Smarter screening and explainable risk
Name search, batch screening, bulk clears with four-eyes per hit, clearer match explanations, and risk ratings you can explain and simulate.
tocOn this page
Screening and customer risk rating are where compliance teams spend much of their time and where examiners ask the hardest "why" questions. This release makes both faster to work and easier to explain, without loosening any of the controls around them.
Screening
Name search and batch screening
Name search (/screening/search) looks a name up against your lists without creating anything. Enter a full name and, optionally, a date of birth, country and party type, set the fuzziness between 70 and 95, and tick the lists you want. Each result shows its score and the three components behind it, and Profile opens the entry's aliases, dates of birth, nationalities, identifiers and remarks. The search is recorded in the audit trail.
Batch screening (/screening/batch) does the same for a CSV of up to 5,000 rows. Only the name column is required. Results show the best candidate per row with its score and reading, and Export CSV produces a formula-safe file. The audit log records the counts, never the names.
Bulk clear, with four-eyes per hit
On the screening workbench, tick several potential hits and choose Clear as false positive, give one reason code and a narrative, then Clear hits. Four-eyes still applies to each hit individually:
- hits on statutory (TFS) lists, or with scores above your policy threshold, become Clear proposed for a checker;
- the rest clear at once, are remembered and are sampled for QA.
The result tells you how many were cleared, sent to a checker or skipped.
Why this hit
The hit review page now sets the subject and list entry Side by side with matched words highlighted, and adds a Reading column for each attribute: agrees, close or differs, for date of birth, nationality or country, identifier and type. Why this hit explains the match in plain English, and Score breakdown shows token alignment, full-name similarity, phonetic match and the threshold. The explanation is advisory and never changes the engine score.
Good-guy exclusions and the PEP register
Exclusions (/screening/exclusions) records known customers who are not the listed party. While an approved exclusion is in force, screening skips that pair in real-time, onboarding and rescreen screening. Each exclusion has an expiry of at most three years, and the page shows what expires in the next 30 days.
The PEP register (/screening/pep) holds your PEP and RCA designations with category, source and rationale, lists reviews due for designations older than 12 months, and handles declassification requests.
Explainable risk
How the score is made up
Every customer's risk tab now shows a waterfall: each bar is a factor's contribution (group weight × factor weight × factor score) with a plain-English reason, plus any floor rules that raised the band. Score history shows how the rating moved, and Explain this rating gives an AI reading of what drives the rating and what would change it.
Overrides with approval
Propose override sets a band with a reason of at least 20 characters and an expiry of at most one year. A different person with model-approval rights approves it with a fresh second factor. The calculated score is still recorded, and a High override opens an EDD case.
The what-if simulator
On Risk models (/risk/models), the What-if simulator lets you change factor group weights and band cut-offs, then Simulate. A band-shift matrix shows "Customers re-scored" and "Would change band", with the customers who would move. Nothing is saved until you choose Save as draft version, and a draft still needs approval before it is activated.
EDD workbench and EWRA
The EDD workbench (/risk/edd) opens a case automatically when a customer's effective band is High. Reviewers mark each checklist item complete with a note and submit; a senior approver signs off the relationship.
EWRA (/risk/ewra) brings the enterprise-wide risk assessment into the product: cycles, a risk register, an inherent-risk heatmap (likelihood × impact) and residual risk by band.
What stays governed
- Statutory lists always need two people for a true match or a clear.
- Exclusions, PEP declassifications, risk overrides and model activations are all maker-checker, with step-up for the checker.
- Nobody approves their own proposal. The server enforces it on every item.
- AI explains, people decide. The match explainer and risk explainer never change a score or record a decision.
How to try it
- Open Screening, then Name search, and look up a name with a date of birth.
- On the Hits workbench, open a hit and read Why this hit and the Reading column.
- Open a customer, go to the risk tab and read How the score is made up.
- If you hold model rights, open Risk models and run the What-if simulator.