Money laundering

Peer-group outlier (behaviour unlike similar customers)

A customer moves far more money, or makes far more transactions, in a month than other customers on the same product with a similar profile, even if that is normal for the customer.

One customer moves far more money, or makes far more transactions, in a month than similar customers on the same product Money or transactions in a month This customer Similar customers, same product

What it catches

A customer whose activity is far out of line with customers like them: same product, same kind of customer, same declared occupation or line of business, same region. A student whose wallet moves as much each month as a small shop’s, a personal account that receives as many payments as a merchant’s, a borrower who repays far more in a month than others on the same loan product. The comparison is with other customers, not with the customer’s own past, so it catches customers who have looked like this from their first full month: an account opened to be used as a mule or a front, a business run through a personal account without the institution knowing, a declared occupation that does not explain the money. Accounts used for only a few weeks and then dropped are never judged over a full month; New account activity watches those.

Regulators and standard setters name the comparison directly. Mexico’s provisions for fintech institutions list, among the circumstances to consider when deciding whether an operation is unusual, “comportamiento diferente o discrepante en comparación con otros Clientes que celebren Operaciones similares o tengan un perfil semejante” (behaviour that differs from other clients who carry out similar operations or have a similar profile). The EBA says transactions may be unusual when they have “an unusual or unexpected pattern compared with the customer’s normal activity or the pattern of transactions associated with similar customers, products or services”. Nigeria’s central bank requires automated AML solutions to “Support related-party mapping, network analysis and peer-grouping to improve identification of unusual or suspicious patterns”. The JMLSG’s guidance for the UK financial sector says monitoring may be done “by comparing their activity or profile with that of a similar, peer group of customers”.

The rule finds an outlier, not a crime. In every group someone is at the top, and most of them are simply the biggest honest users. What makes the rule work is a peer group that really holds similar customers.

How the rule works

Once a period, put customers into peer groups by attributes they share: the product, the type of customer (individual or business), the declared occupation or line of business, the region, and the risk grade where you hold one. For each group, work out the spread of the period’s totals across its active customers: those with at least one completed transaction, in the direction the version watches, during the period. Alert on a customer whose total is above a high percentile of their group and above an absolute minimum, provided the group is large enough for the percentile to mean something. Run value and count versions separately, and money in and money out separately. A customer is first judged at the end of their first full period.

Totals of money are skewed: a few customers move far more than all the rest. A line at the mean plus a few standard deviations therefore flags a different share of customers than its maths suggests; a percentile line does not have that problem.

Groups

Customers into peer groups by product, type of customer, declared occupation or line of business, region and risk grade, using only as many of these as keeps each group large enough.

Measures

Each customer’s completed transactions in the period, in the direction this version watches (money in or money out), as a total value in one currency or as a number of transactions.

Compares

Each customer’s total with the totals of the other active customers in their peer group for the same period: those with at least one completed transaction in this direction.

Leaves out

  • Customers who were not open for the whole period. A new customer is first judged at the end of their first full period.
  • Peer groups below the minimum size: compare those customers with the next broader group instead.
  • Transactions that did not complete, fees, reversals and moves between the customer’s own accounts with you. Transfers to the customer’s own account at another institution stay in.

Alerts when

Both of these are true:

  1. The customer’s total is above the 99th percentile of their peer group.
  2. The customer’s total is at least your minimum for the segment.

How often

At most one alert per customer per period for each version.

The alert shows

The customer’s total, the peer group’s definition and size, and the group’s median and 99th percentile for the period.

Highlighted values are settings you choose. Hover over one, tap it or tab to it for a suggested value.

Common variants:

  • Mix, not size. Compare the share of the customer’s activity that goes to one kind of transaction (cash, cross-border, crypto, agent cash-out) with the share for the peer group. A customer can be ordinary in total and still an outlier in what the money does. For cash and the cash share, see Cash use that does not fit the customer’s activity.
  • Peer and self together. Alert only when the customer is above the peer line and also above their own usual level (see Sudden change from a customer’s baseline). Fewer alerts, but it drops the customer who has been out of line from the start, which is the case this rule is for.
  • Stand-in for new customers. For customers too new to have their own history, compare each full period with the peer group until they have one. Before their first full period, New account activity watches them.
  • Groups learned from behaviour. Instead of fixed attributes, group customers by how they actually transact and refresh the groups regularly. The Wolfsberg Group notes that traditional segments built on static factors such as customer type, industry, type of business and inherent customer risk “do not capture common transaction activity among the segmented entities consistently and rely mainly on outdated information”, and that “A combination of known attributes and a dynamic statistical clustering could lead to more meaningful segmentation”.

Data it needs

ShowHide
FieldUsually comes fromIf it’s missing
Customer idaccount or walletRequired. One id per person, or one person’s activity is split across several small totals.
Productcore systemRequired for the peer key. Without it, customers of a payroll wallet and a merchant wallet are compared with each other.
Customer type (individual, business, agent)onboarding (KYC/KYB)Required. Mixing individuals and businesses puts ordinary businesses at the top of every group.
Declared occupation or line of businessonboarding (KYC/KYB)Needed for the occupation key. Often blank, free text or “other”: map it to a short list first, and keep customers with no value in their own group.
Regionaddress or KYCOptional key.
Risk gradethe institution’s customer risk assessmentOptional key.
Account opening datecore systemNeeded to leave out customers who were not open for the whole period.
Amount and currencyledgerRequired. Convert to one currency, or group by currency or market.
Timestamp, status, direction, transaction typeledgerRequired. Count completed transactions only, and separate money in from money out; without transaction type, fees and moves between the customer’s own accounts with you count as activity.

Thresholds

No regulator sets a percentile, a period or a group size. Mexico’s provisions make behaviour unlike similar clients a circumstance to consider, and Nigeria’s baseline standards require the AML solution to support peer-grouping; the values below are starting values.

ParameterStarting pointWhere it comes from
PeriodOne calendar monthA starting value. Smooths day-to-day noise; every customer in the group is compared over the same days. Shorter for products used daily, if groups stay large.
LineAbove the 99th percentile of the peer groupA starting value. Totals of money are skewed, so a line at the mean plus a few standard deviations flags a different share than its maths suggests; a percentile does not. The line sets the alert volume: about 1% of each group’s active customers per version per period. Raise it for fewer alerts, lower it to reach further down. Test on your own history.
Minimum group sizeEnough active customers that the top 1% is more than a handfulArithmetic: in a group of 100, only the largest customer is above the 99th percentile; with about 500, about five are. Below the minimum, use the next broader group.
Minimum totalSet per segment and currency from your own data; for count versions, a minimum value as well as a countNo public source gives a value. Keeps groups of very small users from alerting on small amounts, and count versions from alerting on many tiny payments.

By country

  • Mexico. Fintech institutions (crowdfunding and e-money institutions, ITFs), when deciding which operations are unusual, must consider, among other circumstances and whether they arise alone or together, “Si existe comportamiento diferente o discrepante en comparación con otros Clientes que celebren Operaciones similares o tengan un perfil semejante” (whether there is different or discrepant behaviour compared with other clients who carry out similar operations or have a similar profile) (SHCP provisions under art. 58 of the Fintech Law, art. 70(XIII)). The same article lists the client’s characteristics, among them the risk grade and “principal ocupación, profesión, actividad, giro del negocio u objeto social” (main occupation, profession, activity, line of business or corporate purpose) (art. 70(I)), and “Los usos y prácticas mercantiles de las Operaciones en el sector” (the commercial uses and practices of operations in the sector) (art. 70(V)). Their automated systems must classify the types of operations, products or services they offer “a fin de detectar posibles Operaciones Inusuales” (in order to detect possible unusual operations) (art. 56(V)). No threshold is set.

    Reporting deadline

    The Communication and Control Committee rules on an unusual operation within 60 calendar days of its detection, and the report goes to SHCP, through the CNBV, within 3 business days of the end of that session; within 24 hours of the alert or staff notice where there are signs or concrete facts that the funds could come from illicit activities or be used for terrorist financing or money laundering (SHCP provisions under art. 58 of the Fintech Law, arts. 69 and 73).

  • Nigeria. The CBN’s Baseline Standards for Automated AML Solutions (10 March 2026), which apply to banks and other financial institutions under the CBN’s purview, require the AML solution to “Support related-party mapping, network analysis and peer-grouping to improve identification of unusual or suspicious patterns” and, where AI or machine learning is used, to give the key factors behind each alert “(like unusual transaction patterns or behaviour compared to similar customers)” (section 5.5(a)(v)). Monitoring must use “customer segmentation” (section 5.5(a)(iii)) and make risk-based use of KYC and KYB attributes “such as occupation, declared income or turnover range, customer segment, product profile, geography, and delivery channel” (section 5.5(a)(iv)). The standards define transaction monitoring as “Surveillance of transactions using scenarios, thresholds, and peer grouping to detect patterns” (section 8, Glossary). Full compliance is due 18 months from issue for deposit money banks and 24 months for other financial institutions (issuing circular BSD/DIR/PUB/LAB/019/002).

    Reporting deadline

    Report to the NFIU immediately, and within 24 hours after the transaction draw up a written report and report the transaction and the action taken, whether the transaction is complete or not (Money Laundering (Prevention and Prohibition) Act 2022, s.7(1)–(3)). The NFIU’s December 2024 STR guidelines read the 24 hours as running from when the transaction is found suspicious, after an examination of no more than 72 hours, and require written reasons for every alert closed as not suspicious (STR-NFIU-2024-A0001, Part Six, para 2).

  • South Africa. FIC Guidance Note 4B lists among its indicators of unusual business “Transactions do not appear to be in keeping with normal industry practices” (para 42). Where the rule runs in an automated transaction monitoring system, FIC Directive 5 of 2019 requires the system’s detection methodology, “including algorithms, scenarios, threshold settings or rules” (this rule among them), to be documented (para 3.1.16) and, taken as a whole, “sufficient to address the associated money laundering and terrorist financing risks applicable to the reporter” (para 3.1.17), and sets the time limits for the system’s alerts given under the reporting deadline below (paras 2.1–2.3).

    Reporting deadline

    As soon as possible, and no later than 15 days excluding weekends and public holidays (FIC Act s.29; MLTF Control Regulations, reg. 24(3), via FIC Guidance Note 4B para 71). The 15 days run from when you became aware of the facts, which “may be, and in the majority of cases will be, before a suspicion is formed” (GN4B para 80), and an institution “may not add additional timeframes for its internal transactional monitoring alert system processes, and / or internal investigation and review processes to the prescribed reporting period” (GN4B para 81). Where the alert comes from an automated transaction monitoring system, the reporter “is deemed to have knowledge” of the possible suspicious activity when the alert is generated: the report is due “no later than 15 days after the alert has been generated”, and work on the alert must start within 48 hours, which the FIC counts as two days without weekends and public holidays, inside the 15 (FIC Directive 5 of 2019, paras 2.1–2.3; FIC PCC 45, paras 5–8).

  • Canada. FINTRAC’s indicators for financial entities, among the indicators surrounding transactions in relation to the person’s or entity’s profile, include “The volume of transactional activity exceeds the norm for geographical area”, and give as an example of activity out of line with a profile “an entity involved in an industry that is not normally cash-intensive receiving excessive cash deposits”.

    Reporting deadline

    Submit the report “as soon as practicable after you have completed measures that enable you to establish that there are reasonable grounds to suspect” (FINTRAC STR guidance, section 4).

  • EU. The EBA’s risk factor guidelines say transactions may be unusual because “they are larger than what the firm would normally expect based on its knowledge of the customer, the business relationship or the category to which the customer belongs”, because “they have an unusual or unexpected pattern compared with the customer’s normal activity or the pattern of transactions associated with similar customers, products or services”, or because they are very complex compared with similar transactions of similar customer types, products or services without a known economic rationale. A firm that detects such transactions must apply enhanced due diligence (EBA/GL/2021/02, guideline 4.60), including finding out the background and purpose of the transactions and monitoring the relationship more closely (4.61).

  • United Kingdom. The JMLSG’s guidance for the UK financial sector says “Monitoring may be by reference to specific types of transactions, to the profile of the customer, to networks of connected persons, or by comparing their activity or profile with that of a similar, peer group of customers, or through a combination of these approaches” (Part I, para 5.7.5), and lists among the transaction characteristics effective monitoring is likely to be based on “the unusual nature of a transaction: e.g., abnormal size or frequency for that customer or peer group” (para 5.7.10).

  • Argentina. For payment service providers and non-financial lenders, operations or transactions “que por su magnitud, habitualidad o periodicidad excedan las prácticas usuales” (that by their size, frequency or regularity exceed usual practice) are among the circumstances to weigh when setting alerts (UIF Resolution 200/2024, art. 34(b), circumstance 2). The resolution does not say whose practice; its definition of unusual operations measures departures in frequency, regularity and amount against “los usos y costumbres en las prácticas de mercado” (market custom and practice) (art. 2(n)). A peer group is one way to measure what is usual.

    Reporting deadline

    Money laundering, within 24 hours of concluding that the operation is suspicious, and no later than 90 calendar days from the operation, performed or attempted; terrorist financing and proliferation financing, within 24 hours of the operation (UIF Resolution 200/2024, art. 36(c)).

  • Ghana.

    Reporting deadline

    Within 24 hours of the suspicion, attempted transactions included, whatever the amount (BoG/FIC AML/CFT/CPF Guideline 2025, 2.8.1(6) and 2.8.3(2)).

  • Kenya.

    Reporting deadline

    Within two days after the suspicion arose (POCAMLA s.44(2), as replaced by the AML/CFT Laws (Amendment) Act 2023, Act No. 10 of 2023).

False positives and tuning

ShowHide

The rule is as good as its peer groups. Typical causes of misfires:

  • Mixed groups. A group that mixes individuals with businesses, or a product used by both consumers and merchants, puts the ordinary businesses at the top every period. Split by customer type first.
  • The line sets the volume. A line at the 99th percentile flags about 1% of each group’s active customers in every version, every period, whatever happens in the group: the alert count is a setting, not a measure of risk. The versions add up, overlapping only in part, so value and count, in and out, together flag more than 1% of customers. Size the line and the number of versions to what your team can review.
  • Groups too small. In a group of a few dozen customers, the top percentile holds one customer or none, depending on how the percentile is computed, and the same one every month. Set a minimum size and fall back to a broader group.
  • A fallback that makes a mixed group. Falling back to a broader group can itself mix customers who do not belong together: a small group of customers who earn more than most lands at the top of the broader group. Keep a separate line for such a group, even a rough one, rather than pooling it.
  • Count versions on small payments. Where money comes or goes in small pieces, such as mobile-money top-ups or small payouts to family, the customers with the most transactions often move little money. Give the count versions a minimum value as well.
  • Wrong or stale keys. Occupation and line of business are usually declared once, at onboarding, and left blank, as free text or as “other”. A customer filed under the wrong occupation is compared with the wrong peers. Map free text to a short list, give customers with no value their own group, and refresh the keys at each KYC review.
  • A key that stands for something else. A field filled in only in some markets or on some channels splits customers by market or channel, not by behaviour. Check each key’s fill rate by market and channel before using it.
  • Customers who are big by design. Resellers, agents, payroll accounts, and sole traders whom the institution allows to trade through a personal account and whose trade it has recorded, sit at the top of an individuals’ group by nature. Once reviewed and the activity documented, move them into a group that fits them, with its own line, rather than whitelisting them: a whitelisted account is no longer watched at all. A business run through a personal account that the institution does not know about, or does not allow, is what the rule is for.
  • Own accounts elsewhere. Leave out moves between the customer’s own accounts with you, not transfers to the customer’s own account at another institution: money collected from others and moved quickly to the customer’s account elsewhere is what the rule should see.
  • Mixed currencies. Totals summed in different currencies make customers who use a stronger currency look large. Convert to one currency, or group by currency or market.
  • Partial periods. A customer who joined or left mid-month looks small, and a customer who joined at the start of a busy month looks large against a group of half-month customers. Leave out customers not open for the whole period, and judge a new customer from their first full period.
  • A crowd that moves together. If many customers in a group are abused at once, for example mule accounts recruited onto one product, the group’s percentile rises and hides them. Keep fixed-threshold and pattern rules running beside this one, and watch the group’s own percentile over time.
  • The same customer every period. A customer reviewed and cleared will be at the top again next month. Record the reason at review, move them to a fitting group, and re-alert only if their total moves well beyond what was cleared.

How to tune: start with few keys (product and customer type), add occupation and region only where groups stay large, set the line per group from your own history, keep the minimum total per segment and currency, and review the groups whenever products or onboarding fields change.

When it fires

ShowHide
  1. Check the peer group: how it is defined, how many customers are in it, and whether this customer belongs in it (is the declared occupation or business type current and right?).
  2. Look at the customer’s own history. A customer who has always been far above the group points to a profile that never fitted; a customer who only recently rose is also a sudden change (see Sudden change from a customer’s baseline).
  3. Look at what drives the total: many senders, one large counterparty, cash, cross-border payments, crypto, transfers to the customer’s own accounts elsewhere.
  4. Compare with what the customer declared: occupation, income or turnover, expected activity. Ask for the source of funds and the purpose of the activity when it is material and the profile does not explain it.
  5. If the activity is explained, record why, and update the customer’s profile or move them to a peer group that fits. If it is not, escalate. Each country’s reporting deadline is under By country.

An alert is a reason to review, not a suspicion in itself. Mexico’s provisions list behaviour unlike similar clients as one of the circumstances to consider when deciding whether an operation is unusual (art. 70), and an unusual operation goes to the Communication and Control Committee, which rules on it before it is reported (art. 69), except in the 24-hour cases of art. 73.

Sources

15 sourcesHide
  1. Secretaría de Hacienda y Crédito Público (Mexico), “Disposiciones de carácter general a que se refiere el Artículo 58 de la Ley para Regular las Instituciones de Tecnología Financiera”, DOF 10 September 2018, arts. 56(V), 70(I), 70(V) and 70(XIII); arts. 69 and 73 (report deadlines). https://dof.gob.mx/nota_detalle.php?codigo=5537449&fecha=10/09/2018 (accessed 2026-11-12).
  2. CBN, Baseline Standards for Automated Anti-Money Laundering, Combating the Financing of Terrorism and Countering Proliferation Financing Solutions, circular BSD/DIR/PUB/LAB/019/002, 10 March 2026, sections 5.5(a)(iii)–(v) and 8 (Glossary, “Transaction Monitoring”); the circular (full-compliance periods). https://www.cbn.gov.ng/Out/2026/CCD/CBN%20issues%20Baseline%20Standards%20for%20Automated%20Anti-Money%20Laundering%20Solution.pdf
  3. Money Laundering (Prevention and Prohibition) Act 2022 (Nigeria), s.7(1)–(3). Listed at https://www.nfiu.gov.ng/LawsAndRegulations
  4. NFIU, Guidelines for the Identification, Verification and Reporting of Suspicious Transactions … for Financial Institutions, ref. STR-NFIU-2024-A0001, 13 December 2024, Part Six para 2. Listed at https://www.nfiu.gov.ng/AdvisoryAndGuidance
  5. EBA, Guidelines on ML/TF risk factors (EBA/GL/2021/02, consolidated version amended by EBA/GL/2023/03), guidelines 4.60 and 4.61. https://eba.europa.eu/sites/default/files/2023-11/EBA%20GL%202021%2002%20-%20consolidated%20%28amended%20by%20EBA%20GL%202023%2003%29_EN.pdf
  6. JMLSG, Prevention of money laundering/combating terrorist financing: Guidance for the UK financial sector, Part I, October 2026 revised version, paras 5.7.5 and 5.7.10. https://www.jmlsg.org.uk/wp-content/uploads/2026/10/JMLSG-Guidance-Part-I_Oct-2026.pdf
  7. FINTRAC, Money laundering and terrorist financing indicators—Financial entities, modified 2025-11-24, “ML/TF indicators surrounding the financial transactions in relation to the person/entity profile”. https://fintrac-canafe.canada.ca/guidance-directives/transaction-operation/indicators-indicateurs/fin_mltf-eng
  8. FINTRAC, Reporting suspicious transactions to FINTRAC, section 4. https://fintrac-canafe.canada.ca/guidance-directives/transaction-operation/str-dod/str-dod-eng
  9. FIC South Africa, Guidance Note 4B on reporting of suspicious and unusual transactions and activities, 26 March 2019, paras 42, 71, 80 and 81. https://www.fic.gov.za/wp-content/uploads/2023/09/2019.03-Guidance-Guidance-Note-4B-STRs.pdf
  10. FIC South Africa, Directive 5 of 2019 on the usage of an automated transaction monitoring system for the detection and submission of regulatory reports, 29 March 2019, paras 2.1–2.3, 3.1.16 and 3.1.17. https://www.fic.gov.za/wp-content/uploads/2023/09/2019-DIR-Directive-5-of-2019-ATMS.pdf
  11. FIC South Africa, Public Compliance Communication 45 on Directive 5 of 2019 in relation to the use of automated transaction monitoring systems, 20 March 2020, paras 5–8. https://www.fic.gov.za/wp-content/uploads/2023/09/2020.03-PCC-PCC-45-On-Directive-5-of-2019-in-relation-to-the-use-of-ATMS.pdf
  12. Unidad de Información Financiera (Argentina), Resolución 200/2024 (RESOL-2024-200-APN-UIF#MJ), 18 December 2024, Boletín Oficial no. 35571, 19 December 2024, art. 2(n); art. 34(b), circumstance 2; art. 36(c) (report deadlines). Text quoted from argentina.gob.ar (content licensed CC BY 4.0): https://www.argentina.gob.ar/normativa/nacional/norma-407429/texto
  13. Bank of Ghana and Financial Intelligence Centre (Ghana), AML/CFT/CPF Guideline for Accountable Institutions, September 2025, published by the Bank of Ghana on 12 December 2025, 2.8.1(6) and 2.8.3(2). https://www.bog.gov.gh/wp-content/uploads/2025/12/BOG-FIC-AML-GUIDELINE-2025-FINAL-5-TO-BE-PUBLISHED.pdf
  14. Kenya, Anti-Money Laundering and Combating of Terrorism Financing Laws (Amendment) Act, 2023 (Act No. 10 of 2023), replacing POCAMLA s.44(2). https://brs.go.ke/wp-content/uploads/2023/09/AntiMoneyLaunderingandCombatingofTerrorismFinancingLawsAmendmentAct2023.pdf
  15. The Wolfsberg Group, Statement on Effective Monitoring for Suspicious Activity, Part I: Moving Beyond Automated Transaction Monitoring, 2024, p.5–6 (segmentation). https://db.wolfsberg-group.org/assets/e3d83d2f-fad9-46d2-b5a9-3cf4e932f53f/Wolfsberg%20Group%20Statement%20on%20Effective%20Monitoring%20for%20Suspicious%20Activity.pdf

Ask us

A question about this rule?

A person at Archer reads every question and answers by email.

The Fraud & AML Library is a practical reference, not legal advice. What binds is the regulator’s own text, which every rule and regulation page links.

Get new rules and regulation changes by email

When we add a rule or a regulation changes, we’ll email you.

You can unsubscribe at any time. Privacy policy