Wiserfunding

The model is the whole company, and the model is substantially statistics rather than learning. On centrality there is no question: strip out the scoring and nothing remains, since the platform exists to produce an SME Z Score, a bond rating equivalent, a probability of default and a loss given default, with the portfolio and monitoring features arranged around those outputs. What holds it out of the top band is technique.

The flagship score descends directly from a discriminant analysis model published in 1968, adapted for smaller companies through peer reviewed academic work, and a statistical scoring model of that lineage is a formula with coefficients rather than a learned function.

The company describes weaving artificial intelligence and machine learning into the platform to constantly update and improve the offering, and learned methods plausibly do the work of extracting signal from unstructured and non financial sources such as governance and management quality. But the asset the company is known for, and the one it publishes, is a statistical model.

An analytical instrument rather than a decision engine, and a customer describes exactly the right way to use it. The platform produces measures, a score, a bond rating equivalent, a probability of default, a loss given default, a debt capacity figure and a suggested credit limit, and hands them to a lender who decides. Nothing here approves, declines, prices or books anything, and no automated decisioning or straight through processing capability is described anywhere.

Lenders set their own parameters and thresholds. The clearest statement of position comes from a customer rather than the vendor, describing the value as a third lens, objective and methodologically distinct, used to triangulate risk assessments and validate internal thinking, which is a description of a second opinion strengthening a human judgement rather than replacing it. What is unpublished is any vendor statement of that boundary, including whether it would object to a lender wiring the score straight through to an automated decline.

The methodology is in the peer reviewed literature, which is a standard of transparency essentially nothing else in this index meets. Papers by the founders and named academic collaborators cover SME credit scoring, the adaptation of the original bankruptcy prediction model for smaller firms, and low default portfolios, so a lender's model risk function can read the actual approach, examine the assumptions and consult the criticism, rather than relying on a vendor's characterisation of its own work.

Interpretability is a design choice rather than an afterthought: expressing output as a bond rating equivalent deliberately renders assessments comparable across industries and markets and maps them onto a scale credit committees already understand, and a twelve month probability of default is a defined, standard measure. Input breadth is quantified at more than 100 financial metrics.

What is absent is current performance: across two passes no discrimination statistic, accuracy figure, validation result, sample, observation period or drift disclosure was located for the models actually in production today.

Reputation carries this record where customer evidence does not. Every customer testimonial published is unattributed, identified by no institution and no individual, and while several are specific about what the product did, including one describing tailored loss given default models built to satisfy a regulatory requirement and another describing the platform as a third and methodologically distinct lens used to validate internal thinking, none can be verified.

Named commercial relationships exist and are checkable, covering an invoice financing marketplace and an embedded lending platform. A company milestone timeline records reaching 30 clients with a team of 20, which is the only client count located and is not dated in current terms. Coverage of more than 45 countries and a Series A of 3 million pounds from a substantial United Kingdom growth investor establish the business is real and funded. What is entirely absent is outcome measurement: no default reduction, no approval rate change, no portfolio performance figure of any kind.

The methodology is open to scrutiny in a way that indirectly serves this axis, and nothing else is documented. Because the scoring approach is published in peer reviewed journals, the assumptions behind it have been examined by parties with no commercial interest in the outcome, which is a form of external review that no internal safety process replicates.

Customer control is offered through configurable parameters, customised alerting and optional bespoke modelling, so a lender can bound what the platform surfaces. Beyond that the record is empty. Across two passes no model card, evaluation methodology, red team result, incident disclosure or acceptable use boundary was located, and nothing describes how the machine learning components said to keep the models current are tested before they change a score. Nor is there any statement on what happens to a client's portfolio data, which for a monitoring product means an ongoing feed of a lender's exposures rather than a one time upload.

A structural advantage this record should credit, and no documentation around it. The subject being scored is a company rather than a consumer, so the compiled dossier problem underlying most of this index, where an individual is assessed without consent and never sees the result, largely does not arise. Corporate financial data drawn from registries and filings is substantially public information, and the vendor is not building profiles of private individuals.

The exception is the non financial assessment, which the company describes as covering corporate governance and management capacity, judgements that necessarily concern the named directors and officers of a business and that a small company owner would experience as personal.

Across two passes no privacy policy content, data processing description, retention schedule or subprocessor list was located, and no reference to the United Kingdom or European data protection regime appears despite incorporation in two European jurisdictions and data drawn from more than 45 countries.

Nothing was located. Across two passes no certification, attestation report, trust centre, security page, penetration test summary, subprocessor list or security contact appears on any surface, and no framework is named anywhere in the published material. This is the complete absence the grade describes rather than a stale or thin disclosure. The usual mitigating inference is unavailable here.

At larger vendors an empty security page can be discounted because dozens of named supervised institutions have plainly run third party risk assessments before onboarding, so the controls demonstrably exist even if the evidence is private. This company names no customer at all, so nothing establishes that any demanding party has examined it, and the backing of a growth investor speaks to financial diligence rather than to information security. A lender monitoring its live portfolio through this platform would be placing its exposure data with a vendor whose controls are undocumented in public.

An unregulated analytics supplier with one concrete piece of regulatory work evidenced from the customer side. A published testimonial states that the company built tailored loss given default models so the customer's risk function could become compliant with new requirements from the United Kingdom financial regulator, which is specific, checkable in principle and describes regulatory capital or provisioning work rather than generic compliance assistance.

Loss given default modelling to a supervisory standard is a demanding exercise and being trusted with it says something. It remains work performed for a client rather than standing held by the vendor. Across two passes no authorisation, registration, supervisory examination outcome or named statutory citation was located.

One absence is notable given the footprint: with incorporation in the United Kingdom and Italy and operations across multiple continents, no position on the European artificial intelligence regulation appears, despite creditworthiness assessment being expressly designated high risk under that regime.

A narrower exposure than most of this index, and still unaddressed where it exists. Scoring a company rather than a person removes the consumer fair lending question that dominates this lane, since a limited company has no protected characteristics and the equal credit rules do not reach corporate financial ratios. What does not disappear is the non financial half of the assessment.

The company states it evaluates intangibles including corporate governance and management capacity, which are judgements about the people running a business, formed from unstructured sources, and applied to firms across more than 45 countries with very different disclosure norms, corporate cultures and languages. A management capacity assessment that travels badly across jurisdictions would systematically disadvantage founders in some markets while appearing entirely objective.

Across two passes nothing published describes how those intangible factors are derived, weighted, validated or tested for consistency across countries, and no model card or fairness analysis was located.

No commercial instrument is published. Across two passes no terms of service, master agreement, warranty, indemnity, liability cap, service level or uptime commitment was located on any surface. The free trial reduces the risk of buying something that does not work, which is a practical protection and not a contractual one, and it addresses the wrong failure: the concern is not that the scores are useless but that a wrong score contributes to a bad lending decision.

Nothing published states what a lender is owed if a probability of default is materially misstated, if a monitoring alert fails to fire on a deteriorating borrower, or if a bespoke loss given default model commissioned to satisfy a regulatory requirement proves inadequate when the regulator examines it.

That last case is the sharpest, because a customer describes exactly that engagement and the consequence of it failing would fall on the lender's regulatory standing rather than on its loan book alone.

Intellectual provenance disclosed completely, data provenance not at all. The provenance disclosure is the strongest in this lane and is externally verifiable rather than asserted: the score descends from a named bankruptcy prediction model published in 1968, the adaptation for smaller companies was carried out by that model's own creator with named academic collaborators, and the resulting work sits in peer reviewed journals a buyer can read.

Nobody else in this index lets a customer trace the lineage of the model deciding their exposures back to its published origin. Input categories are described at a useful level, covering more than 100 financial metrics alongside non financial and macroeconomic signals drawn from public, private and unstructured datasets. What is never named is where any of it comes from.

Scoring companies across more than 45 countries requires registry, filings and financial statement data from many suppliers, none identified, so a buyer cannot assess coverage quality by market or concentration in any provider.

Delivered as hosted software described as easily integrated into existing systems, with two named commercial relationships demonstrating that the scores can be consumed inside somebody else's product rather than only through the vendor's own interface: an invoice financing marketplace using the assessments for default evaluation, and an embedded lending platform partnership.

Being embedded by partners is meaningful evidence that an integration path exists and works, and for a scoring product that is the deployment shape that matters, since the score has to arrive inside the lender's workflow to be useful. Beyond those two, nothing is inspectable. Across two passes no public interface documentation, developer portal, sandbox, connector catalogue or specification was located, and no origination system, core platform or data provider is named as supported. A lender cannot establish before contact how the score reaches its own systems, what latency to expect, or whether batch and single assessment routes both exist.

Hosted software with no placement detail published. The platform is described as a software as a service solution integrated into a customer's existing systems, and nothing further is stated. Across two passes no hosting provider, region list, tenancy description or residency commitment was located, and no customer hosted option is mentioned. The question has more weight here than the company's size suggests, for two reasons specific to this business.

Data is drawn from more than 45 countries, several of which restrict where corporate registry and filings data may be processed or resold, and the company is incorporated in both the United Kingdom and Italy with early operations established in India, so processing plausibly spans jurisdictions with different regimes.

And the monitoring product means a client's live portfolio exposures sit with the vendor on an ongoing basis rather than passing through once, which makes where that portfolio rests a supervisory question for a regulated lender.

No price, unit or tier is published, and two passes across the company's site, its product and platform pages, its insights archive and third party directories produced nothing on how the platform is charged. What lifts this above a bare contact form is that a free trial is offered through a dedicated page rather than mentioned in passing, which gives a prospective lender a route to evaluate the scores against its own portfolio before any commercial discussion, and for an analytics product that evaluation is the whole purchase decision.

Bespoke risk modelling and customised alerting are offered as options, and nothing indicates whether those are included, chargeable or scoped separately, which matters because at least one customer describes commissioning tailored loss given default models, which is consulting work rather than software configuration. Nothing published indicates whether charging follows companies scored, portfolio size, seats, countries covered or a platform subscription.

One subject, several lender types, and no institution named. The subject is narrow by design: small and medium enterprises only, which is the market the founders set out to serve and where they argue a five trillion dollar funding gap exists.

Within that, buyers are described by category rather than by name, spanning traditional and alternative lenders, neobanks, asset managers, embedded lenders and invoice financing marketplaces, and the two named commercial relationships fit that pattern, being a marketplace and an embedded lending platform rather than a bank.

Geographic coverage is genuinely wide for a company of this size, with data across more than 45 countries, incorporation in both the United Kingdom and Italy, an early entry into the Indian market and operations described across multiple continents. Against that: no named institution in any segment, a client count of 30 that appears only in an undated milestone, and a team of roughly 20 people.