Omnisient

Two technologies carry this product and only one of them is machine learning. The privacy layer is cryptographic rather than statistical, using privacy enhancing technologies and tokenisation so that parties never exchange records, and stripping the models leaves that clean room intact as a working product.

What the models do is the part the bank actually buys: discovering which alternative data sources predict repayment, building and testing scoring models inside the environment, and refining risk assessment, with a conversational interface letting non technical users query anonymised data directly. The outcome a lender takes away is a working score, not a secure room, which is why this sits at B rather than lower, but the cryptography would remain a business without a single model.

Autonomy exposure is structurally low because the platform builds and tests models rather than making lending decisions, and the institution deploys whatever it takes away. That division is not described anywhere in public material, which is the gap.

Nothing states who approves a scoring model developed inside the environment before it goes into production, what validation the platform supports on the way out, or whether the conversational interface over anonymised consumer data has any constraint on the questions it will answer. A tool that lets non technical users interrogate consumer behaviour in natural language needs a stated boundary and does not publish one.

Testing is a product function rather than an afterthought, which is what earns the grade. The platform is described as identifying, testing and validating whether a candidate alternative data source actually predicts repayment before an institution commits to it, so a bank can establish predictive value in a controlled environment rather than discovering it in production.

A published lift figure exists, with prediction of loan repayment reported as improving by 41 percent in a named collaboration, which is more measurement than most vendors here offer on any axis. What is absent is the rest of the package: no accuracy or error rates on scores produced, no validation documentation an institution could hand its own model risk function, and no statement of how a model built inside the environment is monitored after deployment.

The measured outcomes are among the strongest in this index and they are stated in the units that matter rather than in time saved. More than 8 million previously unscorable consumers have been scored through the platform, of whom 3.2 million now qualify for credit on alternative behavioural data alone, which is a population level result rather than a customer testimonial.

A named collaboration between banks and a major grocery retailer is reported to have improved prediction of loan repayment by 41 percent with a 29 percent projected increase in credit revenue. Deployment is stated with top tier banks and insurers in international markets, with a case describing Africa's largest bank.

The corroboration is unusual: TransUnion, a global credit bureau, co led the 12.5 million dollar Series A and holds a strategic relationship dating from mid 2025, which means a party with the expertise and incentive to test these claims examined them first. No individual institution is named, which is the one gap.

This vendor answers the question the rest of the index leaves open, and answers it structurally. Everywhere else the cross party data boundary is either unstated, described as anonymisation without a mechanism, or contained by tenancy.

Here the boundary is the entire architecture: participants collaborate without exchanging records, the environment is described as independent and neutral rather than operated in one party's interest, and the vendor positions itself as escrow between parties whose commercial interests conflict.

What makes it credible rather than aspirational is that both sides are protected against the same risk, with intellectual property and consumer privacy named as things the platform preserves for the data owner as well as the buyer. Fourth benchmark answer alongside Mitigram segregation, DiligenceVault consent, Saphyre permissioning and DwellFi tenant containment, and the only one where the parties are competitors by construction.

The strongest privacy position in this index because privacy is not a policy applied to the product, it is the product. Data is never exchanged or shared between collaborating parties and only insights move, enforced through privacy enhancing technologies, advanced cryptography and tokenisation as a service, with data anonymised before it is uploaded and computation happening inside an independent, neutral, certified environment neither party controls.

The architecture removes the failure mode rather than governing it: a retailer's customer records and a bank's applicant file are never in the same place, so there is no combined dataset for either party or the vendor to lose, misuse or be compelled to produce. Third A on this axis after Zeplyn and Rowspace, and the only one where the mechanism is cryptographic rather than procedural.

One framework is named and it is named in the right place, with the collaboration environment itself described as certified to the international information security standard rather than the certification sitting as a corporate badge. The company has also received external recognition specifically for security and consumer privacy protection, which is third party scrutiny of the thing it sells.

What holds this below the top grade is that only one framework appears: no service organisation control report, no trust centre, and no published detail on encryption, key management or how the tokenisation service is independently assured, which is the assurance a bank would want given that the entire value proposition rests on those controls holding.

Regulatory compliance is the central commercial claim and no instrument is named. The platform is positioned as the compliant route to consumer data under rising privacy standards, and it holds a named international information security certification, but no data protection statute, financial supervisor or consumer credit reporting regime appears in located material.

The gap is specific and consequential for the stated expansion: alternative data used to score consumers for credit in the United States engages the federal consumer reporting regime with its own accuracy, dispute and permissible purpose obligations, and the strategic partner co leading the funding round is itself a regulated bureau operating under exactly that regime. Naming compliance without naming the rule is describing the market.

The inclusion case and the proxy discrimination risk are the same fact, and that is the finding. Grocery shopping behaviour predicts loan repayment precisely because a basket encodes household size, income level, dietary and religious practice, health status, neighbourhood and shopping cadence, and several of those are protected characteristics or close proxies for them that fair lending law exists to keep out of credit decisions.

A model that works well on this data works well for that reason. Against that sits the strongest measured inclusion outcome in the index, with 3.2 million people gaining credit access on behavioural data alone, which is a real answer to the objection that alternative data merely relabels exclusion.

What is missing is any published fairness testing, proxy analysis or disparate outcome monitoring on models built in the environment, and the platform is where that testing would naturally live since it already tests each source for predictive power.

No guarantee, indemnity or falsifiable commitment on model or match quality was located. What sits above the floor is genuine and unusual: the architecture limits what any error can expose, since a wrong match or a failed model cannot leak identifiable records that were never exchanged, and both commercial parties retain their own data throughout. The party with no route is the consumer.

Someone whose grocery purchases were used to assess their creditworthiness gave that data to a retailer for an entirely different purpose, is not told a lender scored them on it, and has no way to see the inference, challenge it or ask that their behaviour be excluded. Anonymisation protects identity, not agency, and nothing published addresses the second.

Partial and useful. The most consequential third party is named and its role is unambiguous, with a global credit bureau both co leading the funding round and holding a strategic relationship to expand alternative data use, and bureau data appears explicitly alongside first party retailer data in a published use case. Supplying industries are characterised concretely rather than vaguely, covering retail, telecommunications, healthcare and smaller consumer categories.

What is not disclosed is the model layer: no provider is named for the machine learning tooling or the conversational interface over anonymised data, and no subprocessor list or hosting arrangement was located, which leaves open whether any external model provider touches data inside an environment sold on the premise that nothing leaves it.

The platform is an environment rather than a connector, and integration is the least developed part of the published story. Data owners are described as anonymising and uploading their data, which is a light touch path in, and no core banking system, loan origination platform, decisioning engine or bureau interface is named on the way out.

Nothing describes how a score or model developed inside the environment reaches the systems where a bank actually makes decisions, and no developer documentation or application programming interface reference was located. For a product whose output must land in an origination flow to be worth anything, that is a material absence.

The deployment property that matters most is stated and it is architectural: consumer data never leaves the owner in identifiable form, computation happens inside an independent and neutral environment, and only insights move between parties. That answers the question a data owner asks before participating at all, and it constrains exposure in a way a hosting statement could not.

What is not published is the conventional half: no cloud provider, region selection or residency commitment appears anywhere, which matters for a company operating across African, other international and now United States markets where cross border transfer rules differ and where the platform's whole proposition is regulatory safety.

No pricing, packaging or basis of charge is published for either side of the marketplace and every route in is a demo request. The question is more interesting here than for most vendors because there are two payers with different economics, the financial institution buying insight and the consumer brand monetising data, and the platform describes itself as sitting in an escrow like position between them. Nothing states how value is split, whether the data owner is paid per query or per outcome, or what a bank pays to test a source that turns out not to predict anything.

On the buying side the platform names banks, insurers and credit bureaus, which is three distinct institution types with genuinely different uses of the same alternative data. On the supplying side it reaches retailers, telecommunications operators, healthcare organisations and smaller consumer businesses including gyms and pharmacies, and matchmaking between the two is part of the offering.

Named uses span credit risk scoring, fraud and financial crime, customer acquisition and payment media network monetisation, so the platform is not confined to lending. Geographic evidence is concentrated in African and other international markets with United States expansion stated as underway rather than established, which is what holds this at B.