Eighty-Four Reports a Day

The Money Laundering Reporting Office Switzerland received 21'087 suspicious activity reports in 2025. Spread across roughly 250 working days, that is 84 reports a day, one every six minutes of a working day, filed by banks that are legally obliged to investigate before they file and to keep investigating after.

The level is manageable. The trajectory is the problem. MROS received 5'334 reports in 2020, 7'639 in 2022, 15'141 in 2024, and 21'087 last year: almost four times the 2020 volume, a compound growth rate of about 32 percent per year, with 39 percent growth in 2025 alone (MROS Annual Report 2025). Any compliance function planning its headcount linearly against a workload compounding at 32 percent is solving the wrong equation. At that rate, today's volume doubles roughly every two and a half years.

The scissors

The Swiss numbers are one blade of a global pair of scissors. ACAMS, SAS, and KPMG (2025) estimate USD 4.4 trillion laundered in 2025 against USD 206 billion spent fighting it: 21 dollars of crime for every dollar of defence. Illicit flows are growing at roughly 19 percent a year while compliance budgets stay flat (ACAMS/SAS/KPMG 2025; Verafin 2026).

And most of the flat budget chases noise. LexisNexis (2023) puts the false-positive rate in sanctions, watchlist, and adverse media screening at 99 percent. Ninety-nine alerts cleared to find one real signal. The industry has responded to rising crime by generating more alerts for the same people to clear, which raises cost without raising catch rates. The scissors are widening: the crime compounds, the spend does not, and the spend that exists is consumed by false alarms.

You cannot hire your way out

The traditional lever is headcount, and it is broken in three places.

Labour is already the dominant cost: 57 percent of financial crime compliance spend goes to people (LexisNexis, 2023). The people do not stay: WorkFusion (2025) puts typical tenure for a first-level analyst at around 12 months, which means an analyst is trained for months, productive for months, and gone. And replacements are scarce: ManpowerGroup (2026) reports that 68 percent of financial services employers cannot find the talent they need.

The churn is not mysterious. First-level alert handling is the work of clearing 99 false alarms in the hope of finding the hundredth alert that matters: repetitive, high-stakes, and professionally unrewarding. The job that the volume growth demands more of is precisely the job that people leave fastest. A workload compounding at 32 percent cannot be met by a labour pool that shrinks through churn and cannot be refilled through hiring. This is arithmetic, and it has already made the decision.

The bar rises anyway

While the volume compounds, the standard tightens. Switzerland adopted the revised AMLA and a new federal register of beneficial owners in 2025. The revised AMLO-FINMA enters into force on 1 January 2027, just ahead of the next FATF evaluation of Switzerland. In Frankfurt, the EU's new Anti-Money Laundering Authority has been operational since July 2025, and its supervisory expectations create convergence pressure on every Swiss group with EU operations.

None of these developments reduce the work. Each expands what must be screened, documented, and reported, on a timeline the bank does not control. Demand is compounding, the labour supply is broken, and the regulator is raising the bar on both. Few functions in banking face all three at once.

Adopt what you can govern

The obvious response is agents, and here the industry's caution is instructive. The Bank of England and FCA found in 2024 that 75 percent of financial firms already use AI. PYMNTS Intelligence (2025) found that only 15 percent of CFOs consider themselves ready to deploy agentic AI. The gap between those numbers is a governance statement: banks adopt what they can explain to model risk, audit, and the supervisor, and nothing more.

That is the right instinct, and it defines the path. What passes governance today in financial crime operations are pre-trained, explainable models with a human in the loop: systems that resolve first-level alerts, document their reasoning, and route everything above a defined confidence threshold to a person. This is Level 2 to 3 on the autonomy ladder, a lane of well-bounded decisions with escalation designed in by cost of error. Full autonomy is earned later, alert type by alert type, as the track record accumulates.

The market has noticed that this workforce needs workforce management. Vendors now ship compliance agents with names, defined scopes, and performance histories, which is exactly the job description question arriving in production: what is this agent hired to do, against which policy, reviewed by whom.

What stays human

The reporting decision itself does not move. Filing under Art. 9 AMLA is a judgement the bank must own, and FINMA Guidance 08/2024 is explicit that responsibility for decisions cannot be delegated to AI. A named person decides what goes to Bern and defends it afterwards.

What changes is everything before that decision. Agents assemble the case: transactions gathered, counterparties resolved, adverse media read and weighed, the 99 noise alerts closed with documented reasoning. The human receives investigations instead of alerts. The compliance officer's day shifts from clearing false alarms to judging real cases, which is both the job the regulation actually requires and the job people might stay for.

The arithmetic at the top of this post does not resolve itself. By 2028, on the current growth path, MROS volumes imply well over 150 reports per working day, before AMLO-FINMA widens the funnel further. The institutions that keep pace will be the ones that stopped asking whether agents belong in compliance and started defining the envelope in which they operate. Eighty-four a day was the 2025 number. The operating model that handles the 2028 number is being built now, or not at all.