ClearSale

The removal test lands differently here than for any other vendor in this lane, and the company's own positioning is why. ClearSale sells AI plus human rather than AI alone: every transaction is scored by models, but every flagged transaction is then reviewed by a person, under a stated policy of never declining automatically, staffed by more than 2,000 specialist analysts covering more than 160 countries around the clock.

Strip the models from Riskified, Signifyd or Sift and there is no product left. Strip them from ClearSale and a large, experienced manual review operation remains, working from twenty five years of accumulated fraud data across a stated five trillion dollars of analysed transactions. It would be slower and far more expensive, and it would still function.

The models are real and do the volume work, the marketing describes cutting edge statistical technology, and the data asset behind them is genuine. But the differentiator the company itself sells, and the thing a buyer is actually paying for, is the analyst layer sitting on top. That is a materially less model centric business than its competitors and the grade should say so.

The oversight design is the strongest in this lane and it is structural rather than optional. ClearSale operates a stated policy of never declining a transaction automatically: models score every order, and any order the models flag is routed to a human fraud analyst before any refusal is issued, backed by more than 2,000 specialists working around the clock across more than 160 countries.

Compared with the closest equivalent elsewhere in this lane, where a merchant may resubmit an already declined order for expert re review, this places the human before the adverse decision rather than after it, which is the difference between an appeal and a safeguard. Enterprise customers additionally get custom decision rules, and real time reporting is offered to make the decisioning visible. Two things hold it at this grade rather than above.

The merchant has less control than the score based alternatives, since the company states plainly that customers do not manage fraud rules themselves and that it handles decisions on their behalf. And no specification is published for the review itself: no turnaround commitment, no analyst authority definition, no criteria, and no statement of what proportion of flagged orders a human actually overturns.

A falsifiable accuracy commitment carries this, and it is rare enough to be worth stating plainly: the company advertises a 99 percent accuracy service level, with priority service level guarantees named as part of the enterprise tier, which means performance is a contractual term a customer can hold the vendor to rather than a marketing adjective. Almost nothing else in this lane offers a number of that kind at all.

Supporting it are a training base described as twenty five years of fraud data across more than five trillion dollars of analysed transactions, and the human review layer, which functions as a documented control on model output since a person inspects what the models flag before action is taken. Against that, the documentary half is missing entirely.

No validation report, precision or recall measure, false positive rate, model documentation or monitoring statement was located, and no methodology is published for the 99 percent figure. Confidence in published numbers is further weakened by the customer count moving from 5,000 to 6,000 to more than 100,000 across three sources without explanation.

Named customers with quantified published results, at scale, across a long history. Three case studies carry figures: Bloom reaching a 99.8 percent approval rate alongside a 386 percent revenue increase, Motorola recovering 2.4 million dollars annually from false declines previously rejected under strict fraud rules, and Samsonite holding a chargeback rate below 0.1 percent across fifteen countries without any merchant side fraud review team.

The named logo set is blue chip and international, covering Motorola, Samsonite, Victoria's Secret, AutoZone, Office Depot, Bloomberg, SSENSE, Azul and Dafiti, with the last two giving genuine Latin American depth that nothing else in this lane has. Scale is stated at more than five trillion dollars of transactions analysed across more than 160 countries with a 99 percent customer retention rate.

The company was a listed public issuer on Brazil's B3 exchange until April 2025, so audited historical financial disclosure exists, and a third party estimates 2024 revenue near 115 million dollars. One caution: the customer count does not reconcile across sources, running at 5,000 in 2022, 6,000 in 2024 and more than 100,000 merchants on the current site, which most likely reflects the Brazilian book being counted differently after the Experian combination but is nowhere explained.

The usual network question applies and a second one arrives with the new owner. On the first, twenty five years of fraud data across a stated five trillion dollars of transactions and more than 160 countries is a shared asset built from customers' order flow, described as localised fraud intelligence adapting as a merchant scales, and nothing published states whether a merchant can decline to contribute, how one merchant's data is separated from another's view, or what happens to contributed data when a merchant leaves.

On the second, and this is specific to this record, ClearSale now sits inside Experian and its Brazilian operation runs through Serasa Experian, a credit bureau. A fraud platform holding merchant transaction data inside a credit reporting group raises an obvious and consequential question about whether data flows either way between the fraud models and the bureau's products. Nothing published addresses it in either direction, and a buyer with obligations about how their customers' transaction data may be used should ask before signing.

Compliance is asserted and not evidenced. The site states GDPR compliance as a badge on the home page and publishes a privacy policy, and the Brazilian operation necessarily falls under that country's data protection regime through Serasa Experian, but no regime beyond GDPR is named specifically anywhere located.

Across this session's retrieval no data processing addendum, subprocessor list, retention schedule, named supervisory authority, transfer mechanism or standard contractual clause provision was obtainable. The exposure deserves better documentation than that.

This platform ingests consumer order, identity and behavioural data across more than 160 countries and, distinctively, routes flagged transactions to human analysts, which means named individual consumer records are read by people rather than only processed by machines.

Nothing published describes what analysts see, how that access is controlled or logged, how long reviewed case material is retained, or where the review operation is physically located relative to the consumer whose order is being examined.

Two frameworks are named and neither is claimed as a certification, which is exactly the failure the credential test exists to catch. The company's own knowledge base states that it is PCI compliant and follows ISO 27001 best practices, and its about page states that the platform adheres to Payment Card Industry Data Security Standards guidelines set forth by the standards council. Following best practices is not certification, and adhering to guidelines is not an attestation.

No compliance level is given for the payment card standard, no Attestation of Compliance or qualified security assessor is named, no ISO certificate or registrar appears, no SOC 2 or SOC 3 report exists in any form, and there is no trust centre, penetration test summary or subprocessor disclosure. Home page badges reading global compliance and secure by design carry no underlying artifact.

Single sign on and a service level guarantee are sold at the enterprise tier, which are operational features rather than assurance. Pre emptive negative finding: restating the same two framework names more prominently will not move this grade. Naming the payment card level with an Attestation of Compliance, or converting best practices into a certificate with a named registrar and scope, is what would.

No licence, no supervisory relationship and no regulatory approval was located, which is expected for a technology supplier and is not itself a deduction. What is worth recording is a loss rather than an absence. ClearSale was a listed public issuer on Brazil's B3 exchange under the ticker CLSA3, carrying the audited financial disclosure and continuous reporting obligations that listing imposes, until Experian acquired it on 1 April 2025 and took it private.

A buyer evaluating this vendor today therefore has materially less independently attested information available than one evaluating it in 2024, and the acquisition price was never disclosed. Compliance claims that do exist are asserted against standards rather than authorities: adherence to Payment Card Industry guidelines and GDPR compliance, both credited or discounted on the axes where their evidence sits. No named regulator appears as a counterparty and no payments directive or transaction risk analysis regime is claimed.

One published figure that looks like governance and is not, and then nothing. The company advertises 99 percent accuracy as a service level, which is unusual and is credited on the model risk axis as a contractual commitment, but it is a promise rather than a measurement: no methodology, denominator, time period, false positive breakdown or independent verification accompanies it, so a reviewer cannot tell what is being counted as accurate. Beyond it the axis is empty.

No error rate decomposition, confidence measure, fairness testing, disparate impact analysis or coverage statement exists. That matters more than usual for this vendor, because human analysts make the final call on flagged orders and human review carries its own bias exposure alongside the model's, yet nothing documents review criteria, analyst calibration, consistency measurement between reviewers, or how a decision is escalated when analysts disagree. A 2022 initiative on more inclusive fraud science terminology addressed the language of the field rather than the behaviour of the models.

Two commitments run in parallel here where most guarantee vendors offer one. The chargeback guarantee covers the full cost of a fraud chargeback on any order ClearSale approved, is included in the published Growth plan rather than sold as an extra, and the company claims to have originated the model.

Alongside it sits a 99 percent accuracy service level, so the vendor is on the hook both financially for the orders it wrongly approves and contractually for the rate at which it is right, which is a more complete answer than a guarantee alone. ChargebackOps extends the posture into managing disputes end to end on the merchant's behalf. For the consumer the position is better than anywhere else in this lane, though still not a remedy.

Because no order is declined without a human analyst reviewing it first, a shopper refused by ClearSale was refused by a person who looked at the case rather than by a threshold, which is a meaningful difference in kind. It is still not recourse: that shopper is not told, cannot see the evidence, has no appeal route and cannot reach the company, and no service level attaches to correcting a wrong decline.

Nothing in the chain is named, and the new ownership makes one omission conspicuous. No third party data supplier, identity provider, device intelligence source or enrichment service is identified anywhere, no subprocessor list exists in public, no cloud infrastructure provider is named, and no model or foundation model provider is stated for any component.

The company describes its own accumulated fraud data across twenty five years and five trillion dollars of transactions as the input, which identifies the internal source and none of the external ones. The specific gap worth a buyer's attention is the parent.

ClearSale is now an Experian company with its Brazilian operation inside Serasa Experian, and Experian is one of the largest consumer data businesses in existence, so the single most plausible external data source for these models is the group that owns them. Whether bureau data feeds the fraud models, and on what basis, is disclosed nowhere. Named corporate acquisitions covering logistics, consulting and chargeback management describe how the company was assembled rather than what feeds its decisions.

Deep on the commerce layer and absent on the payments layer, which is precisely what separates this from the top of the axis in this lane. More than eighteen native integrations are claimed and the significant ones are named: Shopify, Magento, BigCommerce, WooCommerce, Shift4Shop, VTEX, Oracle Commerce, Salesforce Commerce Cloud and PrestaShop, alongside a REST API and published developer documentation.

Deployment is characterised as one click install or plugin, with most teams live within days and setup described in minutes, which is a genuine difference from the event instrumentation the score based competitors require. An AI Agents Portal exposes the platform for machine readable commerce, with the site itself offering agent oriented affordances for opening and copying page content, which is a forward looking integration surface few competitors have built at all.

What is not present is any named payment gateway or acquirer connector. The competitors holding the top grade here reconcile chargeback data directly from named processors, and for a vendor whose core product is a chargeback guarantee the absence of that layer is conspicuous.

Hosted software consumed through platform plugins or a REST API, with no private, single tenant or on premise option located. On residency nothing is published: no cloud provider, region, data centre or country of processing is named, no region selection is offered, and no transfer mechanism appears for customers outside the operating jurisdictions.

The gap is larger here than for a single market vendor because the operation is explicitly multi jurisdictional, spanning more than 160 countries with a distinct Brazilian entity running through Serasa Experian and a separate Spanish language Latin American presence, so processing plainly happens in several places and the buyer is told which none of them are.

The human review layer sharpens the question further, since flagged consumer orders are read by analysts whose physical location is never stated, and a European or Latin American merchant cannot complete its own transfer assessment without knowing where those analysts sit.

This is the only vendor in the ecommerce fraud lane that publishes an actual number, and it is worth being precise about what that number does and does not tell a buyer. Growth plans carrying the chargeback guarantee are advertised as starting as low as 250 dollars a month, with pricing stated to scale on monthly transaction volume and risk profile, no upfront fees, no setup fees, and chargeback protection on approved orders included rather than sold separately.

A second tier is defined by a stated threshold, merchants processing more than 50,000 orders a month, and its additions are enumerated as advanced reporting and controls, custom integrations, service level guarantees, single sign on and volume discounts. A buyer therefore knows the floor, the metering basis, the tier boundary and what upgrading buys, all before contact.

What they still cannot determine is their own rate, because the percentage or per order figure that applies at their volume and risk profile is never published and every path ends at a custom quote. A published floor is a real disclosure and it is not a rate card.

One buyer type, served across an unusually wide band within it. The customer is the online merchant, and unlike the enterprise focused vendors in this lane ClearSale addresses that customer from small business upward, explicitly positioning for teams without in house fraud expertise and describing a path from startup through mid market into enterprise on the same platform.

Vertical coverage is enumerated across electronics, fashion and apparel, travel and airlines, automotive parts, sporting goods, home and garden, food and beverage, luxury goods, and marine and robotics. Geographic reach is the genuine distinction: more than 160 countries with 2,000 analysts, a dedicated Brazilian operation running through Serasa Experian and a separate Spanish language Latin American presence, which is regional depth no competitor in this lane demonstrates.

None of that changes the buyer. This vendor does not sell to banks, payment providers, insurers, regulators or capital markets firms, and the second platform aimed at payment providers that lifts one peer above this grade has no equivalent here.