AML checks: what they are, in plain terms
An AML check (Anti-Money Laundering) is a set of procedures and controls that financial organizations and online services use to detect and prevent risky scenarios: dealing with sanctioned persons, attempts to circumvent restrictions, dubious counterparties, and other signs of the laundering of funds. In real-world processes, AML relies on collecting identification data (KYC), screening against external databases, and risk assessment. In solutions like NeuroVision, this logic is implemented as a single flow: data is extracted from documents via AI-OCR, normalized (including transliterations), then checked against connected sources and compiled into a report with an audit trail.
The international AML system functions through the unified standards of FATF (the Financial Action Task Force), which are implemented in the legislation of 200+ countries. In Russia, the requirements are enshrined in Federal Law No. 115-FZ, which obliges banks, payment systems, crypto exchanges and marketplaces to check their customers and their operations. Violations carry fines of up to 1 million rubles for organizations and the blocking of licenses.
As part of an AML check, NeuroVision analyzes dozens of parameters of the customer and their operations in fractions of a second — from the geography of payments and the frequency of transactions to connections with counterparties and beneficiaries — and simultaneously matches the data against connected sources. The platform uses machine learning and fuzzy-matching algorithms (including working with transliterations and alternative spellings) to find hidden risk scenarios that are hard to detect manually. In practice, NeuroVision performs online screening in less than a second against 1,700+ databases, and the accuracy of recognizing and matching names in different spellings is stated at 99.74% when sufficient identifiers are available.
The main goals and tasks of an AML check
The main goal of an AML system is to create a financial barrier for criminal capital at the entry point into the legal economy. The check blocks attempts to turn the proceeds of drug trafficking, corruption, terrorism and fraud into legitimate assets through bank accounts, investments or the purchase of real estate.
The primary task is the automatic filtering of transactions by signs of risk. The system instantly matches each payment against typical laundering schemes: splitting large sums into small transfers, chains of transit operations through offshores, a sharp increase in turnover without economic justification. When patterns match, the operation is blocked for an additional check.
A critically important function is real-time sanctions screening. In NeuroVision it is built on regular synchronization with key sources (for example, the UN, OFAC, the EU and other registries relevant to the business) and on precise matching technologies: data normalization, the generation of spelling variants, and fuzzy matching, in order to find not only “exact matches” but also close variants (typos, transliterations, rearrangements of parts of a name). When a significant match is found, the system can stop the operation, generate an event card, and pass it on for manual verification by a compliance specialist together with the source, the type of list, and the match score.
A mandatory component is the assessment of the reputational and compliance risks of the customer and related persons. In practical scenarios, this includes checks against sanctions, PEP lists and other relevant registries/sources. In NeuroVision, such checks are supported through connected databases and flexible attribute configuration (full name in different languages, date of birth, documents, addresses), and the results are recorded in a report — this helps make a decision before starting cooperation, without relying on manual search and scattered sources.
Regular screening of the existing customer base ensures compliance with current legal requirements. A customer’s status can change: falling under sanctions, a change in connections/affiliation, the appearance of new restrictions in the lists. Therefore, it is important not only to “check at onboarding” but also to build in re-screening and change monitoring. In NeuroVision, this is implemented through repeated checks against updated sources and the accumulation of compliance decisions (including whitelists and confirmed cases), which helps to more quickly distinguish real risks from matches “by name”.
What data and participants are checked within AML
An AML check covers all categories of participants in financial operations — from individuals to international companies and their counterparties. At the basic level, personal and registration data is analyzed (full name/company name, date of birth, citizenship, address, identifiers), which is then matched against external sources. In NeuroVision, this data can be obtained both from uploaded documents (via AI-OCR) and passed via an API — after which the platform performs screening against connected databases and returns a structured result with the details of the matches.
For legal entities, the check is significantly deeper: the entire ownership chain down to the ultimate beneficiaries is analyzed, and the company’s structure, types of activity, and financial statements are studied. Particular attention is paid to companies from offshore jurisdictions and organizations with a complex multi-level ownership structure, which are often used to conceal the real owners.
Transactional data is subject to continuous monitoring: transfer amounts, the frequency of operations, the geography of payments, the payment purpose, counterparties. The algorithms detect anomalies in payment behavior — a sharp change in volumes, atypical routes of the movement of funds, operations in high-risk jurisdictions.
Politically exposed persons (PEP) and their immediate circle are subject to checks: members of governments, deputies, judges, heads of state corporations, their relatives and business partners. Enhanced monitoring is applied to this category, with an analysis of income sources and business interests. The PEP database is regularly updated and includes officials from 240+ countries.
Documentary verification includes passports, companies’ constituent documents, licenses, statements, contracts. In NeuroVision, this stage is accelerated through a combination of computer vision + AI-OCR: the platform recognizes the document and extracts the key fields automatically (the article below gives an example of a recognition speed of 0.1 seconds), after which it immediately uses this data for screening. This reduces the share of manual entry and decreases the number of errors, and also helps to more quickly identify discrepancies in the customer’s documents and data.
Behavioral analysis tracks the user’s digital footprint: IP addresses, access devices, the time and frequency of logins, the geography of service use. A mismatch between the declared data and actual behavior serves as an indicator of potential fraud or the use of front persons to conduct operations.
AML and KYC: what the differences are
The terms AML and KYC are often used together, which causes confusion: where customer identification ends and ongoing compliance control begins. Distinguishing these concepts is important not only “in theory” but also for practical implementation: KYC is responsible for correct source data, while AML is responsible for risk control based on this data and external sources. In products like NeuroVision, these processes are conveniently assembled into a single scenario: KYC data, normalization, screening against databases, report/audit trail, compliance decision
What a KYC check is
KYC (Know Your Customer) is a procedure for identifying and verifying a customer’s identity when establishing business relations. At the core of KYC is the collection and confirmation of basic information about a person or organization: full name, date of birth, registered address, passport data, and INN for legal entities.
The procedure includes three mandatory stages. The first is identifying the customer through the provision of identity documents. The second is verifying the authenticity of these documents and the correspondence of the data to a real person. The third is assessing the customer’s risk based on the collected information.
Modern KYC systems use biometrics and computer vision: matching the face with the photo in the document, AI-OCR for the automatic reading of data, as well as document integrity checks (including the analysis of features and machine-readable zones). In NeuroVision, this approach is used as the “entry” into the compliance loop: data is extracted automatically, then immediately applied in sanctions and risk screening. As a result, KYC ceases to be a separate manual procedure and becomes part of fast onboarding — from seconds to a minute depending on the scenario and the set of checks.
KYC is mandatory when opening bank accounts, registering on cryptocurrency exchanges, taking out loans, buying securities, and making transfers above the established limits. The requirements for the depth of the check vary depending on the jurisdiction and the type of operations, but the basic principle remains unchanged — the organization must know exactly who it is dealing with.
The role of KYC within the broader AML system
KYC functions as the entry point into the comprehensive AML system, creating the foundation for all subsequent checks and monitoring. Without quality customer identification, it is impossible to track suspicious transactions, identify connections with sanctioned persons, or detect signs of money laundering.
In the architecture of an AML system, KYC plays the role of a primary filter. The data collected during identification becomes the basis for building the customer’s risk profile, which is then used to configure the transaction monitoring parameters. Customers with an elevated risk are subject to enhanced control, and their operations are checked against stricter criteria.
KYC data is integrated with the other components of AML: sanctions screening, transaction monitoring, verification of the sources of funds. For example, the customer’s country of residence established during KYC determines the applicable sanctions lists for the check. Their professional activity affects the threshold values for detecting unusual operations.
The regular updating of KYC information keeps the entire AML system current. A change in the customer’s status, a change of citizenship or place of residence can radically change the risk level and the control requirements. Therefore, KYC is not a one-time procedure at onboarding but a continuous process of keeping customer data up to date.
Key differences between AML and KYC for businesses and customers
| Category | Description |
|---|---|
| Business | For businesses, the differences between AML and KYC manifest in the scale of implementation and operating costs. KYC requires investment in identification and verification systems, integration with government databases, and training staff in document verification procedures. AML involves creating a full-fledged financial monitoring infrastructure with transaction analysis systems, a staff of compliance officers, and regular reporting to regulators. In terms of time frames, KYC happens at specific moments: when registering a customer, updating data, or conducting large operations. AML works continuously, analyzing every transaction, tracking changes in sanctions lists, and detecting atypical patterns of customer behavior. |
| Customer | For customers, KYC is experienced as the need to provide documents and pass verification when starting to work with a service. This is the visible part of the checks that every user encounters. AML remains largely invisible — the customer may be unaware of the constant monitoring of their operations until the triggers of suspicious activity fire. |
The legal consequences of violations also differ. A poor-quality KYC entails fines for violating identification procedures, usually of a fixed amount for each case. Failures in AML can lead to accusations of aiding money laundering, the revocation of licenses, and the criminal prosecution of a company’s management.
In technological terms, KYC indeed focuses on the accuracy of document recognition and biometric verification — processing speed, OCR quality and the reliability of liveness checks are critical. AML is broader: it adds sanctions/PEP screening, verification of connections, and risk assessment based on external sources and internal policies. NeuroVision closes the key “compliance loop” at the intersection of these tasks: AI-OCR + normalization + fuzzy matching + screening against 1,700+ databases + reports and an audit trail, which reduces the manual burden and speeds up decision-making.
The cost of implementation also differs substantially. Basic KYC solutions are accessible even to small companies through cloud solutions such as NeuroVision, with pay-per-check. A full-fledged AML system requires substantial investment in software, integration with numerous external data sources, and the creation of internal procedures and regulations.
Sanctions screening: how it works in an AML check
Sanctions screening is a critically important component of the anti-money-laundering system, aimed at identifying persons and organizations that are under international or national restrictions. This process protects financial institutions from reputational risks, multimillion-dollar fines and the loss of licenses by automatically checking every customer and transaction against current sanctions lists.
Unlike basic identity verification, sanctions screening focuses on matching the customer’s data against constantly updated databases of persons subject to economic restrictions. Modern AML systems conduct such a check in real time, analyzing not only exact matches of names and details but also similar spellings, transliterations, aliases and related persons.
Which sanctions lists and databases are used for screening
| Category | Description |
|---|---|
| International sanctions lists | International sanctions lists form the basis of screening in most financial systems. The US OFAC (Office of Foreign Assets Control) list contains more than 12,000 records of individuals and legal entities whose economic activity is restricted on the territory of the United States and in dollar transactions worldwide. The EU consolidated list combines the sanctions of all European Union member states and is updated almost weekly, covering about 13,000 subjects. |
| UN lists | UN lists include persons connected with terrorism, the proliferation of weapons of mass destruction, and violations of international law. The United Kingdom maintains its own UK Consolidated List after leaving the EU, containing more than 2,000 records. Switzerland, Canada, Australia, Japan and other developed countries maintain national sanctions registries, which often overlap with international ones but contain unique records. |
| The Rosfinmonitoring list | Russian financial organizations are required to work with the Rosfinmonitoring list, which includes extremists and terrorists under federal legislation. For full-fledged screening, systems also connect lists of politically exposed persons (PEP), Interpol databases of wanted criminals, registries of unreliable companies, and industry blacklists from the regulators of specific jurisdictions. |
Commercial aggregators of sanctions data, such as Dow Jones Risk & Compliance, Refinitiv World-Check and LexisNexis, combine information from hundreds of sources, providing a single point of access to global lists. These databases contain not only official sanctions but also adverse media mentions, court decisions, and information about beneficial owners and corporate connections.
The stages of sanctions screening: from data collection to decision
How systems reduce false positives in sanctions screening
Contextual data analysis significantly improves screening accuracy by taking additional information about the customer into account. The system matches not only the name but also related attributes: profession, place of work, transaction history, the geography of operations. If Ivan Petrov from the sanctions list is a politician from a certain region, while the customer being checked is a programmer from another city with a documented biography, the probability of a false positive is sharply reduced.
Machine learning on historical data allows systems to independently identify patterns of false matches. The algorithms analyze thousands of previous checks where the initial trigger turned out to be erroneous, and automatically adjust the threshold values for similar cases. Neural networks learn to distinguish real risks from technical matches, constantly improving the quality of filtering.
Multi-level identification uses biometric data for the final confirmation of identity. Matching the photograph from a passport with reference images of faces from sanctions lists through face recognition algorithms gives a verification accuracy of up to 99.9%. This practically rules out situations where a bona fide customer is blocked because of a name match with a sanctioned subject.
Intelligent exclusion filters automatically cut off knowingly irrelevant matches based on logical rules. The system ignores matches with deceased persons if the date of death in the sanctions list precedes the customer’s date of birth. Geographic filters exclude checks against sanctions not applicable in the operation’s jurisdiction — for example, unilateral sanctions of third countries that do not affect domestic transactions.
Dynamic sensitivity tuning allows organizations to independently regulate the balance between the level of protection and the number of false positives. For high-risk operations, strict check parameters are set, requiring a full match of multiple attributes. For routine internal payments, lighter criteria are applied, reducing the load on the compliance unit without a significant increase in risks.
Online sanctions checks: how they work in practice
Modern sanctions screening systems work in real time, processing thousands of requests per minute. Unlike a manual check, which could take hours or days, automated online screening delivers a result in fractions of a second. The NeuroVision platform checks a customer against 1,700+ databases in less than a second, including the international sanctions lists OFAC, UN, EU, as well as the Russian databases of the FTS, the MVD and Rosfinmonitoring.
Sanctions screening is built into the overall AML check process and works in parallel with other control mechanisms. The system analyzes not only direct matches but also related persons, beneficiaries, affiliated companies. This is critically important, since sanctions often extend to whole groups of companies and individuals connected with the main subject.
What data is needed for an online sanctions list check
The minimum set of data for an individual includes the full name (in the original language and in transliteration), the date of birth and the country of citizenship. To improve accuracy, the system requests additional identifiers: passport number, INN, registered address. The more attributes provided, the more accurately the algorithm cuts off false positives.
For legal entities, the mandatory fields are the full name of the organization, the registration number (OGRN, INN for Russian companies or their equivalents for foreign ones), the country of registration. The system also analyzes data on managers, founders and ultimate beneficiaries with an ownership stake of 25% or more. Modern platforms automatically pull related information from public registries and databases.
Checking transliterated names presents a particular challenge. The same name can be written in dozens of ways in different documents and databases. NeuroVision’s algorithms use phonetic matching algorithms and machine learning for the correct matching of names in various spellings, achieving a recognition accuracy of 99.74%.
A step-by-step scenario for the online check of a customer or counterparty
The process begins with uploading the customer’s documents through a web interface or an API. The system recognizes the document in 0.1 seconds, automatically extracting all the necessary data. NeuroVision’s technology correctly processes documents from 200+ countries in 90 languages, including complex cases with damaged or poorly scanned documents.
After the data is extracted, a parallel search across all connected databases is launched. The algorithm checks for exact matches, then expands the search to similar records, taking into account possible typos, alternative spellings, abbreviations. The system analyzes not only current sanctions lists but also historical data — a person may have been under sanctions earlier or connected with sanctioned persons.
The check results are compiled into a structured report indicating the data sources, the match percentage, and the type of sanctions (blocking, sectoral, secondary). When a match is found, the system shows detailed information: the grounds for inclusion in the list, the date of entry, the term of the sanctions, and the related persons and organizations.
The entire process, from uploading the document to receiving the full report, takes 1-3 seconds. NeuroVision retains a full audit trail of the check for subsequent compliance control and regulatory reporting.
What happens if sanctions screening fires
When a match is found, the system immediately blocks the operation and sends a notification to the compliance officer. Automatic blocking prevents the accidental execution of a prohibited transaction, which is critically important for avoiding fines and reputational losses. The algorithm assigns each trigger a risk level: high for a full match, medium for a partial one, low when only individual attributes match.
The compliance specialist receives a detailed card with the check results, showing all the matches found, their sources and the degree of reliability. The platform provides tools for in-depth analysis: checking additional attributes, matching photographs through biometric algorithms, analyzing connections through an interaction graph.
If the check confirms that the customer is indeed under sanctions, the system automatically generates a report for the regulator in accordance with the requirements of the jurisdiction. Under Russian law, the organization is obliged to send a message to Rosfinmonitoring within three business days. In parallel, all of the customer’s accounts and operations are blocked, and the asset freeze procedure is activated.
In the event of a false positive, the compliance officer adds the customer to a whitelist with a justification of the decision. Machine learning analyzes such cases and adjusts the algorithms, reducing the number of false positives in the future. NeuroVision statistics show a 40% reduction in false positives after three months of the system’s operation thanks to the accumulation of data on the specifics of a particular business.
AML checks and sanctions screening solve a critical task — they protect a company from financial crimes, reputational losses and regulatory sanctions. KYC verification acts as the first line of defense, forming the correct customer identifiers, while the sanctions/AML loop checks the customer, counterparties and related structures against external sources and risk criteria.
In practice, the maximum effect comes from automation: when KYC, data normalization, fuzzy matching and screening are “stitched” into a single process, compliance stops holding back growth. NeuroVision makes this loop practical: a check against 1,700+ databases takes less than 1 second, data is extracted from documents via AI-OCR, transliterations and spelling variants are supported, and compliance decisions accumulate and help reduce false positives (the article above gives an example of -40% false positives after three months of operation). As a result, the business gets both compliance and fast onboarding without a multiple increase in the costs of the compliance team.
