Transparently.AI

The models are the entire product and they are the company's own, described as around 200 proprietary financial models trained on millions of data points and built by replicating the analytical behaviour of forensic accountants, activist short sellers, credit analysts, equity analysts, auditors and academics. The company states explicitly that its algorithms are its own creation rather than an off-the-shelf solution. A generative assistant sits on top for conversational interrogation. Remove the models and there is no product, since it sells neither data nor workflow.

The design is explicitly investigative rather than conclusive. Findings are grouped into 14 risk clusters so a team can prioritise where to look, forensic reports carry detailed explanations of every red flag with suggested next steps, and the assistant returns the specific question to put to a chief financial officer rather than a verdict. The company frames this as knowing where to look being half the problem solved. Held at B because a letter rating and numerical score are still assigned to named public companies, and no threshold or review process governs how those propagate into investment decisions.

Method disclosure is well above category norm. The architecture is stated as roughly 200 factor models organised into 14 named risk clusters covering areas such as accrual anomalies, asset quality and actions taken during the financial year to engineer outcomes, the score is defined precisely as the joint probability of manipulation and consequent collapse, and reports explain every red flag with its underlying data points.

The company states its research evidences predictive power for both corporate collapse and shorter-term security returns. Held at B because that research is referenced rather than published, and the 90 percent collapse prediction figure carries no stated sample, period or false positive rate.

Customers are characterised rather than named, covering one of the world's largest sovereign wealth funds, a global commercial bank and one of the four largest audit firms, with one asset manager named in a published case study. A major United States asset management group invested in the pre-Series A round, which is a form of validation from within the buyer category. Coverage extends to more than 85,000 listed companies. Held at B because the headline accuracy claim of predicting over 90 percent of corporate collapses up to three years ahead is self-reported with no methodology, sample or period disclosed.

No boundary statement was located. The input data is public, which removes the usual concern, and the residual question is the client's own activity: a platform that sees which issuers a sovereign fund, a global bank and a major auditor are each examining holds a view of institutional intent that would be valuable if aggregated, and nothing states whether query activity is isolated, retained or used.

No data protection agreement, retention schedule or subprocessor list was located. Personal data exposure is low by construction, since the analysis operates on published corporate financial statements rather than customer records, and what the platform does hold is a record of which companies each client is investigating, which is commercially sensitive in its own right for an investor building a position.

No attestation, certification, trust centre or enumerated control set was located. A sovereign wealth fund, a global bank and a major audit firm have presumably completed supplier review before adopting, so assurance exists privately, and nothing is published for a prospective buyer, which is a gap made more visible by the self-serve purchase route the company promotes.

No regulator, statute, accounting standard or supervisory expectation is named. The product operates adjacent to audit and financial reporting regulation and to market abuse considerations, since ratings on listed companies inform trading decisions, and none of that framework is mapped. The company publishes commentary criticising audit quality and assurance standards without identifying the rules its own output sits within.

No fairness testing or governance disclosure was located, and the exposure here falls on the companies being scored rather than on consumers. A model assigning a public manipulation probability to a named issuer can move its cost of capital and its share price, and systematic tendencies in what the models flag, whether by jurisdiction, accounting regime, sector or company size, would propagate at scale. Nothing addresses false positive rates or differential behaviour across those dimensions.

No guarantee, indemnity or correction process was located, and the affected party here is unusual in this index: the scored company, which has no commercial relationship with the vendor. A listed issuer assigned a high manipulation probability has no described route to see the analysis, correct a misread accounting treatment, or contest a rating that institutional investors and auditors may act on. Nothing states whether issuers are notified or given any right of reply.

The models are the company's own and it says so plainly, describing the algorithms as its one-of-a-kind creation rather than an off-the-shelf solution, which removes external model dependency for the risk engine. No base model or provider is identified behind the generative assistant, and no financial data provider is named for the statement data covering more than 85,000 companies, which is the input determining what the models can see.

Interface access is offered specifically so that fraud detection and predictive analysis run inside the client's own risk management tools rather than in a separate destination, which the company presents as the route for analysing very large portfolios, and auditors are described as reaching risk assessments within their existing tools. A browser-based product covers users who do not integrate. Held at B because no portfolio, risk or audit system is named individually and no developer documentation was located.

No hosting provider, region, residency commitment or private deployment option was located. Residency matters less than usual because the analysed data is public company filings rather than client or customer information, and the location of client query activity and generated reports is still undisclosed.

A self-serve route exists and the company explains why it built one, stating that customers can purchase and activate the software without going through a lengthy sales process, which is a meaningful commitment for an institutional analytics product where procurement usually takes months. Held at B because no rate, tier or basis of charge accompanies it, so a buyer knows access is easy without knowing what it costs.

Buyer types span asset managers, banks, auditors, exchanges, sovereign investors and risk functions across lines of defence, each addressed with a distinct use case from stock selection through due diligence to audit planning. Company coverage is global across more than 85,000 listed issuers. Held at B because the product addresses one analytical problem rather than a range, and institutional presence is stated by category rather than evidenced by count.