Perfios

Models do the work that matters but a very large platform survives without them. Machine learning reads bank statements, tax filings and financial statements and turns unstructured documents into income, cash flow and risk assessments, with the CAM AI underwriting platform layering generative and agentic tooling on top.

Strip all of it and 75 products, more than 500 application programming interfaces, the account aggregator plumbing and the integrations into national identity and tax infrastructure remain, which is a substantial data and connectivity business in its own right. This is the Codat and Alloy position, where models enrich a platform whose value is also its reach.

One published statement does real work here and it is unusually candid for a vendor selling automation. The company describes a hybrid model combining artificial intelligence with subject matter expert intervention in document scanning, which concedes that machine extraction alone is not sufficient on the input that everything downstream depends on, and places identifiable people inside the pipeline rather than only at the end of it.

The CAM AI framing is consistent, presenting the benefit as processing twice the applications with existing teams rather than as removing them. What is not published is where the boundary sits: no confidence threshold triggering expert review, no statement of what share of documents receive it, and no description of who signs off on an underwriting recommendation.

No accuracy figure, validation evidence, error analysis or model documentation was located, and the gap sits on the component everything else rests on. Document extraction is the foundation of the entire platform, so a misread figure in a bank statement or a tax filing propagates into an income assessment, a credit decision and a regulatory record, and no extraction accuracy rate is published for any document type.

The hybrid model with expert intervention is a real quality control and it is unmeasured, since nothing states what share of documents it touches or what error rate remains after it. Published outcomes are throughput, with an 85 percent reduction in turnaround time measuring speed rather than correctness.

Among the strongest evidence surfaces in this index on every measure the axis takes. More than 1,000 financial institutions across 18 countries, with named customers spanning the top of Indian banking including the largest private and public sector banks and major non bank lenders.

Scale is stated in verifiable units rather than adjectives: 8.2 billion data points delivered to institutions annually, 1.7 billion transactions processed a year, and 36 billion dollars of assets under management on the platform. The CAM AI launch carries a specific outcome claim of up to 85 percent reduction in underwriting turnaround and twice the application throughput with existing teams.

Unusually for this index the company is profitable with published revenue and profit growth, and has raised more than 400 million dollars across fifteen rounds from institutional private equity, which means its financial claims face investor scrutiny rather than only marketing review.

No data boundary statement was located. More than a thousand institutions run underwriting on one platform and many of them compete directly for the same borrowers in the same market, so whether document extraction, fraud patterns or risk models improve for one lender from another's applicant flow is the question a chief risk officer would put first.

Nothing published addresses it, and the acquisitions that brought identity, fraud and collections capability in house make the combined data surface larger rather than more legible. The benchmark answers to grade against remain Rulebase and DwellFi.

The privacy position rests partly on architecture the company did not build but operates inside, and that is a genuine strength rather than a technicality. A significant share of data access runs through the national account aggregator framework, a consent based sharing architecture supervised by the central bank, under which the customer authorises each flow and can withdraw it, so the data subject is a party to the arrangement rather than an unwitting one.

The company also names compliance with the national data protection statute and publishes a registration identifier as a licensed electronic system operator in a second jurisdiction. What holds this below the top grade is that consented flows are only part of the picture, since documents, bureau records and tax data also enter by other routes, and no retention schedule, subprocessor list or deletion commitment was located.

No attestation, certification, trust centre or enumerated framework was located. Third party review notes that consolidating multiple capabilities under one master agreement simplifies a customer's data protection vendor governance and security review cycles, which describes a procurement benefit rather than the vendor's own assurance position.

For a platform holding bank statements, tax filings and identity documents for the customers of more than a thousand supervised institutions, a published attestation set is the first document a security review requests and none is available to read.

The deepest regulatory integration in this index, and it is integration rather than assertion. The platform is wired directly into national financial infrastructure: the central bank's consent based account aggregator framework, the national identity authority for verification, the goods and services tax network and income tax return data as sources, and the small business development bank for enterprise lending support.

Each of those is a supervised system with its own admission requirements, and being inside them is a passed process rather than a claim. Compliance with central bank guidelines and the national data protection statute is stated, and the company publishes a business identification number as a registered electronic system operator with a named ministry in a second country, which is a formal registration a reader can check. Fourth A on this axis after OnFinance AI, Akur8 and NICE Actimize.

Credit assessment at national scale with no fairness disclosure of any kind. Models produce income verification, cash flow assessment and creditworthiness scoring that feed underwriting decisions at more than a thousand institutions, and the CAM AI platform is described as a credit underwriting platform rather than an input supplier, which puts it inside the decision rather than beside it.

Fraud and document tampering detection adds a second exposure with a harsher consequence, since a tampering flag is an accusation against a named applicant. Nothing published addresses fair lending testing, outcome analysis across borrower groups, adverse action handling, or how extraction accuracy varies across document types, languages, regional formats and the informal record keeping typical of the small business and agricultural borrowers the platform explicitly serves. Eighth instance of this pattern after Alloy, Sardine, Taktile, Oscilar, UPTIQ, EnFi and Lentra.

No guarantee, indemnity or falsifiable accuracy commitment was located. Genuine subject facing agency exists for part of the data flow, because access through the consent based aggregator framework is authorised by the customer and can be withdrawn, which is the property that earned Codat its higher grade. It does not extend to the rest.

An applicant whose income was misread from a document, or who was flagged for tampering by a detection model, is not told a machine produced that conclusion, cannot see the extraction, and has no described route to correct it, and a fraud flag in particular is an adverse finding that follows a person without their knowledge.

The data chain is named to an unusual degree because much of it is public infrastructure the company must identify to explain what it does: the consent based account aggregator framework, the national identity authority, the tax network, income tax return data and the small business development bank all appear explicitly, and a satellite intelligence partner is named for agricultural lending.

Several capabilities that would otherwise be third party dependencies were brought in house through acquisition, covering identity verification, fraud detection, collections and health claims, which shortens the chain and is disclosed. What remains unnamed is the model layer behind the generative and agentic tooling in the newer underwriting product, and no subprocessor list or hosting arrangement was located.

Integration is the company's defining property and it is evidenced at every level. More than 500 application programming interfaces across 75 products, direct connection into national identity, tax and consent infrastructure, credit bureau access, and stated integration timelines of two to four weeks.

The consolidation effect is the substantive claim: one contract replacing four or five vendor relationships covering income verification, identity, fraud and collections, which simplifies both procurement and the vendor governance burden a supervised institution carries.

Independent commentary describes the company as the backend layer beneath a national lending market, which is the position this grade is meant to capture, and partnerships such as satellite derived intelligence for agricultural lending show the surface still extending.

No hosting provider, region selection or residency commitment was located. The question is live rather than theoretical at this footprint, since the home market imposes data localisation on financial data, the company operates across 18 countries with expansion into two more regulatory regions underway, and it holds a local operating registration in at least one additional jurisdiction, which implies local establishment arrangements exist without describing them. Nothing published tells a buyer where applicant financial records rest or whether the location is configurable.

No pricing is published. Third party review describes an enterprise only sales motion with no self serve option, custom negotiated pricing and local currency billing, alongside integration timelines of two to four weeks and around eight weeks for account aggregator and open banking connections.

That gives a buyer the shape of the engagement and the effort involved, which is more than most enterprise vendors here concede, but it reaches the reader through an independent review rather than the company's own material and contains no rate or unit of charge.

The buyer set covers banks, non bank finance companies, fintechs, neobanks, co lending platforms, insurers and global lenders, and the lending journeys covered run from retail through small and medium enterprise to agricultural credit, with insurance claims added through acquisition. Geographic reach is 18 countries with an established position in India, presence across the Middle East and Southeast Asia, and stated expansion into North America and Europe.

The consolidation argument is a coverage claim in itself: a single contract replacing four or five separate vendor relationships across income verification, identity, fraud and collections, which is only possible because each of those is a maintained product rather than a checkbox.