Transaction monitoring, filing drafts, and thresholds you set yourself — what shipped
Nine releases since the last of these — v1.44 through v1.53. Rather than list them, here are the four things that change what you can actually do, and a handful of smaller items at the end.
Transaction monitoring, and it is not AI
This is the biggest addition. Until now ARGOS screened names; it did not look at payment flows. It does now.
You upload a transaction file — a CSV export from your ledger, with sender, beneficiary, amount and date — and get alerts back against five patterns: structuring, velocity, funnel (many senders converging on one beneficiary), fan-out (one sender paying many beneficiaries), and high-risk corridor exposure. Each alert names the subject, gives a plain-English reason, and lists the exact transactions that triggered it. One click opens it as a real case file, pre-filled.
The detection is rules-based, deliberately. No model is involved in deciding whether something is an alert. That’s a design decision rather than a limitation we’re working around: an alert you have to justify to a regulator is worth more when the reason it fired is a threshold you can point at, not a score you have to explain. Every alert states which rule fired and why.
Two honest limits. It reads an uploaded file — it is not yet plugged into a live payment feed, so this is a review you run, not a stream that watches. And there is no per-customer historical baseline behind it: the rules compare against thresholds and time windows, not against what a given customer normally does. If a customer’s activity is unusual for them but ordinary in absolute terms, these rules won’t catch it. We’d rather say that than let you assume otherwise.
You set the thresholds now, not us
Shipped a day after monitoring itself, and the more useful half of it for anyone running this seriously.
Every number the rules use — the structuring amount, the time windows, the minimum counts of transfers, senders and beneficiaries — is now editable in Team & Tenancy, per workspace. Leave a field blank and it uses the built-in default, which is shown to you as the placeholder. Change it and it applies to your next upload.
The reason this matters: a sensible structuring threshold in one corridor is noise in another, and the version of this that shipped first required us to change a setting for you. That’s not a workable arrangement for a compliance function that needs to tune its own alerting and record why. Now the numbers are yours, and changing them doesn’t involve us at all.
Filing drafts, with a second pair of eyes required
A completed investigation can now be drafted into a regulatory filing — structured to the regulator’s own form rather than as free text: the UK NCA SAR Portal and India’s FIU-IND formats are in.
The part worth knowing is the workflow, not the drafting. A draft cannot be marked ready by the person who created it. A second reviewer is required before it can be exported — four-eyes, enforced by the system rather than by procedure, and recorded in the audit trail either way. A filing is one of the few things in this job where being fast and being wrong is worse than being slow.
Screen the whole book at once
Batch screening arrived in v1.48. Upload a CSV of names, watch live progress, then export the results — or push every hit straight onto ongoing monitoring rather than re-screening it by hand next quarter.
Smaller things
- Two-factor authentication is available on your account (authenticator app, plus one-time recovery codes), along with a page listing your active sessions and a button to revoke any of them. It is optional and stays optional — nobody gets force-enrolled on their next login.
- Case files got a lifecycle: statuses, a required second reviewer to close, file attachments, and a customer risk rating carried on the case.
- The Overview page became a work queue — your open cases, your recent screens, and anything overdue — rather than a wall of counters. System health moved to its own page, where it belongs.
- Watchlist and ongoing monitoring merged into one place, with on-demand re-screening available from it.
- A chat answer containing two separate bullet lists could render them merged into one malformed list. Fixed.
- Chat answers and investigation reports now record which model produced them, visible in the audit trail.
Coming, and honestly not here yet
The live-feed version of transaction monitoring — where ARGOS watches transactions as they happen instead of when you upload them — is a real piece of work we haven’t started, because it needs a connection into your payment systems and that deserves its own conversation rather than a quiet release note. If that’s the version you need, tell us and it will shape the order we build in.
If anything here doesn’t behave the way this describes, say so. That’s still how the most useful fixes get found.