PHONETIC FINGERPRINT
Phonetic fingerprint is a mathematical representation of the pronunciation and phoneme of a name (individual person or other entity).
Using automated Real-time technology, supported by AI-ML layers, using over 48 mathematical algorithms, tracing phoneme & accurate name-matching across 38 different languages, transliterations and spelling variations.
ML TREND ANALYSIS & PERSISTENCE MECHANISM
An automated process, designed for ongoing verification resolving 60% of Potential Alerts, Automatically!
On each time there is a change in Sanction and PEP lists – changes will be verified automatically, and only real-alerts will be forwarded for further resolution; so unless there is a change of data the Compliance team will not be Alerted for no reason, whilst if there is a change it will not be missed.
Indexing sanction lists/PEP/Blacklists using Phonetic Fingerprint technology. Names translated to singular numeric ID, enable rapid and accurate matching in real-time. The use of mathematical representations ensures high-accuracy for both FP reduction and avoid FN.
applying AI/ML engines. Machine Learning looks at the frequency of Names, weighing and differentiating common names, unique names or combinations of names.
The use of Phonetic based methodologies (rather than rule based, enables Comparison of Entities in a very wide funnel; more name combinations, languages, transliterations.
I) Phonetic Threshold – checks name composition; i.e. how many naming words are included in the entity-search; b) II) Name-Distance Threshold, resolve the difference between names that spelled or sounds alike.
Based on Past Approvals Data, using ML Trend Analysis. automatic ongoing monitoring, resolution of re-alerts on the same entity, validates changes in watchlists on a daily basis, whilst reducing the workload on compliance teams
provide a wider name-screening coverage. Using advanced ‘name-understanding’, enable us to break thru the limiting traditional trade-off, that enforces compromising between FP and FN (as is in common fuzzy matching AI’s).
Pre-process: Indexing names in Sanction & PEP lists into “Phonetic Fingerprint” mathematical representation (Indexed Phonetic Mirror).
In non-automated systems, Alerts are resolved by the compliance team, and no further info is kept for future on-going verification. Moreover, once verified, past Alerted Names are forwarded to a White-listing, and thus have a major risk-exposure of missing real Alerts. Also – ongoing verifications are based on sample and random periodic batch files.
The complex task of screening the customer roster results in many false positives that re-appear and requires manual resolution, causing back log and heavy workload that needs to be processed.
In trying to tackle this issue FIs have incorporated “White Listing” methodology which essentially after a customer was flagged and it was resolved in a positive manner, they will white list that customer so he will not re-appear; the problem is that white listed customers can in time become sanctioned, simply due to more information surfacing that would identify them as sanctioned persons.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
Lorem ipsum dolor sit amet, consectetur adipiscing elit. Ut elit tellus, luctus nec ullamcorper mattis, pulvinar dapibus leo.
