Quantexa

Entity resolution is not a supporting function here, it is the product, and it is model work. The company resolves identities across messy, inconsistent records using a predictive model that deliberately does not depend on pre resolved entity lists for training, which is the design choice that lets it work where deterministic matching fails, on poor data quality, transliterated names and complex ownership hierarchies. Network analytics then generate features from the resulting graph. Apply the removal test and what remains is rules based record matching, which is the legacy approach the platform exists to displace rather than a version of the same product.

The product is built to inform a human rather than replace one. Its stated value is intelligence led investigation, giving an analyst a single resolved view with the surrounding network already assembled so the work shifts from gathering data to judging it, and the generative assistant is scoped to report generation and research support rather than decisions. That is a sound posture and the effect on an investigator's day is real.

The disclosure is thinner than the leaders on this axis, with no described approval gates, no account of which determinations are automated versus surfaced, and no sampling or quality control mechanism published.

The transparency claim is unusually concrete for this axis. Describing the resolution model as transparent and tuneable, and stating that it operates without pre resolved entity lists, tells a validator both that the logic is inspectable and where the model's behaviour comes from, which is more than a general explainability assertion. Performance is quantified, citing accuracy improvements above 90 percent and analytical model resolution 60 times faster than traditional approaches. The formal package is still missing: no model documentation, no validation summary, no error rates for resolution itself and no stated position on supporting a customer's own validation.

The strongest evidence combination in this lane. Deployments are named at the top of global banking, including a globally systemically important bank automating counterparty data gathering, a second global bank, a large European bank refocusing customer due diligence resource on real financial crime, and a Nordic bank that came through a major money laundering failure, alongside national tax authorities.

Quantified outcomes are published across both the enterprise platform and the newer product for smaller banks, including false positives down 75 percent, investigations 80 percent faster and investigative effort more than halved. Third party validation runs to an analyst evaluation naming it a leader in its category and a commissioned economic study reporting 228 percent return over three years.

Two design choices carry real weight. The resolution model is described as transparent and tuneable, so an institution can inspect and adjust matching behaviour rather than accept a fixed black box, and explainable output is positioned as the company's differentiator rather than a compliance afterthought. Avoiding pre resolved training lists also avoids inheriting whatever bias sat in a curated match set.

What is absent is the error account: no published rate for false merges or false splits, no statement on which model providers sit behind the generative assistant, and no description of the boundary between customers on shared infrastructure.

The privacy surface here is distinctive and larger than a customer monitoring product's. Building a resolved view means consolidating internal records with third party feeds, public records and corporate registries to profile not only the institution's own customers but their counterparties and beneficial owners, which means constructing detailed relationship profiles of people who have no relationship with the bank and no way to know they have been resolved into its network.

That is defensible under financial crime obligations and it deserves an explicit account. None was located: no published privacy framework, retention schedule, subprocessor list or position on how counterparty data is treated differently from customer data.

This pass located no trust centre, enumerated certification list, attestation scope or audit period on the public site. The customer base includes globally systemically important banks and national tax authorities, whose third party assurance requirements are among the most demanding anywhere, so the actual control environment is certain to be substantial and to have been examined repeatedly. The grade records what an outside buyer can verify without entering diligence and should be revisited if a trust surface is published or located.

Quantexa supplies technology and holds no licence, the expected posture. Its regulatory grounding is demonstrated by deployment rather than by claim, since institutions adopt it precisely when supervisory scrutiny is most intense, including a bank rebuilding after a major money laundering failure, and it is trusted by national tax authorities and government agencies for their own enforcement work.

That produces the same structural position as the blockchain intelligence vendors in this index: the platform serves supervisors and the supervised. It is disclosed openly and buyers should understand it.

Entity resolution has a fairness problem that is specific to it and rarely discussed. A false merge conflates two people, which can attach one person's sanctions exposure or adverse media to another; a false split hides a real network. Name matching accuracy varies systematically by naming convention, transliteration, script and the use of patronymics, which means resolution error rates differ by national origin as a property of the technique rather than as a defect.

The company states its resolution is tuned for the variation patterns of global financial services, which acknowledges the issue implicitly. No per population accuracy, no error rate disclosure and no route for a wrongly resolved individual to seek correction were located.

A transparent and tuneable resolution model means an institution can inspect why two records were merged and adjust the behaviour, so a wrong outcome is diagnosable and fixable within the customer's control. Nothing extends beyond that. No accuracy guarantee, no published error rate for false merges or splits, and no route for an individual wrongly resolved into a network, or wrongly attached to another person's adverse media or sanctions exposure, to learn of it or seek correction.

The categories of input are stated clearly, covering internal systems, third party feeds, public records and corporate registries, and the cloud product for smaller banks names the hyperscaler it runs on, which is more infrastructure disclosure than most peers give. The specifics are absent: no individual data feeds or registry providers are named, no model provider is identified for the generative assistant, and no subprocessor list is published.

Ingestion breadth is the strength and it is genuinely hard engineering, pulling from internal systems, third party feeds, public records and corporate registries and reconciling them despite poor data quality, inconsistent identifiers and complex ownership hierarchies at the scale of a global bank's estate. Supporting multiple use cases from one instance means the same foundation serves financial crime, credit and customer intelligence rather than requiring separate deployments.

What was not located in this pass is the outward facing surface the strongest integrators publish: no named core system connectors, no public developer documentation, no status page and no partner directory.

The cloud product for mid size and community banks is delivered on a named hyperscaler, which is more than most vendors here disclose, and enterprise deployments at global banks and government agencies imply arrangements that meet their own hosting requirements. None of it is documented publicly: no hosting regions, no residency options, no transfer mechanisms, no tenancy separation description and no subprocessor list. The gap matters given customers operating under several supervisory regimes and public sector users whose requirements usually differ from commercial ones.

No rates, tiers or minimums are published and enterprise engagement runs through contact paths. One structural change is worth noting for buyers below the tier one level: the company has productised a cloud delivered anti money laundering offering aimed specifically at mid size and community banks on a named hyperscaler, which implies a more packaged commercial shape than the bespoke enterprise programme and makes the technology reachable for institutions that could never fund the latter. Neither the packaged nor the enterprise price is disclosed.

The range is wider than anything else in this lane and it runs in two directions at once. Vertically it spans globally systemically important banks down to community banks through a purpose built cloud product, which is a span very few vendors attempt.

Horizontally it covers banking, insurance, government including national tax authorities, and telecommunications, and within an institution the same resolved data foundation serves financial crime, customer due diligence, fraud, commercial credit risk, master data quality and customer intelligence, so the platform is bought by more than one function.