General AI failed to parse 6,000 complex budget codes. A custom model now maps every line to green goals, cutting manual work by 30%.
A French municipality with a population that triples to 250,000 in summer manages 6,000 annual budgetary operations across diverse public services.
A new law mandated the classification of every budget line against environmental goals, but manual Excel processes were error-prone and unscalable....
“The ‘green’ budget is a perfect use case because it’s mandatory, time-consuming, and requires an understanding of French and the business context. General-purpose AI would produce poor results, as it doesn’t recognize budgetary codes or the acronyms we use. So, we had to train a specialized model—tiny LLMs—which works perfectly.”
Municipal government and public administration for the city of Antibes.
SAP is an enterprise software company that provides comprehensive business solutions for data management and digital transformation across various industries.
Related implementations across industries and use cases
Staff spent two days manually hunting for legal codes. Secure AI agents now cross-reference millions of documents in two minutes.
Manual mapping took a year and cost millions. One scientist now completes the work in three days using local AI, saving €1.2M.
Staff spent 15 minutes researching building codes. GenAI now queries the full corpus, cutting response times to under 3 minutes.
Aligning 70k laws to OECD standards meant manual checks across languages. AI translates and flags gaps, cutting work from years to <1.
Answering one regulatory question averaged two hours. AI now provides instant, cited answers in seconds.
Data policies blocked collaboration. Federated AI detects signals 10x faster, syncing global observatories without exposing raw data.
Drowning in 80,000 monthly alerts, analysts struggled to secure data. AI now prioritizes threats, cutting response time by 99%.
Manual pivot tables missed complex fraud rings. A graph database now links scattered evidence, cutting investigations from weeks to minutes.
100 manual touchpoints caused 12-day backlogs. AI now validates IDs and disburses funds autonomously, cutting processing to 30 minutes.