Challenge
1 —
A detection model based on A complex business rule engine that is challenging to maintain and adjust
2 —
A sophisticated technical stack capable of handling large volumes of data in real-time
The Key Questions
1 —
Comment faire en sorte que mes équipes métiers puissent travailler main dans la main avec des Data Scientists et Data Engineers ?
2 —
How to integrate advanced AI models into an architecture based on real-time processing?
Approach
1 —
Acculturation of teams on AI topics and on agile functioning integrating different types of profiles (business, data, IT)
2 —
Retrieval of historical data constituting the training base of the model
3 —
Data analysis and co-construction with business teams of the major variables influencing the prediction
4 —
Training and fine-tuning of the model on historical data
5 —
Performance measurement and backtesting of the model on past periods
6 —
Packaging of the model and integration within the client's technical stack
Results
1 —
A 25% improvement in prediction accuracy at an equivalent recall compared to the rule engine
2 —
Construction of an AI model in about 15 weeks
3 —
A user-friendly monitoring tool to manage the detection model and track performance on D+1 (the day after)
