Refine Intelligence

The removal test leaves investigators telephoning customers ad hoc, which is the state the product replaces. Models are trained on a proprietary dataset of genuine customer activity drawn from millions of financial records to establish a baseline of ordinary behaviour, then map each alert to the life stories most likely to explain it, ranked by probability. Recognising that a pattern of deposits looks like a house sale rather than structuring, and ranking that against alternatives, is the whole product.

The design keeps the investigator in place and makes them better informed rather than replacing them. Alerts are mapped to candidate explanations ranked by probability, which is a recommendation the investigator weighs rather than a disposition, and the team receives a real time explanation of the activity and source of funds within a clear audit trail.

Customers can resolve straightforward alerts themselves through structured inquiry, which removes work without removing judgement from the cases that need it. Held at B because no threshold is described for what clears automatically versus what escalates.

Outputs are described as probability ranked, which implies calibrated confidence and is the right presentation for a recommendation an investigator must weigh, and no accuracy, precision or validation figure is published for life story matching.

The error asymmetry deserves attention the company does not give it: wrongly greenflagging genuine criminal activity clears an alert that should have proceeded, which is the failure a supervisor would examine most closely, and nothing describes how often that occurs or what guards against it.

A 13 million dollar seed round was co-led by two established venture firms with participation from a security specialist fund and, notably, the corporate venture arm of a United States commercial bank, which signals that an institution assessed the approach before investing.

The most substantive evidence is derived rather than asserted: the company reports that analysis of its banking partners' alert volumes found 64 percent attributable to five everyday scenarios, a figure that could only come from live deployments. Founders are described as serial entrepreneurs who previously built fraud detection companies. No customer is named and no volume or handling time reduction is published.

No boundary statement was located, and the architecture makes the question central rather than incidental. The core asset is described as a unique proprietary dataset of genuine customer activity built from millions of financial records, and a model that improves as it observes which explanations prove correct.

Nothing states where those records came from, whether one bank's customers and their explanations inform life story matching for another institution's alerts, or what a client contributes by deploying the platform.

No data protection agreement, retention schedule, subprocessor list or consent framework was located. Two distinct holdings raise the question: the customer responses gathered through outreach, which describe source of funds, relationships with beneficiaries and personal circumstances such as a house sale or a family gift, and the proprietary training dataset assembled from millions of financial records whose origin and permissions are not described. Both are highly sensitive and neither is addressed.

No attestation, certification, trust centre or enumerated framework was located. A bank's venture arm invested and banking partners are in production, so security assessment has been passed privately, and for a platform that contacts a bank's customers directly and holds their explanations of personal financial activity, a published control set is the disclosure other institutions would expect before permitting outreach under their own brand.

One detail demonstrates genuine legal domain knowledge rather than borrowed vocabulary: the outreach process is explicitly designed to avoid tipping off risk, which is the statutory prohibition on alerting a customer that they may be subject to a suspicious activity investigation. Building customer contact that gathers context while respecting that prohibition is a real design constraint most vendors would not think to name. Enhanced due diligence is identified as a supported workflow. Held at B because no regulator, statute or jurisdiction specific rule is named across three regions of stated operation.

The entire premise is a fairness argument and the supporting figure is empirical: 64 percent of alerts at partner banks trace to five ordinary scenarios including selling property, cash intensive work, gifts, buying a car and paying construction costs, meaning most people caught by these systems are doing something unremarkable, and disproportionately those paid in cash or moving money for family reasons.

The stated mechanism is right too, since context is gathered through structured, consistent questions described as bias free, which removes the investigator discretion where differential treatment usually enters. Held at B because no testing, subgroup analysis or clearance rate by population is published, and because the training baseline of good behaviour will reflect whichever customers it was built from.

This is among the few vendors in the index where the affected individual has an actual route, and it is built into the product rather than promised in policy. Automated digital inquiries let customers explain their own activity and resolve alerts themselves, which means the person whose account was flagged participates in clearing it instead of waiting invisibly while an investigator decides.

The company describes making the customer a partner in fighting financial crime, and one playbook covers educating customers about cash structuring so they understand why ordinary behaviour triggered scrutiny. Held at B because this is alert resolution rather than a formal right of appeal, and nothing describes recourse for a customer whose explanation is not accepted.

The training corpus is characterised only by scale and character, described as a unique proprietary dataset of genuine customer activity patterns drawn from millions of financial records, with no statement of origin, licensing, permissions or whether it derives from customer institutions. For a company whose differentiator is that dataset, its provenance is the material disclosure, and no model provider, hosting arrangement or subprocessor list appears either.

The product is positioned deliberately as an overlay, described as bolstering existing monitoring rather than replacing it, which is the correct architecture because transaction monitoring systems are deeply embedded and validated with regulators, and displacing one is a multi year exercise. Prepared playbooks for distinct alert types suggest configuration rather than bespoke integration per use case. No named monitoring, case management or core banking system appears and no developer documentation was located.

No hosting provider, region selection, residency commitment or private deployment option was located. Stated operations span North America, Europe and Latin America, and the platform holds customer explanations of personal financial circumstances, which several of those jurisdictions treat as requiring local processing, so residency is a question a buyer would raise early and published material does not answer.

No pricing, packaging or basis of charge was located. The value case is framed around investigator time, customer churn and freeing staff for revenue work, without figures attached, and nothing indicates whether charge follows alerts processed, customers contacted or institution size, which matters for a product whose value scales directly with alert volume.

The platform serves fraud, anti money laundering and compliance teams inside banks, with prepared playbooks spanning cheque fraud, money laundering investigation, enhanced due diligence, scams and cash structuring education, so one deployment covers several distinct alert types rather than a single workflow. Expansion is stated across North America, Europe and Latin America. Held at B because no institution is named in any of those markets and the buyer remains banks specifically rather than the wider set of regulated firms carrying the same obligations.