Complead

Fusion's agents investigate and close routine alerts, models resolve whether differently spelled names belong to one person, and language models classify adverse media by offense, role and severity across more than 300,000 articles a day. Fraud runs on Complead's own device and behavioral signals, with a supervised model trained on each institution's confirmed fraud beside unsupervised anomaly detection.

Underneath sits the data layer, the consolidated lists and the country by country PEP dataset, which its founder calls the company's largest asset. That data and the rules engine ran for years as Sanction Scanner before the agents arrived, and the agents now do the investigative work on top of it.

Agents gather the data on an alert, check it against the customer's profile and prior decisions and write the reasoning. They act on confidence bands each customer sets, for example closing above 90 and escalating below 70, and Complead puts routine closures at 77 percent. The terms bar refusing, restricting or reporting anyone on automated output alone, relationship decisions stay with a person, and a SAR leaves only after an MLRO signs it. The agent works as its own user in the customer's account, so every step it takes is logged with its inputs, its output and the policy or person that authorized it.

Risk rating weights are versioned with author and date, so the model in force on any past date can be reproduced, and changes are simulated against the live portfolio before release. Transaction monitoring backtests 12 months of history in about 30 minutes and shows the rule and version behind every alert. Agent rules can dry run on live alerts, and fraud decisions record the model version used. Complead reports false positive cuts of up to 97 percent from entity resolution, but no figure for missed matches or recall.

Octet, a trade network of 4,500 member companies in 72 countries, cut false positives by 70 percent through whitelisting and case management, screening in about 150 milliseconds. AirHelp cut daily operational time by 62 percent on passenger payouts, and Moldcell screens about 100,000 transactions for 1.5 million users. DLL Group runs its AML controls automatically through the API, and an officer at each of these customers speaks to the results.

Ateshgah Life, GonderAL, Quick Finans, BPN and Unity also vouch for the product, without figures. Complead reports more than 800 customers in over 80 countries, and Chartis named it a Disruptor in the RiskTech100 for 2027.

Complead says customer data never trains a model serving anyone else and that nothing is pooled across tenants, including for benchmarking. Several features draw on shared signals. Risk assessment templates come preweighted from consortium data, and peer benchmarks set a customer's weightings against anonymized data from similar institutions. The Nova assistant compares signals with sector medians, and monitoring rules can be shared across members on an opt in basis. The terms also license submitted content to operate and improve the service, and Complead does not name the language model provider or its retention terms.

Retention on screening records is set by the customer, and on exit Complead exports the data, deletes it within 30 days and confirms the deletion in writing. A data processing agreement with standard contractual clauses and a subprocessor register naming each entity, purpose and region are released under NDA within one business day, and subprocessor changes are announced 30 days ahead.

The privacy policy, updated in September 2026, sets retention periods, such as five years after a relationship ends for AML and compliance records. It covers transfers under standard contractual clauses and the UK addendum.

ISO 27001 and ISO 9001 certificates for 2026 to 2027 download directly from the trust center. A SOC 2 Type 2 report, penetration test summary, security whitepaper, CAIQ and SIG questionnaires and a cyber insurance certificate are released under NDA within one business day. Data is encrypted with TLS 1.2 or above in transit and 256 bit AES at rest, and the first status update on an incident comes within 30 minutes. An independent firm tests the platform every year, and customers may run their own test against a dedicated environment.

Complead has classified its platform under the EU AI Act and lists the obligations it accepts regardless of the result, though it does not disclose the classification. Filings go out in goAML XML, FinCEN SAR and CTR, Türkiye's MASAK format and UK and EU national templates, with direct submission where a regime allows it and a second reviewer signing off first.

Models are versioned, and a change can be tested against the customer's own historical cases before it reaches production. Entity resolution weighs diacritics, abbreviations and transliteration across Latin, Cyrillic, Arabic, Greek and Chinese scripts, which shapes how screening treats names that do not follow English spelling. Complead does not break out match rates or false positives by demographic group.

Complead's site terms, which cover its free tools and trials, cap liability at the greater of fees paid over the prior 12 months or $100 and exclude regulatory fines, reputational harm and data loss. They warn that screening results may be incomplete or inaccurate and make the customer solely responsible for its decisions, while paid services run under a customer agreement that is not public. Every agent action is logged against its inputs and authorizer, and audit trails export for a regulator.

List data is collected from the authorities that publish it rather than bought from resellers, and dated snapshots keep every past state. Complead does not name the language models, or their provider, behind adverse media reading, agent narratives and the Nova assistant. Its public subprocessor list gives only categories and regions, with entity names under NDA, and Microsoft Azure hosts the platform.

Thirty three named integrations include Temenos, Thought Machine, Mambu, Finastra, Oracle FLEXCUBE, Intertech and Fimple. Data providers LexisNexis, Moody's, Dow Jones and LSEG, identity vendors such as Sumsub, Persona, Onfido, Jumio and Veriff, and data platforms from Snowflake to Kafka are among them as well. Payment screening parses SWIFT MT and ISO 20022 messages field by field with a 62 millisecond median, and the REST API is idempotent with replayable webhooks. Sanction Scanner is available as a connector for Claude that acts only with analyst approval.

Hosting is on Microsoft Azure, with customers choosing EU, UK or Türkiye residency and data kept in that region for processing and backup. Production is isolated, defined as code and reached only through logged, time limited access. A US company serves American contracts, though Complead names no US region. Recovery targets are four hours to restore service and no more than one hour of lost data.

The Sanction Scanner screening product carries a public price. Its Starter plan costs €990 a month for 100 screened entities, scaling to 2,000, with two months free on annual billing and cancellation at any time. Starter covers individual, company, vessel, crypto wallet and aircraft screening, PEP lists, adverse media, daily monitoring and basic API access. Transaction monitoring, risk assessment, the rules engine and SSO sit in a custom priced Enterprise plan, overage is billed at standard rates Complead does not list, and integration carries no fee.

Banks and neobanks, payment and electronic money firms, fintechs, insurers, investment firms, lenders, crypto businesses and gaming operators each have their own configuration. Named customers span Kuveyt Türk and QNB in banking, iyzico, Papara and Moka United in payments, GonderAL and BPN as money transfer agents, Zurich, Generali and NN in insurance, and DLL Group in vendor finance.

More than 800 customers in over 80 countries also include corporates such as BMW, Delivery Hero and Unity and a United Nations agency, UNOPS. Screening covers US, EU, UN, UK, Australian and Swiss sanctions regimes. KYB reaches more than 160 countries, going to registries directly where they are reachable and to licensed sources elsewhere, with no fixed limit on ownership depth. Transaction monitoring includes a payments library of 38 scenarios, alongside payment screening and fraud.