Compliance monitoring · Production

CNCC deploys AI-powered document research for France's statutory auditors via Dust

The problem

France's statutory auditors relied on keyword search alone to navigate thousands of pages of standards and regulations, but keyword search could not interpret complex professional questions—only match exact terms.

First attempt

The Elasticsearch-based Sidoni database, despite two major evolutions, still could not interpret auditors' professional questions—only match keywords.

Workflow diagram · grounded in source
1
Auditor submits natural question
trigger
“Auditors can ask natural questions rather than constructing keyword queries”
2
Semantic search retrieves doctrine
ai_action
“returns relevant doctrine, not just documents containing those words”
3
Reasoning transparency displayed
validation
“Sidoni shows its analytical steps, filtered to essential phases rather than overwhelming technical detail. Auditors can follow the logic and verify conclusions.”
4
Source-linked response delivered
output
“Every response links to authoritative documents. No unsourced claims, no AI-generated assertions without foundation.”
5
Hallucination prevention check
validation
“Agents are configured to acknowledge limitations rather than fabricate answers. When the knowledge base doesn't contain relevant information, the AI says so.”
6
Meeting minutes generated
output
“Drag-and-drop transcription processing generates structured summaries in minutes rather than hours.”
Reported outcome

The pilot achieved an 85-87% satisfaction rate among testers from a rigorous profession, with external pilot demand doubling from 20 to 40 auditors, and internal staff experiencing faster meeting minutes generation.

Reported metrics
Pilot satisfaction rate85-87%
External auditors in pilot40
Meeting minutes generation timeminutes rather than hours
daily Sidoni connections2,000
Reported stack
DustSidoniElasticsearchFrameMaker
Source
https://dust.tt/customers/cncc-ai-powered-research-statutory-auditors
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The pilot achieved an 85-87% satisfaction rate among testers from a rigorous profession, with external pilot demand doubling from 20 to 40 auditors, and internal staff experiencing faster meeting minutes generation.

What tools did this team use?

Dust, Sidoni, Elasticsearch, FrameMaker.

What results were reported?

Pilot satisfaction rate: 85-87%; External auditors in pilot: 40; Meeting minutes generation time: minutes rather than hours; daily Sidoni connections: 2,000 (source-reported, not independently verified).

What failed first in this deployment?

The Elasticsearch-based Sidoni database, despite two major evolutions, still could not interpret auditors' professional questions—only match keywords.

How is this compliance monitoring AI workflow structured?

Auditor submits natural question → Semantic search retrieves doctrine → Reasoning transparency displayed → Source-linked response delivered → Hallucination prevention check → Meeting minutes generated.

Related compliance monitoring cases