Party Consolidation (Resolution)⚓︎
Automated Party consolidation (resolution), is a process which attempts to consolidate multiple reported parties into consolidated party. A reported party represents the unique combination of characteristics provided on a report, while a consolidated party is intended to pull together reported parties to represent real world party.
This logic is based on determining what linkages satisfy the criteria for two reported parties to be considered the same real world party for the purposes of automated monitoring. The logic that determines if two reported parties should be consolidated is based on:
- The similarity of individual linkages. For example: "Johnny Smith" and "John P. Smith" (both with an address of 15 Treewood Street Sydney 2000 NSW) may be considered similar enough.
- Frequency of the linkage. For example: name "J. Smith" or DOB "1/1/1970" maybe linked to so many reported parties that it may not be able to be used for party consolidation at all or provide only very low weight in the consolidation of reported parties.
- The number and types of linkages. For example: reported parties with exactly the same name and email may be consolidated, while a reported parties with similar addresses and dates of birth may not be consolidated.
Party Consolidation Examples⚓︎
A simplistic consolidation algorithm is used within FI-Comp to generate the consolidated party resources. The below illustrates some examples of reported parties that have been consolidated together due to shared linkages.
The notebook used to perform the consolidation can be found here.
Party Consolidation Processing⚓︎
As mentioned earlier there is no universal identification for a reported party, so every time a new reported party is loaded into a financial intelligence datastore an automated process attempts to evaluate if the reported party should be consolidated with all the previously reported parties. This contrasts with the Single View of Customer
(SVoC) challenge which most businesses are familiar with, where data across disparate systems is matched against a single register of customers.
A key element of the consolidation process is how to determine the order that reported parties should be processed through the consolidation process. Noting that once a reported party has been consolidated with another reported party moving them to a different consolidated party may result in automated monitoring re-triggering on the new consolidated party and confuse analytical outputs. It may also result in very different results for an analyst who is performing interactive analysis (e.g. today the analysis identifies 50 reports associated with a consolidated parties with a specific address, but tomorrow they may only identify 10 reports associated with consolidated parties with the same address). That is, there is a balance between the stability of consolidated parties and the creation of optimal consolidated parties.
Chaining Strengths and Weaknesses⚓︎
Chaining is a key concept in the creation of consolidated parties. Chaining is a beneficial process that consolidates reported parties together even through they do not have direct linkages. This concept is illustrated below, it can be seen that even though the reported parties on the left and the right share no common identifiers they are "chained" together by the central party and therefore consolidated.
However long chains of consolidation can often result in the incorrect consolidation of reported parties, as small variations of identifiers are permitted within each link in the chain. For example:
John Bakerman
andJohn P Bakerman
are consolidated, thenJohn P Bakerman
andJohn Phillip Bakerman
are consolidated, thenJohn Phillip Bakerman
andJon Phil Bakeman
are consolidated, thenJon Phil Bakeman
andJo Phil Bakman
are consolidated, thenJo Phil Bakman
andPhil Pakman
are consolidated, then and so on ....
Highlighting again that the party consolidation process is a balancing act between competing priorities, in the context of chaining the balance is between more consolidation vs potentially lower quality groups.