FinBox

The suite divides roughly in half. Models carry bank statement parsing and interpretation, the alternative data score built from mobile device signals, and the risk score that feeds decisioning, and none of those functions exists without them.

Alongside sits substantial infrastructure that does: a loan origination system, partnership and co lending rails, embedded distribution, identity interfaces and connections into national lending infrastructure, all of which a lender would still value with the models removed. The decisioning component is itself described as a business rules engine with artificial intelligence layered on, which is the honest shape. This is the Perfios and Lentra position, models doing the differentiating work inside a platform whose reach is also its value.

The product is built for speed and the published framing is explicit about it, promising instant underwriting, real time decisioning and credit embedded into a partner platform within days. Genuine control sits at design time, since the decisioning engine lets a lender author its own policies and, according to independent description, run controlled tests between competing underwriting policies before adopting one, which is meaningful ownership of the rules.

What is absent is anything at decision time: no referral threshold, no review queue for marginal applications, no confidence exposure on a score, and no described path for a borrower the models cannot assess confidently. Agentic workflows are funded and in development, which will sharpen this question rather than answer it.

No accuracy figure, discrimination statistic, validation result or model documentation was located, and improved approval accuracy is asserted without measurement. One real capability exists in the decisioning layer, since independent description indicates lenders can test underwriting policies against one another before adopting a change, which is the right mechanism for policy risk.

It does not extend to the models themselves: nothing states how the statement parser performs across bank formats, how the device score behaves across handset types and operating system versions, or how any of it holds up as borrower behaviour shifts.

More than 130 enterprise clients with the largest names in Indian lending stated publicly, including two of the country's biggest private sector banks and three major non bank lenders and diversified financial groups. Volume is given in the unit that matters for this function, with more than nine billion dollars of loan applications facilitated through the platform, alongside 100 percent year on year growth across core products.

A 40 million dollar Series B in 2025 was led by a major growth investor with participation from an established venture firm and a large industrial group's venture arm, of which 35 million was primary capital. Named customers at that tier are the strongest evidence available in this category, because each represents a completed procurement at an institution with a supervisory relationship of its own.

No data boundary statement was located. The question carries real weight at this footprint, because more than 130 lenders compete for overlapping borrower populations in one market and alternative data scoring improves markedly with pooled repayment outcomes, so the commercial incentive to learn across the customer base is strong.

Nothing states whether device signals, parsed statements or performance data from one lender inform models serving another, whether a lender can decline to contribute, or how the shared risk score is trained.

No retention schedule, subprocessor list or deletion commitment was located, and one product makes this the sharpest privacy position in the index. Device based alternative data underwriting draws on signals from a borrower's mobile phone, which depending on implementation can encompass installed applications, message metadata, contact patterns, device characteristics and location history, and that is a materially more intrusive payload than bank statements or bureau records.

The company's published privacy language addresses its clients' data integrity rather than the borrower's. Independent description of the pricing model references a consent flow for the device product, which implies a consent architecture exists without describing what is disclosed or how it can be withdrawn.

A security and compliance section is maintained and describes advanced encryption and real time monitoring, which is more attention than many vendors here give the subject, but no attestation, certification or enumerated framework was located within it.

For a platform holding parsed bank statements, identity documents and mobile device signals on behalf of more than 130 lenders including the largest banks in its market, an independently assessed control set is what those institutions' own examiners will expect to see, and it is not public.

Grounded in named national infrastructure, which is the pattern across every Indian vendor in this index. The platform connects to the central bank's unified lending interface, to the consent based account aggregator framework and to the open network for digital commerce, each of which is supervised public infrastructure with its own admission requirements, so participation is a technical integration a vendor must qualify for rather than a claim it can assert.

Identity verification interfaces sit alongside. What holds this below the top grade is that no regulator, statute or guidance instrument is named directly and no licence or recognition is evidenced, so the position rests on the rails rather than on the rules governing what runs across them.

The most invasive credit input recorded in this index paired with an explicit and unevidenced fairness claim. Device based underwriting infers creditworthiness from what a borrower's phone reveals, and those signals correlate directly with income through handset value, with religion, health and political affiliation through installed applications, and with community and social network through contact patterns, which makes proxy discrimination a property of the method rather than a risk of poor implementation.

It is aimed at thin file and new to credit borrowers, the population least able to contest an outcome. Against that the company states its mission includes extending fairer and more accessible credit, and names its risk score for inclusion, and publishes nothing to support either. No fair lending testing, outcome analysis across borrower groups, adverse action handling or explainability mechanism was located. Tenth instance of this pattern and among the strongest.

Nothing binds the vendor and nothing serves the borrower. No guarantee, indemnity or accuracy commitment was located. A person is scored on parsed bank statements and on inferences drawn from their own mobile device, receives a decision in real time with no human involved, is not told which system produced it, cannot see the signals that counted against them, and has no described route to correction or appeal.

The device dimension makes this worse than the comparable cases, because a borrower is unlikely to understand that their handset, applications or contacts formed part of the assessment at all. Third D on this axis after Lentra and TurnKey Lender, and the same reasoning applies with an additional layer.

The data chain is named where it runs through public infrastructure, with the unified lending interface, the account aggregator consent framework and the open commerce network all identified, and identity verification and bureau orchestration described as parts of the stack. That gives a lender visibility into the regulated rails behind a decision, which is more than most credit vendors disclose.

What is not published is the rest: no individual bureau, data provider or identity source is named, no model provider appears for the scoring or parsing components, and no subprocessor list or hosting arrangement was located, which matters most for the device product where the signal collection mechanism is itself undisclosed.

Integration is the product rather than a feature of it. The platform is interface first and software development kit based throughout, connects into the national digital lending rails including the unified lending interface, the account aggregator framework and the open commerce network, and is stated to slot into the systems already running a lender's ecosystem.

The effect is quantified in the terms a buyer weighs: a new non bank lender's build or buy decision compresses from a twelve month engineering project to a four to eight week integration, and a platform business can have credit embedded in days. More than 130 live enterprise deployments, including at two of the country's largest banks, evidence that the connections work at institutional scale rather than only in principle.

No hosting provider, region selection, residency commitment or private deployment option was located. The home market applies its own localisation requirements to financial data and the company has funded international expansion, so where borrower records rest and whether that changes as it moves into new jurisdictions is a live question. Nothing published addresses it.

The company publishes no list prices, and an unusually specific independent account of its commercial model exists. Contracts are described as structured by product module plus per interface call volume, with indicative unit ranges given for each: a modest per statement fee for bank statement parsing, a per borrower fee for device based onboarding that varies with signal depth and consent flow, and a low per decision fee for the risk score, with committed volume discounts and material savings on multi module contracts.

Total annual contracts for mid sized lenders are placed in a stated band. That lets a buyer model cost against expected volume before any conversation, which is what this axis measures, though the review publishing it explicitly flags the figures as indicative and to be validated by quote.

Within one market the coverage is close to complete. Buyers span private and public sector banks, non bank finance companies, fintech lenders and platform businesses embedding credit into their own products, and the lending lifecycle is covered end to end from multi channel origination and identity checks through underwriting, real time decisioning, disbursal and post disbursement portfolio monitoring with continuous rescoring, plus partnership and co lending arrangements.

Loan types run from personal and working capital lending to buy now pay later and small business credit, with thin file and new to credit borrowers named as a target segment. The concentration is geographic rather than functional, with international expansion funded but not evidenced.