Trends
The direction a value is moving in, measured over a set number of events or over a time window. A steady climb shows up before any single value looks unusual.
trend · last N events, or N minutes, hours, daysFinding patterns no single record shows
Pattern Recognition reads every record against the history behind it, finding the trends, anomalies and changes that no single record shows. Many risks can be seen this way before any one transaction gives them away. Your rules stay in place, and now they can act on patterns as well as single events.
Patterns across events
One record is an event.
Many make a pattern.
A rule that checks a single record sees one moment. Pattern Recognition reads each record against the history behind it, for the same client, counterparty or both. Over time that history settles into a shape of its own, and a change in it shows up even when no single record looks unusual. That makes it an early warning: many risks can be seen before any one transaction gives them away.
Records that fit the patternA change in it
Schematic.
Four kinds of pattern
From the same history, four ways to see a change.
Pattern Recognition calculates values over the history in the datamart, grouped by client, counterparty or any other field, over minutes, hours, days or working days. Four kinds of pattern come out of it. Rules decide what happens when one appears.
Works onClientsCounterpartiesAny field in the datamart
Schematic. Which values are calculated, over which periods, and which rules act on them is configured per institution.
Anomalies
When the latest value breaks from its own history. It is compared with the average, the range or the last values recorded for the same client or counterparty.
latest ÷ average · min · max · last NClusters
Clients grouped by how they behave. One that moves away from their own history or from their group stands out.
behaviour cluster · own history · peer groupNovelty
Something not seen recently: a counterparty, country or channel that doesn't appear among the last N values.
unique among last NHow a pattern is defined
Every pattern is built from four choices.
The values Pattern Recognition calculates are defined in a base model, set up for your institution. Each one sets what is calculated, what it is grouped by, over which window, and under which condition. The result becomes a named field that rules can use, next to the fields that come straight from your systems.
01 · calculationWhat
02 · keyGrouped by
03 · windowOver
04 · conditionOnly when
The value
Total value of records in one category, per client, over the last 30 days.
sum(value)by customer30 dayswhere category = X
Becomes a field rules can usecust_catX_sum_30dExample rule
If cust_catX_sum_30d is above your threshold, open an alert.
Illustrative field names. The base model is set up for each institution.
Where the results go
Alerts, scores and your core system.
A calculated value is only useful once something acts on it. When a rule is met, the result can open an alert for review, update the client’s risk score, or both, and the risk class can go back to the systems your teams already use.
One client, three placesC-20418
AlertViewer
Alert queue
C-20418No open alert
C-11873Name screening
C-07391Pattern rule
Client risk
Risk profile
Risk score42
Risk classMedium
Rules triggeredNone
Your core system
Client record
ClientC-20418
Risk classMedium
SourceDetectX®
Nothing has crossed a rule yet.Rule: cust_catX_sum_30d > 10,000
Illustrative client, fields and values.
Explore the DetectX® Platform
Every pattern needs something to act on it.
Pattern Recognition finds what changed. The other capabilities score it, link it and decide what happens next. Pick one to see what it does.
Pattern Recognition FAQs
What compliance teams usually ask before they choose Pattern Recognition.
Something we have not covered?Talk to our team
No. Rules on single records catch what you already know. Pattern Recognition gives them the history to catch what one record can't show, and both run on the same core.
An account shows rapid high-value transfers, or frequent small withdrawals followed by one large transaction. No single payment breaks a rule, but the pattern does, and the account is flagged for investigation.
Clustering is one of the four kinds of pattern. Pattern Recognition also calculates trends, anomalies and new values, and turns any of them into a field your rules can use.
Yes. Each pattern is set up in the base model for your institution, by choosing what is calculated, how it's grouped, over which window and under which condition.
No. A manufacturer can read machine sensor data and spot temperature or vibration changes that signal a defect early. A logistics operator can read supplier, route and schedule data and detect recurring delays before they become bottlenecks.
Go Deeper
All resourcesSolutionTransaction Monitoring
Where the calculated values are used: each client's activity checked against their own history, with every alert traced to its rule.
See Transaction MonitoringSolutionBehavioural AnalysisHow a change in behaviour is spotted early: each client compared with their usual pattern and with clients like them.
See Behavioural AnalysisCase studyEarly detection of delivery delaysHow a logistics operator read supplier, route and schedule data to spot recurring delays before they became bottlenecks.
Read the case studySee Pattern Recognition in a demo
Book a working session on your own data. Bring your questions about what the engine reads and how a pattern becomes an alert.