.png)
DGFiP
Government · France
20,000+ undeclared pools detected
French tax authority uses computer vision to find 20,000+ undeclared swimming pools
AI, aerial imagery and register matching generated €10M additional revenue in the first year.
Friction
Detecting undeclared value-adding additions (pools, outbuildings) was impossible at scale without unaffordable inspection capacity.
Breakthrough
AI and register matching as a compliance engine: aerial image detection cross-referenced with existing registrations to surface discrepancies for targeted enforcement.
Impact
9-region test revealed 20,000+ undeclared pools. Approximately €10M additional revenue in 2022. Approximately €40M projected for 2023 national rollout.
Unlock the full analysis with breakthrough, impact, what made it smart and its technical approach below!
Problem
Property tax revenues depend on registered objects and modifications. DGFiP needed a scalable way to detect undeclared value-adding additions without unaffordable inspection capacity.
What made it smart
The detect-match-triage-human decision pattern: AI flags the discrepancy, humans confirm and issue the assessment. Neither pure automation nor pure manual work, but intelligent triage at scale.
Technical approach
Computer vision model on aerial images detects pool shapes. Automatic cross-check against land and tax registers. Results go to human handling for validation and assessment. Core principle: detect, match, triage, human decision.
Strategic lesson
Computer vision creates public value when it turns unscalable manual inspection into automated signal generation, with humans making the final call.
Reflection question
Where in your organisation do you know non-compliance or inefficiency exists, but lack the capacity to detect it at scale?