Use case: AI driven
Fraud Detection.
Main areas of benefit.
Budget Savings ๐ฐ
We use open-source frameworks which enables us to onboard the latest AI trends quickly without a budget hit. Integration costs are also minimised.
Continuous improvement ๐
The system learns from your feedback and adapts its decisions accordingly => this enables you to detect new types of fraud quickly.
Customer satisfaction ๐
Our system can not only help with fraud detection, but also with better customer satisfaction.
Employee satisfaction ๐ฅฐ
Our system helps you with routine works so your employees can focus on creative tasks. 24/7 service 100% covered by machine.
Our principles ๐ฉ
Open Source
Zero cost for 3rd party software.
Cloud Ready
No matter if you run on AWS, Google Cloud, Azure or on-premise, we are ready.
Scalable
Prototyping can be done on local machine or small sandbox. Production can run on a huge cluster.
Modularisation
Enable only data sources you need. Detect only the types of fraud you want - a plug-in approach.
Performance ๐
Accuracy
More than 85%-95% precision depends on a case
0.004% - 0.006% false positives ratio
Latency & Performance
Less than 3 minutes to classify 25 million calls
Approximately 2 minutes to prepare data set
Financial Benefits (T-Mobile CZ)
1 FTE spared
24/7 monitoring not needed now
150k - 200k EUR yearly saved (OPEX)
Use cases ๐ฏ
Examples
PBX fraund - Hacked PBX, stolen mobile (SIM) identity, etc.
SIM boxing - Interconnection fraud
Spoofing - Detects fake call-centers, masquerading as a trusted entity
Wangiri - "one ring and cut" using premium numbers
Other features - payments fraud, flash-calls, gambling, etc.
Delivered
Fraud detection in Telco network
Fraud detection on Google payments data
Data analytics platform on huge amount of e-com data (scraped htmls mainly + behavior data)
Customer segmentation for marketing and campaign optimization
Setting up environment + processes for data analytics (hadoop + spark clusters, backend for reporting systems etc.)
Tech. stack ๐ง
How we work ๐จโ๐ป
Analysis/Feasibility
We analyse fraud cases specifics, data source available and decide on feasibility.
Proof of Concept
We prepare deployment of a real, representative scenario to evaluate performance and feasibility, this is cheap and fast.
Detailed design
We define how to integrate with the current systems/data sources and prepare infrastructure.
Integration
System is deployed, AI models trained and fraud detection is up and running.
Activation
System is activated and after certain monitoring period we ensure system is running OK considered in full duty.