Metro’s monitor missed the day the account was opened
The FCA’s November 2024 notice says a timestamp error kept same-day transactions, and everything after them until the account record moved, out of the automated system. More than 60 million transactions, over £51 billion, were not monitored.
The fine is the small number
On 12 November 2024 the Financial Conduct Authority fined Metro Bank Plc £16,675,200. Metro settled at stage one, so the figure includes a 30 percent discount. Without it the penalty would have been £23,821,700. The Authority imposed it under section 206 for a breach of Principle 3: the bank did not take reasonable care to organise its affairs with adequate risk-management systems. The fine is not the fact that matters. The fact is what the system never saw. Metro switched on an automated transaction-monitoring system on 6 June 2016. That is the start of the relevant period. By December 2020, the notice says, the bank had failed to monitor over 60 million transactions, about 6 percent of volume, with a value of over £51 billion, about 7.6 percent of value. A monitoring system that is dark for one transaction in sixteen, by value closer to one in thirteen, is not a system with a gap at the edge. It is a system with a hole in the middle of ordinary business. Many of those transactions were looked at later, in a remediation exercise. The Authority’s point is the delay. Suspicious activity that is found years after the payment is a file for the archive, not a report in time for anyone to use.
Why the same day fell out
The error has a dull name, which is how you know it is real. The notice calls it the Time Stamp Code Logic Error. It sat in the way data was pulled from the bank’s records and loaded into the monitor. If a customer opened an account and transacted on the same day, the account record was not included in the feed. The monitor then rejected the transaction records, and kept rejecting later transactions on that account, until something updated the account record and the system could reconfigure. Referential integrity is the jargon. The plain version is that a payment with no account attached to it, inside the monitor, was thrown away as bad data. The customer existed. The payment existed. The monitor declined to hold both at once. The technical mismatch is specific enough to copy into a test. The timestamp on the account record was “processing day plus one.” The logic that built the batch file looked for “processing day minus one.” Those two clocks never met, so the account record never left the source system on the day it was needed. Customer records and transaction records were extracted. The account record was not. From June 2016 until the error was found, that rule was the production rule. Nobody had written a test that opened an account, paid it the same day, and asked whether the monitor received the row. The notice’s count for this one error, over roughly three years, is in the region of 166,000 accounts, more than 46.5 million transactions, and more than £31.5 billion. That is not the whole 60 million. Other records were rejected as bad data for other reasons, and the Authority treats that second pile as part of the same failure of oversight. When you quote the case, keep the two piles apart. The timestamp error is the mechanism. The £51 billion is the period. Mixing them makes a precise notice sound like a round number.
It was found by accident, and the minutes lost the warning
Metro found the error on 17 April 2019, during testing of an upgrade to its core records system, almost three years after the monitor went live. It told the Authority on 29 May 2019. A tactical fix went in on 21 July 2019. After that fix, the notice says, at least 99.7 percent of transaction records that were supposed to be monitored were consistently fed in. Metro did not know that. It had no ongoing check that would have told it. Ad hoc reconciliations in 2020 looked at customer and account records, not at transactions. A count check that actually compared what left the source system with what arrived in the monitor was not in place until 17 December 2020, four and a half years after go-live. The Authority’s view is that a reconciliation was part of the monitoring system, not a nice-to-have after an incident. If it had existed in June 2016, the clock mismatch could have been seen while the missing population was still small. A lookback started in July 2019. Metro compared accounts in the source system with accounts in the monitor as at 1 July 2019 and found about 238,000 accounts missing. Of those, 166,358 were missing because of the timestamp error. The lookback is the remediation. It is also the proof that, for three years, the bank did not know which accounts the monitor had never met. Junior staff had already said the data was wrong. In 2017 and 2018 people at less senior grades investigated bad data and tried to escalate it. In January 2018 the Financial Crime Steering Group, the committee charged with overseeing this risk, agreed to review the issue. That decision was then taken out of the final minutes, on the basis that the issue did not appear to be substantiated, with the intention of coming back once it was better understood. It was not discussed there again until April 2019, when the error had already been found by the upgrade test. Members of the working group and the operations risk board escalated in 2018 as well. The risk was passed to a more junior member of staff to take to a senior one. The Authority records that Metro has no formal record that the conversation happened, and that no substantive action followed. A risk that leaves the minutes has left the institution. The system kept rejecting the same-day book, and the committee that owned financial crime had agreed not to keep the sentence.
What to ask, and what not to conclude
The useful question for any bank that bought a monitor and a core system from different generations is not “do you have automated monitoring.” It is “what happens to a transaction on the day the account is opened, and who reconciles source counts to monitor counts every day.” Ask for the exception report, not the architecture slide. Ask whether a failed row is queued, rejected, or silent. Metro’s rows were rejected, and rejection looked like the system working. Bad data was a category the monitor knew how to discard. Discarding it was the control failure. Do not conclude that £51 billion was laundered. The notice says those transactions were not monitored, that the delay increased the risk the bank would be used for financial crime, and that many were reviewed later. Unmonitored is not illicit. It is the condition in which illicit can sit without a record. Do not treat the tactical fix in July 2019 as the end. The bank still lacked a consistent completeness check until December 2020, and it did not know, at the time of the fix, that 99.7 percent was the number it had reached. Keep the final notice, the two populations (166,000 accounts and the wider 60 million), and the January 2018 minute that dropped the action. If your own feed can reject a payment because a parent record is a day out, you have this finding in draft. The test is one account, opened and paid on the same day, and a person whose job is to see that it arrived.