Early access · Fintech & banking

Detect fraud and AML risk across institutions, without pooling customer data.

Chefoba lets banks, fintechs, and payment providers train shared fraud and AML models on federated data. Each institution keeps its data in place. The result is governed, audited, and regulator-ready from day one, not a generic ML platform with federation bolted on.

Built on a federated learning system evaluated across two UK fintech platforms, four institutional clients, and 21,458 users.

How it works

One shared model. Data that never leaves your infrastructure.

Chefoba coordinates training across institutions without ever moving raw transaction or customer data between them.

01

Institutions train locally

Each participating institution trains on its own transaction data, inside its own infrastructure. Nothing raw ever leaves the building.

02

Contributions are trust-scored

Chefoba scores each participating organisation and adaptively weights how much its model updates influence the shared model.

03

Aggregation is governed

Model updates are combined under policy controls your compliance team sets, with no-code governance and a full audit trail.

04

Output is audited and evidenced

The resulting shared model ships with automated regulatory evidence, ready for your compliance review and your regulator.

Backed by evaluation, not a pitch deck

Evidence, including where it's uncomfortable

Chefoba's federation core was evaluated end to end on a real 30-day remittance prediction problem across two UK fintech platforms. We report what worked and what didn't.

0.869 vs 0.864

Federated model AUROC matched centralised training on the same problem. Decentralising the data added no measurable performance cost within a fixed model class.

0.797 TPR gap

Subgroup auditing by payment corridor surfaced a severe fairness gap that was invisible in aggregate accuracy metrics. Chefoba is built to find that gap before it reaches production, not after.

21,458 users

Evaluated across four institutional and geographic client partitions with 30 structured transaction features, including honest reporting of privacy-driven accuracy trade-offs.

Read the full methodology & results →

Also supports

Healthcare organisations collaborating on clinical models

The same governed federation core applies where institutions hold sensitive patient records instead of financial transactions. Chefoba is built primarily for fintech and banking, and supports healthcare deployments on the same platform.

Healthcare use cases →

Same core, second sector

Trust scoring, adaptive weighting, and automated evidence generation carry over unchanged. Only the deployment context differs.

Compare

How Chefoba compares to Palantir, Databricks, Snowflake, Flower Labs, and Owkin

Most platforms in this space are either broad data infrastructure with privacy bolted on, or open-source federation toolkits with no governance layer. Chefoba is purpose-built for governed, cross-institutional collaboration in regulated industries.

See the full comparison →

Governed federation vs. the alternatives

Trust scoring, adaptive weighting, no-code policy management, and automated regulatory evidence generation are proprietary to Chefoba among the vendors we track.

Be first to know when Chefoba opens access.

We're onboarding a limited group of fintech and banking partners before general availability.

Join the waitlist