Perfios

Machine learning turns bank statements, tax filings and payslips into income and risk assessments, and the CAM underwriting platform adds generative and agentic tooling. Strip the models and a data and connectivity business of more than 75 products still stands.

Subject matter experts intervene in document scanning alongside AI, placing identifiable people inside the pipeline, and CAM is pitched as handling more applications with existing teams. Nothing says what confidence threshold triggers expert review, what share of documents gets it or who signs off a recommendation.

A misread figure in a bank statement propagates into an income assessment, a credit decision and a regulatory record, yet no extraction accuracy figure, error analysis or model documentation is published. Up to 85 percent faster underwriting with CAM measures speed, not correctness.

Customers named include ICICI Bank, HDFC Bank and Standard Chartered, with executives quoted at Security Bank in the Philippines and Karur Vysya Bank, across more than 1,000 institutions and 2 billion transactions a year. References and numbers never join on the homepage.

Competing lenders run underwriting on one platform, so whether one lender's applicant flow improves extraction or fraud models for another is the first question a chief risk officer would ask. The trust center covers hosting and backups but says nothing about model training or cross customer data use.

Much of the data arrives through India's Account Aggregator framework, where the customer authorizes each flow and can withdraw it. ISO 27701 privacy certification and rolling 30 day backups add structure, though documents and bureau records also enter by other routes and no subprocessors are listed.

ISO 27001 and ISO 27017 certificates dated 2025 and the Cloud Security Alliance's STAR Level 2 can be downloaded without asking. A SOC 2 reference appears without its type, the one item a buyer would still request.

Indonesia's Ministry of Communication and Information lists Perfios as a registered electronic system operator, with a business identification number a reader can check. It also participates in supervised Indian data infrastructure, each part with its own onboarding requirements.

CAM sits inside the credit decision rather than feeding it, and a document tampering flag is an accusation against a named applicant. Fair lending testing, outcomes across borrower groups and how extraction holds up on regional languages and informal farm and small business records are all unaddressed.

Consent over data access is not a way to challenge an outcome, and an applicant whose income was misread is not told which system decided. Perfios publishes no guarantee or remedy and describes no audit trail tracing an extracted figure to its source document, even for the lender.

Public infrastructure supplies much of the data, including identity verification, the goods and services tax network and income tax returns. Acquisitions brought identity, fraud and collections capability in house, while the model behind CAM's generative tooling is unnamed.

More than 500 APIs, plus Journey Builder, Consent Management and Perfios Hub for assembling flows, make integration the defining property. One contract can cover income verification, identity and collections, which simplifies vendor governance for a supervised lender.

Data can be stored and processed within India, the European Union or other regions as mandates require, on AWS, Azure and Google Cloud. Which cloud serves which customer and whether an on premises option exists are not described.

Pricing for a catalog spanning analysis, verification and fraud suites sits with the sales team. Whether charges fall per document, per verification or per API call is unstated, so cost cannot be sized before contact.

Banks, non bank lenders such as InCred and UGRO Capital, and insurers such as Prudential and Manulife are all served, with lending from retail through small business to agricultural credit. Eighteen locations reach from India across the Middle East to Ireland.