scanned documents
Processing speed 1-4 seconds (traditional OCR), 5-10 seconds (LLM vision)
Accuracy benchmarks 99% simple fields, 95-96% complex line items (vendor-reported)
Language support Multilingual, multiple currencies
Integration REST API, SDK, MCP OCR server (24 tools)
Deployment Cloud SaaS, on-premises
Security certifications SOC 2 (vendor-disclosed), GDPR, CCPA, ISO 27001
Data policy No training on customer data
Output formats JSON, XML, CSV
Native integrations Google Drive, Slack, ERP systems
Pricing Extract: €49/month (200 pages), Analytics: $199/month
Free trial 20 free credits for new users
99%Simple field accuracy (vendor-reported)
95–96%Complex line item accuracy
15 minMCP server setup vs. 3–5 days for API
20%Procurement data currently exploited (market gap)

Competitive position

Koncile's published comparison targets Nanonets, Klippa, and Readiris on multi-domain document handling and structured data export. The self-reported claim of being "the only one" to combine a no-code interface, complete API, and structured multi-format export is a positioning statement, not an independently verified benchmark. Klippa is Netherlands-based with strong logistics coverage; Nanonets is India-based with strength in forms processing. Koncile's differentiation argument is breadth across document types rather than depth in a single vertical.

Against Amazon Textract and Mindee, koncile's LLM layer is the primary differentiator. Textract excels at structured forms and tables at scale but applies rule-based extraction without semantic understanding. Mindee targets developer-first API users with pre-built models for specific document types. Koncile positions between these: more semantic understanding than Textract, more document-type flexibility than Mindee's pre-built model approach.

The MCP server launch opens a different competitive dimension. By enabling AI agents to call document extraction as a tool, koncile competes with any IDP vendor that has not yet exposed an MCP interface, regardless of underlying OCR quality.

All accuracy figures (99% simple fields, 95-96% complex line items) are vendor-reported from koncile's own documentation. No independent benchmark has verified these numbers as of April 2026.

Resources

Company information

Headquarters: Paris, France

Founded: 2022

Founders: Jules R., Tristan Thommen, Raphael Balbous (Technical Lead)

Investors: Agoranov, Entrepreneurs First