An examiner who opens a closed alert asks the same question every time: who decided this was not suspicious, on what evidence, and where is that written down? The rules that fire, the models that score, the agents that draft a narrative and the transaction monitoring software that holds the case together are all judged by the record they leave.
What is transaction monitoring in AML?
Anti-money laundering (AML) transaction monitoring is the ongoing review of a customer's deposits, withdrawals, payments and transfers against what the firm knows about that customer, so that activity which does not fit can be examined and, where the suspicion holds, reported. It starts after onboarding and runs for the life of the relationship.
It is not screening, which matches a name or a payment party against sanctions and watchlists before money moves, and it is not fraud detection, which decides in seconds whether a payment goes through. Monitoring reads the money. It asks whether the money matches the story the customer told.
Banks run it across the most products (accounts, wires, cards, cash, trade finance) and keep the alert, the system's output, apart from the case, the analyst's work. Fintechs and payments firms inherit the same shape with fewer products and far more transactions per customer.
In banking as a service the sponsor bank answers for every BSA check a fintech runs, monitoring among them, and the guide to BaaS compliance shows who owns each check.
Picture a neobank customer whose profile says salaried employee, with about 2,400 dollars arriving on the first of each month. In week three the account receives 14 transfers from 14 individuals totaling 31,000 dollars, and within 48 hours the balance goes to one account abroad. No single transaction is large and no name matched a list. Suspicious transaction monitoring exists for this pattern: it compares the flow with the profile and raises an alert that an analyst must explain or close.
Video page: Transaction monitoring explained in 21 seconds | AML Explained
How does the transaction monitoring process work, from data to alert to SAR?
The transaction monitoring process has six stages. Read the last column first, because it is what the examiner reads.
| Stage | Input | Output | Who decides | What the record keeps |
|---|---|---|---|---|
| 1. Data | Transactions, customer profile, know your customer (KYC) risk rating, counterparties | Normalized events on one customer record | Nobody; the data owner signs off the field mapping | Source system, timestamp, completeness checks, missing fields |
| 2. Detection | Rules, behavioral models, the customer's own baseline | An alert naming the rule and the signals matched | The rules engine fires; a named person owns each rule | Rule name and version, threshold, window, matched transactions |
| 3. Triage | The alert, the customer's history, prior alerts | Close with a reason, or open a case | The analyst; an agent for routine cases, with the analyst's sign off | Who, when, reason code, evidence viewed, time to decision |
| 4. Investigation | The case, related alerts, counterparties, documents | A decision to file or not to file | The investigator and the MLRO (money laundering reporting officer) | Timeline, documents requested and received, rationale |
| 5. Reporting | The SAR narrative and supporting transactions | A filing within 30 calendar days of detection (31 CFR 1020.320(b)(3)) | The MLRO signs | Filing date, confirmation, supporting documentation kept five years |
| 6. Feedback | SAR outcomes, closed alerts, false positive counts | Thresholds changed, rules retired or added | The compliance committee | Tuning log, counts before and after, approval, date |
A rule that reads a missing field as zero never fires, and the record cannot show the gap unless ingestion logged it, so the data stage sets the quality of everything after it. Triage and investigation are where the examiner spends time; an alert closed with "reviewed, no concern" and nothing else is a finding in itself.
Reporting has a clock. Under 31 CFR 1020.320(b)(3) a US bank files "no later than 30 calendar days after the date of initial detection", up to 60 days if no suspect was identified (eCFR, current to 6 October 2026). Feedback is the stage teams skip, and it is where a program stops paying for the same false positives twice.
What do rules, behavioral models and agents each do, and who decides?
Automated transaction monitoring has three layers, and an examiner wants to know what each one did to a given alert.
Rules are written conditions: an amount, a count, a window, a segment. A reviewer can repeat the logic in a sentence, and rules catch the money laundering typologies regulators have described. They fire on numbers. A threshold set for merchants drowns a retail segment in alerts unless the rule reads the segment.
Behavioral models score deviation from a baseline built from the customer's own history and a peer group. A model should name the signals that moved the score, the way a rule names its condition, or the analyst who closed the alert cannot say why.
Agents do the work around the alert. In complead Fusion, the engine of the AI native platform, an agent gathers the evidence around an alert, drafts the narrative in the examiner's format and, for routine cases, proposes a closure with the work log attached, so the analyst sees only the cases that need judgment. A named person decides at every step: the agent proposes, the analyst closes, the MLRO signs the filing, and the record shows which of them did what. Explainable AI, to an examiner, is that trace from output to signals to a person.
Run the neobank alert through the three layers. The funnel rule fires on 14 senders in seven days. The model adds that the outbound amount is 13 times the 90 day average. The agent pulls the 15 transactions, notes that none of the senders is a known counterparty, and drafts two paragraphs. The analyst reads them and escalates. The examiner later sees four steps, in order, with names.
Who must monitor transactions, and what do the rules say?
The obligation is written in three places for the three markets most readers work in, each quoted with the article number and the date it was read.
| Jurisdiction | Rule | What it requires (quoted) | Who is covered |
|---|---|---|---|
| United States | 31 CFR 1020.210(a)(2)(v), eCFR current to 6 October 2026 | An AML program includes "appropriate risk-based procedures for conducting ongoing customer due diligence", including "(B) Conducting ongoing monitoring to identify and report suspicious transactions and, on a risk basis, to maintain and update customer information." | Banks under the Bank Secrecy Act; parallel rules in 31 CFR chapter X cover money services businesses, broker dealers and other financial institutions |
| United Kingdom | Money Laundering Regulations 2017, regulation 28(11), current version read 8 October 2026 | "The relevant person must conduct ongoing monitoring of a business relationship, including (a) scrutiny of transactions undertaken throughout the course of the relationship (including, where necessary, the source of funds) to ensure that the transactions are consistent with the relevant person's knowledge of the customer, the customer's business and risk profile" | Relevant persons under regulation 8, from credit and financial institutions to crypto asset businesses |
| European Union | Regulation (EU) 2024/1624, Article 26(1), adopted 31 May 2024, applies from 10 July 2027 | "Obliged entities shall conduct ongoing monitoring of business relationships, including transactions undertaken by the customer throughout the course of a business relationship, to ensure that those transactions are consistent with the obliged entity's knowledge of the customer, the customer's business activity and risk profile" | Obliged entities under Article 3, one list for every member state once the Regulation applies |
The test in all three is fit with what the firm knows about the customer, whatever the amount. "On a risk basis" sets the depth, and the UK's regulation 28(12) says how: the extent of the measures "must reflect (i) the risk assessment carried out by the relevant person under regulation 18(1); (ii) its assessment of the level of risk arising in any particular case". That is risk based transaction monitoring, with the depth set by the firm's own written assessment.
The reporting floor differs. A US bank files under 31 CFR 1020.320(a)(2) when a transaction "involves or aggregates at least $5,000 in funds or other assets" and the bank "knows, suspects, or has reason to suspect" that it "has no business or apparent lawful purpose or is not the sort in which the particular customer would normally be expected to engage". Article 69(1)(a) of the EU Regulation requires a report "regardless of the amount involved", so a rule set built for one market needs a different floor in the other. The whitepapers on anti-money laundering in the United States and anti-money laundering in the United Kingdom cover the legislation and the authorities around these rules.
Until 10 July 2027 the EU duty comes from the AML directives (AMLD), written into each member state's own law. Article 13(1)(d) of Directive (EU) 2015/849 lists among the customer due diligence (CDD) measures "conducting ongoing monitoring of the business relationship including scrutiny of transactions undertaken throughout the course of that relationship".
The duty reaches beyond banks. Regulation 8 of the UK regulations also lists auditors, insolvency practitioners, external accountants and tax advisers, independent legal professionals, trust or company service providers, estate and letting agents, high value dealers, casinos and art market participants, the group usually called designated nonfinancial businesses and professions (DNFBPs). Regulation 7 names their supervisors: the Financial Conduct Authority (FCA) for authorized credit institutions and financial institutions, HM Revenue and Customs for high value dealers and the Gambling Commission for casinos.
In the UK a money service business registers with HMRC before it trades; the guide to UK MSB registration, fees and rules covers the fit and proper test and the FCA's part.
Other markets write the same test into their own rules. In Singapore, paragraph 6.19 of MAS Notice 626 from the Monetary Authority of Singapore (MAS) says "A bank shall monitor on an ongoing basis, its business relations with customers." In the UAE, Dubai's Virtual Assets Regulatory Authority (VARA) requires virtual asset service providers to "continuously monitor business relationships with clients to identify any Suspicious Transactions" and report them to the UAE FIU (Compliance and Risk Management Rulebook, Rule III.F).
Which rules does a fintech run first, with what thresholds?
These are the ten rules a fintech or payments firm typically switches on first, written as transaction monitoring examples with a threshold each. Two numbers are sourced: a currency transaction report is due on a transaction in currency "of more than $10,000" (31 CFR 1010.311, eCFR current to 6 October 2026), and the SAR aggregation line is "at least $5,000" (31 CFR 1020.320(a)(2)). Every other threshold is illustrative. Each is a starting point the firm's own data must confirm or move. The transaction monitoring rules library carries 25 with their logic, data needs and tuning method.
| # | Rule | Logic | Illustrative threshold | False positive note |
|---|---|---|---|---|
| 1 | Structuring under the reporting line | Cash deposits each below the CTR line that together cross it | 3 or more cash deposits of $9,000 to $9,999 in 7 days | Cash heavy retailers deposit daily; read the declared business |
| 2 | Rapid movement of funds | Money leaves before it could have been used | 80% or more of inbound value out within 48 hours, inbound above $5,000 | Payroll, treasury and settlement accounts do this by design |
| 3 | Velocity spike | Count far above the customer's own baseline | More than 20 transactions in 24 hours against a daily average under 5 | Sale days for sellers; read the segment and the calendar |
| 4 | Profile deviation | One transaction out of scale with history | 3 times the 90 day average and above $5,000 | Seasonal businesses, house purchases; whitelist a documented single event |
| 5 | High risk geography | Value to or from a high risk jurisdiction | Any transfer above $1,000 to or from a country on the firm's high risk list | Family remittance corridors; pair with the customer's declared ties |
| 6 | Dormant account reactivation | A quiet account suddenly moves money | No activity for 180 days, then more than $2,000 within 7 days | Savings after a sale or a tax refund; read the source of funds |
| 7 | Round amount transfers | Amounts set by agreement, not by invoice | 5 or more transfers in exact multiples of $1,000 in 30 days | Rent and standing orders are round; exclude recurring payees |
| 8 | Funnel account | Many senders in, one transfer out | 10 or more distinct senders in 7 days, then one outbound of 80% or more of the total | Fundraising, group gifts, marketplaces; read the declared purpose |
| 9 | New counterparty concentration | Most volume goes to a counterparty never seen before | More than 50% of 30 day outbound value to a counterparty first seen within 30 days | A new supplier or landlord; one alert, then whitelist with evidence |
| 10 | Cash intensity mismatch | Cash share does not match the declared business | Cash deposits above 40% of 90 day inflow for a customer not declared cash intensive | The profile may be out of date; refresh it rather than close the alert |
Each rule carries an owner, a version and the date it was last reviewed, and a new or changed rule is backtested on the firm's own history before it goes live. Whatever the vendor, ask the engine for the logic in plain words, the last change and who made it, and the alert to case ratio since.
Rule 1 in practice: cash deposits of 9,400 dollars on Monday, 9,600 on Wednesday and 9,500 on Friday. None crosses the 10,000 dollar line. No currency report is due on any one of them, yet together they total 28,500 dollars in five days. The alert should name all three deposits and the window.
In ZIP codes named in a FinCEN geographic targeting order (GTO), that currency reporting line sits lower for a set period, and a structuring rule has to follow it.
How do you tune thresholds and cut false positives without missing risk?
Tuning is a measured change: measure the rule, move the threshold on history, test below the line, record the change.
The numbers here are illustrative. Picture a payments firm with 2 million transactions a month. Rule 3 raises 4,000 alerts a month; 3,800 close at triage, 200 become cases and 40 end in a SAR, an alert to SAR rate of 1 percent. Rerun over six months of history with the count raised from 20 to 30 transactions in 24 hours, the rule raises 2,500 alerts a month and would still have caught 38 of the 40 SARs. The two misses are marketplace sellers with 22 and 26 transactions on a sale day, a segment problem rather than a threshold problem, so the firm keeps 20 for retail customers and applies 30 to sellers.
Below the line testing is the check an examiner asks about. Each quarter the team reviews a sample of transactions that sat just under each threshold, say 200 per rule. Cases in the sample mean the threshold is too high, and an empty sample still goes on the record as a test done.
What happens after an alert?
After an alert the work moves from the system to people, and the record changes from counts to narrative.
Triage comes first. The analyst reads the alert, the profile and prior alerts, then closes with a reason code or opens a case; for routine patterns an agent proposes the closure with the evidence attached and the analyst signs it. Write the reason code as a sentence. "Three deposits are the customer's documented weekly takings, matched to till reports received 3 March" closes an alert, and "no concern" does not.
Investigation comes second. The investigator pulls related alerts, counterparties and documents, requests what is missing from the customer and writes a timeline. The decision carries its rationale either way, because the examiner reads why a case was not filed as closely as why one was.
The guide to AML case management software sets out what an investigator needs from the case record and compares 8 tools from their own pages.
Then the narrative and the clock. The SAR is filed "no later than 30 calendar days after the date of initial detection", and the bank keeps it with its supporting documentation "for a period of five years from the date of filing" (31 CFR 1020.320(b)(3) and (d), read 6 October 2026). The guide to how to file a SAR covers the narrative and the form.
When the record is missing, the regulator rebuilds it. On 10 October 2024 FinCEN assessed "a record $1.3 billion penalty against TD Bank, N.A. and TD Bank USA, N.A." for violations of the Bank Secrecy Act; the order imposed a four year independent monitorship and a SAR lookback to "remediate the SAR filings that TD Bank missed due to its extensive control gaps" (FinCEN news release, 10 October 2024). A lookback is this section run backwards over years of transactions by outside reviewers, at the firm's cost. In the UK, the FCA's Final Notice to Starling Bank, dated 27 September 2024, imposed "a financial penalty of £28,959,426": the bank's financial crime controls "failed to keep pace with its growth". A skilled person appointed in 2021 had been instructed "to test the adequacy of Starling's transaction monitoring".
Real time or batch: when is each right?
Run in real time where the payment waits for the answer, in batch where it does not. Real time transaction monitoring scores a transaction on ingest, so an alert can hold the payment while the activity is in progress; it costs more in tuning, because a false positive now delays a customer rather than a report. Batch monitoring runs the rules over the day's or the week's transactions and suits patterns that only exist over time, such as velocity over 30 days or cash share of turnover. Continuous transaction monitoring means the review never stops between periodic customer reviews, whichever of the two runs it. Most programs run both. They keep real time rules to the few that justify a hold.
Payments
Payment transaction monitoring deals with volume and speed: many small transactions per customer and little balance, so count, velocity and funnel rules do more work than amount rules. Real time matters most at payout, the moment money leaves the firm's control. Merchants need their own segment, because a merchant's normal day looks like a retail customer's worst alert, and merchant accounts carry their own typology, transaction laundering.
Crypto
Crypto transaction monitoring, at exchanges and other virtual asset service providers (VASPs), adds the chain. The firm scores the wallet addresses a customer sends to and receives from for exposure to mixers, sanctioned entities and known thefts. The logic is the same (velocity, new counterparty, rapid movement), but the counterparty is an address and the lookup runs against chain analytics. In crypto and payments the same control is often called know your transaction (KYT): the per transaction check inside the broader program of rules, cases and filings.
Correspondent banking
Correspondent banking is the hard case, because the correspondent sees the respondent's customers only through payment messages. Rules lean on message fields: originator and beneficiary completeness, nested relationships and jurisdiction. The guide to high risk countries and FATF lists covers the public lists for the geography rules, and the firm's own risk assessment sets the final list.
For SWIFT payments, the GPI tracker follows each payment by its UETR from bank to bank and shows whether it was credited, rejected or put on hold.
How is monitoring different from screening, and from fraud monitoring?
Screening is a gate. Monitoring is a camera. Screening checks a name or a payment party against sanctions, politically exposed person (PEP) and watchlists at a point in time and answers yes or no before money moves. Monitoring watches what the money does over time and answers whether the pattern fits the customer. Both should read the same customer record.
In a cross border wire, SWIFT sanctions screening is that gate: it checks the parties, banks, addresses and free text of the payment message before release.
In the neobank case the monitoring alert says the pattern is wrong. Whether the beneficiary abroad is a listed party is a screening question, and the analyst can answer it in a minute with a free sanctions check on the beneficiary's name. A hit turns a suspicious pattern into sanctions exposure, with its own reporting route.
AML monitoring versus fraud monitoring
Fraud transaction monitoring protects the customer and the firm from loss; AML monitoring protects the financial system from the proceeds of crime. Fraud decides in milliseconds whether to let a payment through, and it blocks. AML decides over days and weeks whether behavior should be reported, and it files. The signals overlap (a new device and a new payee are both fraud signals and mule signals), so the two should read one customer record even as separate rule sets, and a mule account caught by fraud rules becomes a monitoring case with the handover on the record. The fraud detection and prevention guide covers the fraud side.
How do you build and measure the program?
A transaction monitoring program is the rules, people, records and reviews around the system, and the examiner's questions are about the program, whichever software runs it.
- A dated, written risk assessment. All three statutes set the extent of monitoring by it, and no threshold can be justified without one.
- A rule inventory with owners. Logic in plain language, threshold, version and the person who signs changes: the first document an examiner requests.
- A segment map. So a merchant and a salaried customer are not scored alike.
- Reason codes written as sentences. A reviewer who was not there can agree or disagree with the closure.
- The 30 day clock on the case screen. Detection date and days elapsed on every open case.
- Quarterly tuning with below the line testing. Recorded whether or not it finds anything.
- Five year retention of the SAR and its evidence. Readable in year five, with the rule version that fired.
An examiner asks for five measures: coverage (which customers, products and channels have rules), alert to case and case to SAR ratios per rule, backlog age in days, rule review dates, and the share of SARs filed inside 30 days of detection. Nobody counts those by hand at volume. Moldcell, Moldova's leading telecom operator with 1.5 million users, monitors about 1 million transactions a month on complead; the Moldcell case study, published in July 2026, covers its screening side, about 100,000 transactions screened.
A program that can print all five on demand, with the record behind them, has answered the opening question before it is asked.
The transaction monitoring guide: all pages
This page is the hub. Pages linked in the sections above are marked so; the others are linked here once live.
- AML transaction monitoring (this page).
- Transaction monitoring software, the product page with the parallel run offer, linked in the opening.
- Know your transaction (KYT).
- Mirror transaction (mirror trading).
- Smurfing: several people splitting criminal cash so no deposit reaches the reporting line.
- Travel rule: the originator and beneficiary data that moves with a transfer.
- Transaction monitoring rules: 25 examples with thresholds, the data each rule needs and the tuning method, linked in the rules section.
- The dated comparison of transaction monitoring software.
- AML software: the five modules an AML tool runs and what supervisors expect of it.
- AI transaction monitoring.
- Ongoing monitoring in AML: rescreening and profile refresh.
- Red flag indicators.
- SAR and STR filing, linked in the alert section.
- Cross-border transaction monitoring, from correspondent banking to wire transfer rules.
- Electronic funds transfer (EFT): which payments count, why a wire falls outside Regulation E, and the AML risks.
- Payment and transaction screening, the sender and receiver check inside the payment flow. Sanction Scanner, the screening product, is part of complead.
- SWIFT/BIC code: the 8 or 11 character code that names a bank in a payment message.