evidoria

← Back to browse

Good practice Imported

Grand Débat National — France's AI-Assisted Analysis of 2 Million Citizen Contributions

France · Paris · See the France profile · See the Paris profile

Evidence: Descriptive / self-reported Top 36% 67/100 · Ask Evidence Copilot about this practice

France's 2019 Grand Débat National used a seven-firm consortium's NLP pipeline to classify and synthesise nearly 2 million online contributions and 16,000+ local registers from 1.5 million citizens into 19 public knowledge trees within about two weeks.

1.93 million contributions
Online contributions analyzed (January–April 2019)
16,000+ registers
Municipal registers digitized (2019)
~1.5 million citizens
Citizens represented in local registers (2019)
19 knowledge trees
Thematic knowledge trees produced (2019)
Grand Débat National — France's AI-Assisted Analysis of 2 Million Citizen Contributions

Details

Maturity
Discontinued
Promoter
French Government (Grand Débat National) with Roland Berger, Cognito, Bluenove, OpinionWay, Qwam, Res Publica and Missions Publiques
Period
Jan 2019 – Apr 2019
Keywords
citizen consultation, natural language processing, deliberative democracy

Context

Following the 'gilets jaunes' protests, France's government launched the 2019 Grand Débat National, inviting citizens nationwide to submit views on taxation, public services, democracy and the environment. The government contracted a seven-firm consortium led by Roland Berger to process the resulting responses using natural language processing.

Activities

The consortium digitized handwritten local registers (with Numen and the Bibliothèque Nationale de France) and analyzed nearly 1.93 million online contributions plus over 16,000 municipal registers representing around 1.5 million citizens, using a contextualised lexicological-analysis pipeline to classify text into predefined categories and produce 19 thematic knowledge trees, with human synthesis teams completing the final written analysis in roughly two weeks.

Results

Institut Montaigne's independent review found the pipeline successfully handled the scale of the exercise but showed technical limits: it struggled with irony, ambiguous language and near-duplicate phrasing, participation skewed toward older, more educated, home-owning men, and the system could flag but not fully distinguish coordinated campaigns from organic grassroots repetition.

Implementation
Indicative cost
Medium (€50k–€500k)
Time to results
Short (< 1 year)
Staffing & skills
Seven-firm consortium led by Roland Berger (Cognito, Bluenove, OpinionWay, Qwam, Res Publica, Missions Publiques), Numen and Bibliothèque Nationale de France for register digitization, Human synthesis teams for final written analysis

Conditions for success

  • Rapid multi-firm mobilization within a compressed timeframe
  • Predefined taxonomy/categories to structure classification
  • Human synthesis teams to complete final analysis rather than relying on NLP output alone

Common failure modes

  • NLP pipeline struggled with irony, ambiguous language and near-duplicate phrasing
  • Could not fully distinguish coordinated campaigns from organic grassroots repetition
  • Participation skewed toward older, more educated, home-owning men, limiting representativeness

Where it fits

Governance type
national government consultation
Scale
national
Income level
high-income

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful