Databricks pitches Lakebase and Genie as a real-time fraud stack for telecoms
Databricks has outlined a telecom fraud detection setup that combines its Lakebase Postgres engine, Databricks Apps and the Genie AI coworker on one governed platform. According to the company, the aim is to close the gap between fraudsters who move at machine speed and detection systems that still run on batch jobs.
- Databricks combines Lakebase, Databricks Apps and Genie into one telecom fraud detection stack under Unity Catalog governance.
- Lakebase syncs lakehouse gold tables continuously, which Databricks says gives sub-millisecond feature retrieval for real-time scoring.
- Databricks cites $41.82 billion in telecom fraud losses, based on the CFCA's 2025 survey.
- The Telecom Fraud Prevention Solution Accelerator deploys the risk engine, workbench and AI investigation agent via Databricks Asset Bundles.
- The company says the approach augments existing fraud management systems and can be adapted to other industries.
The problem: fraud at machine speed, detection at batch speed
Databricks opens its case with loss figures. Citing the Communications Fraud Control Association's 2025 survey, it says telecom fraud losses have climbed to $41.82 billion. It adds that U.S. consumers reported losing more than $12.5 billion to fraud in 2024, a 25% jump in a single year, and that the global average cost of a data breach hit a record $4.88 million in 2024. According to the company, every dollar of fraud costs a carrier roughly 3.4x once chargebacks, replacement costs and investigation overhead are counted.
The company argues that a telecom account has become a gateway to a subscriber's other assets, because banking apps authenticate through the phone number, password resets flow through it and two-factor codes land on it. Once an account is compromised, Databricks says, a fraudster can execute a SIM swap, change the billing address, order expensive devices and move into linked financial accounts within minutes. It describes the root issue as "fraud moving at machine speed while detection moves at batch speed."
Why existing architectures fall short
Databricks says account takeover and SIM swap attacks were up 240% in 2024 in Australia, citing Stockhead, and that roughly 90% of them now happen with no contact with the victim. It also points to International Revenue Share Fraud, which it says pumps $4 to $7 billion a year into premium-rate numbers, and SIM box bypass fraud at about $3.1 billion a year. In most cases, the company says, the detecting signal already exists but is split across six or seven systems, including CDRs, CRM, billing, device and SIM data, network data and digital channel logs.
The company does not call for scrapping current fraud management systems, which it says successfully block known attack patterns. Its argument is that they should be augmented. It names cross-channel correlation, behavioral modeling, device emulation and fast-evolving patterns as the areas those engines were never built for.
A central concept is adversarial drift. Databricks says fraudsters change behavior as soon as they are caught, so a deployed model's recall can start degrading right away, sometimes within 24 hours. It describes the usual setup as a split between an operational OLTP system and an analytical OLAP system, linked by scheduled batch jobs that copy finished "gold" data back to the operational database. That copying, the company says, leaves the decision engine working from a stale snapshot, and an AI agent added on top inherits the same gap.
The proposed stack: Lakebase, Apps and Genie
Databricks says Lakebase, its managed Postgres engine, removes the OLTP/OLAP split. Gold tables in the lakehouse sync continuously into Lakebase, which the company says gives the real-time scoring engine sub-millisecond feature retrieval next to the analytical models that produced the features. With no batch job in between, it argues, a fix for a drifting model propagates at the speed of the drift rather than the next scheduled sync.
On the analyst side, Databricks says its agentic platform and Genie let teams analyze historical incidents, synthesize operational playbooks and automate complex investigative work. Genie acts as an AI coworker that offers natural-language querying over governed fraud data. Databricks Apps is positioned as a way to deploy custom operational tools, such as a case-management workbench, without waiting on a separate build cycle when a new fraud pattern appears.
Governance sits under Unity Catalog, which the company says manages governance, lineage and security from raw network ingestion to the live transaction decision. Databricks claims this removes the trade-off between giving an AI agent rich context and keeping sensitive data controlled.
What it means for carriers
Databricks is also offering a Telecom Fraud Prevention Solution Accelerator. According to the company, it deploys the full pipeline, risk engine, workbench and AI investigation agent into a customer's workspace through Databricks Asset Bundles, and it can be adapted to other industries.
For telecom operators, the pitch is a way to build on existing fraud systems instead of replacing them, while using real-time network telemetry as fraud signals and shortening the monitor-tune-redeploy loop. The claims about speed, including the shift from days to something much closer to real time, come from Databricks itself, and the source gives no customer results or benchmark figures to test them against.
