IDfy

Reading documents, matching faces and spotting manipulated media are model work, and a live Credit Underwriting Agent orchestrates sub agents over bureau, government and bank data to return a credit recommendation, with Auto-CAM spreading the financials, testing covenants and drafting the credit memo in two to four hours against three to four days by hand. Much of the platform is not model work, though: direct registry and bureau queries plus days of human investigation in background checks would remain without the models.

A reconsideration route for a rejected applicant, case management and the threshold at which a human must review are not described. The Credit Underwriting Agent applies the lender's own credit policy and hands a recommendation with its evidence trail to a human reviewer, who steps in where a decision needs judgment, and every agent call is logged and replayable. Video customer identification needs a trained officer in a live session, background checks rely on human investigators over hours or days, and customers set thresholds through no code workflows.

False acceptance and false rejection rates, an evaluation method and model documentation are unpublished for document reading, face matching and liveness. What IDfy does publish is behavior: which checks return in seconds and which take days, and a reason attached to every OneRisk score.

The quotes, from Kirti Patil, chief technology officer at Kotak Mahindra, and leaders at Yubi, Hero Enterprise and Blackbuck, speak to accuracy and fraud prevention without a measured outcome. IDfy's own figures carry the weight instead: more than 70 million verifications a month, over 15 million frauds prevented and 500 active clients in seven countries.

OneRisk's machine learning is trained on more than 400 million records, including criminal, court and fraud network data IDfy calls exclusive, and its fraud intelligence uses device fingerprinting and velocity signals across clients. The privacy policy sets no limit on that pooling and says nothing on training, so a bank cannot tell whether its own customers feed models that serve other clients. Deepfake and manipulated media detection are named capabilities.

Baldor Technologies Pvt. Ltd., the company behind IDfy, treats client data as processed on client instructions, but its privacy policy gives no retention period and does not address Aadhaar data. The stakes are high because national identity numbers, facial images, credit records and court histories make a wide personal data footprint under India's Digital Personal Data Protection Act.

A trust center, an audit period and a scope statement are not linked, and the auditor goes unnamed. IDfy states ISO 27001 certification and SOC 2 Type 2, has announced a completed Type 2 audit covering confidentiality and integrity, and quotes 99.9 percent uptime.

Prescriptive local rules give the product concrete grounding: Reserve Bank of India know your customer directions, the video identification procedure and Aadhaar authentication limited to entities the Unique Identification Authority of India approves. IDfy does not state which approvals it holds itself, though Privy is sold for data protection compliance. Its video identification keeps audit trails built for the Securities and Exchange Board of India, the Reserve Bank of India and the Insurance Regulatory and Development Authority of India.

IDfy publishes nothing on the accuracy or false match rate of its face matching or liveness checks, both of which vary by skin tone, age and capture quality, and nothing on fair lending, disparate impact or adverse action for the live Credit Underwriting Agent that recommends on small business loans. Court record matching carries a bias surface of its own, since common names, transliteration between scripts and patchy record keeping make false matches structural.

A false criminal record match can cost a person employment, credit and banking access at once, and IDfy publishes no accuracy guarantee or remediation commitment. Individuals can review, update and correct their data through a named grievance officer and data protection officer, which fixes what a future check reads, while a check already run stands.

Sources are named by category, from national identity authentication and tax registries to credit bureaus and court records, so the authorities behind a result are visible. Document reading, face matching and the AI agents have no named provider, and cloud and signing vendors appear only by category.

IDfy names no core banking or lending platform partner. Its integration work goes into identity, tax and company registries, each with its own access rules, reached through API Central and no code workflows, and its onboarding product states go live in one day.

Payment system data must be stored domestically under Reserve Bank of India rules, and the Aadhaar framework constrains where authentication data may sit. Service providers operate both in and outside India, per the privacy policy, with no hosting region or transfer mechanism stated for Southeast Asian and Middle Eastern work.

A self serve background verification platform sits beside the application programming interface (API) catalog, but no rates, tiers or billing unit are published. At tens of millions of checks a month, volume banding decides cost, and a buyer cannot model it from outside.

Non banking financial companies get their own treatment alongside banks, insurers and fintech lenders, a locally specific segment a generic platform would skip. Operations span India, Southeast Asia and the Middle East, with no European or North American coverage.