Risk often hides in the connections between records rather than in the records themselves. A single account can look ordinary on its own, until you notice that it shares a device with three other customers, sends money to an IBAN that two of them also use, and gives an address that appears on a fourth profile. Seeing that picture used to mean exporting data, joining tables and drawing it by hand.
Names have a similar problem in screening. The same name can be written in many ways, and two spellings that look very different on paper can sound exactly the same when spoken. A screening that only compares letters can miss them, or catch too much when the threshold is lowered to compensate.
We built the July release around connections, coverage and precision. The highlights:
- Network Analysis shows transfers and shared device, IP, IBAN, address and phone links up to three hops away.
- Agentic AI and AI Suggest work across name screening, transaction monitoring, ongoing monitoring and adverse media.
- Phonetic screening catches names that sound alike, and each list can have its own minimum match rate.
- Nova opens in a full-screen workspace, and conversations are saved.
- AI translations open in the language you choose.
Network Analysis
See a customer's connections
The problem. Mule accounts and fraud rings show up in what accounts share, but those links were spread across transactions, devices and customer records, with no single view of them.
What changed. Network Analysis draws the connections on a visual network.
How does Network Analysis work?
Open the Network Analysis tab on a customer, or from a transaction monitoring case. The selected customer sits at the center, and the graph draws the accounts and customers connected to it. You choose how far to look, one, two or three hops. One hop shows direct counterparties, and three hops can reveal a structure that no single transaction would show.
Which connections does it show?
- Transfers: money moving between accounts.
- Device and IP: customers who used the same device or IP address.
- IBAN: customers who sent to or used the same IBAN.
- Address and phone: customers who gave the same address or phone number.
- Ownership: links between owners and the companies they own.
Each link type has its own look, so a cluster of shared devices stands out from a chain of ordinary transfers.
As an illustration, take a new customer whose first transfers look ordinary. At two hops, the graph shows that the customer's device was also used by four other recent customers, and that all five sent money to the same IBAN within a week. None of those transfers would raise an alert on its own, but together they look like a mule network.
What can you do from the graph?
When you find a suspicious node, you can expand it to see its own connections, add it straight to a case, or isolate it by adding it to the whitelist when the link turns out to be harmless. A full-screen mode gives a large network the room it needs. The investigation happens where the connections are visible, without leaving the screen.
Why it matters. Networks become visible in a few clicks, and the analyst can act on a suspicious link the moment it appears.
AI across the platform
Agentic AI and AI Suggest in four modules
The problem. Agentic AI started in name screening and AI Suggest in name screening and transaction monitoring, so ongoing monitoring and adverse media queues still relied on fully manual review.
What changed. Both now work in four core modules: name screening, transaction monitoring, ongoing monitoring and adverse media.
What does that mean in daily work?
In any of the four modules, an analyst can see a decision suggestion with a confidence score, then accept or reject it. Teams can also build agents for each module on the same drag-and-drop canvas: conditions on the AI's decision and confidence, followed by actions such as closing the case, setting the match status or risk level, assigning it or adding a note. An agent can run on demand or on a schedule. An ongoing monitoring agent, for example, can close clear namesake alerts each morning with the reasoning attached, and leave real changes for the analyst.
Why it matters. The same AI support follows the work across the main queues, under rules the team controls.
Screening precision
Phonetic screening and per-list match rates
The problem. Names spelled differently but pronounced the same could slip past a letter-by-letter comparison, and one match rate for every list forced a choice between missing hits on critical lists and drowning in noise on others.
What changed. Screening gained a phonetic layer and per-list thresholds.
How does phonetic screening work?
In a search profile, switch on Use Soundex Phonetic Search. Names are then compared by how they sound, not only by how they are written, so variations such as Mohamed and Muhammad are found together even when the spelling differs a lot. It is a setting per search profile, so you can use it where transliterated names are common and leave it off elsewhere.
How do per-list match rates work?
Each list in a search profile can have its own minimum match rate between 50 and 100. If a list has no rate of its own, the profile's default applies. For example, a profile can screen at 90% overall and at 70% for a list you consider critical, so that list is searched more broadly without widening every other search. A bulk option sets the rate for all filtered lists at once.
Why it matters. Screening is tuned to the risk of each list, and names that sound alike don't slip through.
Nova
A full-screen workspace and chat history
The problem. Nova lived in a side panel, which is handy for a quick question but tight for a longer analysis, and a conversation ended when the panel closed.
What changed. Nova has a full-screen workspace and keeps your conversations.
What is new in Nova?
A Nova AI item in the left menu opens the assistant on a full page, with room for longer answers and tables. Conversations are saved, so you can reopen an earlier analysis from the history and continue where you left off, or start a new chat. Suggested prompts on the opening screen show what Nova can help with, and support questions about how to use the platform are answered by the same assistant. The side panel stays available for quick questions.
Why it matters. Longer analyses have the space they need, and earlier work isn't lost.
Everyday work
AI translations in your language
Set the Default AI Translation Language in your user profile, and AI translations on profile and case summary screens open directly in that language, with no need to pick it each time.
A change log for customers and transactions
Activity Reports record changes to customer scores and transaction decisions field by field: who made the change, when, and the old and new values. In an audit, the history of a score or a decision can be read in one place.
Smaller changes you'll notice during the day
- Repeated identical amounts: a rule can detect several transactions of exactly the same amount within a period, for example ten successful transactions of the same amount in 24 hours.
- Internal and external customers: a customer can be marked as external when it is added only for screening, separate from your own customers.
- Deactivate instead of delete: customers can be deactivated and reactivated later with their scans, alerts and logs intact.
- Filters that stay: filtered screens remember your filters when you come back.
- Time in date filters: date filters accept hours and minutes.
- Guided tours: step-by-step tours walk new users through screens such as Name Screening and Customer Management.
- Match list details: match list entries can carry a note and show the date they were added, and deleted entries remain visible with their deletion date.
- Safe list history: safe list changes log the profile IDs involved and can be exported to Excel.
- Swift tables: Swift cases can be handled in bulk, and Swift reports show the case status.
- Transaction reports: the triggered rule description and the transaction IP address appear in lists and exports.
- Search profiles: auto resolution and secondary sanctions can be set in each search profile.
- Go back: detail screens return you to the page and filters you came from.
If you already use complead, the details of each update are in the Help Center inside the app. For other months, see the Product Updates category.
Previous month: June 2026 updates. Next month: August 2026 updates.