Third-Generation Release of Evelyn and Tara: GenAI + Enhancements Drive Higher Automation Rates, Improved Performance and Better Customizability 

WorkFusion has announced the general availability of its third generation of Digital Workers, Evelyn, an AI adverse media and sanctions screening alert review analyst and Tara, an AI transaction screening analyst. The latest release, which incorporates GenAI, provides significant product enhancements and further revolutionizes the way financial crime compliance teams perform their adverse media and sanctions payment screening alert review functions. 

“We’ve introduced game-changing functionality in this third- generation release of the Evelyn and Tara Digital Workers,” said Peter Cousins, WorkFusion’s Chief Technology Officer. “Incorporating LLMs into Evelyn’s adverse media decision-making process in a way that pushes automation rates to over 90% is just one of many significant enhancements in Gen-3 Digital Workers. With Evelyn 3, out-of-the-box integration with Thomson Reuters, a new connector framework, the choice to use multiple search tools simultaneously, and customizable decision narratives give compliance teams the flexibility they need to optimally perform their vital work.” 

“Similarly, the enhancements to Tara 3 for transaction sanctions screening are quite exciting for our customers. Tara 3 can automatically adjudicate 60 to 80% of transaction sanctions alerts at speeds that are 30% faster than in Gen-2. But the true differentiator in Tara 3 is her ability to retain and access historic decisions and apply them to new payment alert decisions. This eliminates the need for ML compute in up to 50% of all payment sanctions decisions, making it less resource intensive, both in terms of people and technology.” 

Three Key Innovations of Evelyn 3 for Adverse Media and Sanctions Screening Alert Review 

(1) GenAI: Boosting automation rates while reducing errors. GenAI is a big addition to Evelyn 3. The adverse media alert review process can take five to 20 minutes per review. Using AI, the standard automation rate to review adverse media for Digital Worker Evelyn typically reduces the manual work by 60-80 percent. But with the addition of LLMs, Evelyn 3’s automation rate can reach up to 95 percent straight-through processing (STP), when/where an organization’s risk appetite allows. 

Harnessing the power of GenAI, much like a patient seeking a second opinion from a doctor, Evelyn can now leverage two streams of decision-making, dramatically lowering the odds of an error in either of them. People only become involved when there is a tie-breaking decision to be made. “Many people recognize the strength of GenAI to drive up automation rates. Yet, it can also drive down error rates to near zero,” said Peter Cousins. “WorkFusion adds an LLM into Evelyn’s adverse media alert adjudication process. Just as banks have always leveraged humans in the loop, Evelyn leverages ‘LLM in the loop’ to inform the results when Evelyn is less than 90% certain of a specific alert adjudication decision.” 

(2) Simplified Integrations: Thomson Reuters out-of-the-box integration and a new Connector Framework. Evelyn 3 provides an out-of-the box integration to the Thomson Reuters Adverse Media and Sanctions solution (aka Thomson Reuters CLEAR). The integration’s impact comes in the ability to perform lookback and ongoing monitoring. “For financial institutions with a million or more customers, rather than having to go every day or every period that they determine and start from scratch with a lookback window on customers, they gain efficiency by allowing Thomson Reuters to notify them when there is anything new to see. This allows Evelyn to gain a boost from ongoing monitoring in a much more scalable way,” explained Peter Cousins. 

The new Connector Framework makes it easy for Evelyn 3 customers to flexibly add new sources of data and integrate Evelyn into both upstream and downstream systems of virtually any kind. Using the Connector Framework, customers simply provide an API spec or download one from an open repository, then choose the fields from which Evelyn 3 will pull content. 

(3) Business and Technical Flexibility: Analytics, search and decision narration. Evelyn 3 introduces a new analytics platform with more capabilities and flexibility. The dashboards have migrated to Apache Superset, a widely available product that allows for more meaningful analytics data views. WorkFusion has programmed flexible filtering and customizable dashboards into Evelyn 3. 

Evelyn now gives customers the flexibility to use multiple search providers in any configuration. So, customers use their preferred search providers, such as Google and Thomson Reuters, or Thomson Reuters and Refinitiv. Customers can also pull data from any internal database. 

Compliance teams can now customize their alert decision narratives within the WorkFusion user interface to align with their organization’s language norms. Previous generations of Evelyn provided one-size-fits-all decision narration. While effective, customized narratives bring a more ‘human’ feel to reporting. Being able to summarize decisions and explain them succinctly in a way that human colleagues can understand, including supervisors, risk officers, and regulators ensures explainability and trust. Detailed narratives allow AML analysts to move from being authors to editors, collecting data from various sources and simply reviewing it for better understanding. This speeds up the time for an analyst to make a decision, be it the speed to adjudicate an alert or the speed to file a suspicious activity report (SAR), etc. 

Three Key Innovations of Tara 3 for Transaction Screening Alert Review

(1) Decision Retention and Access. Tara 3 For banks and other FIs that process millions of payments per year, applying historical decisions to current alert reviews means significant resource savings. Not only does it eliminate the need for a person to review an alert with an applicable historical decision, but it also eliminates the need for computing resources surrounding that alert because the decision has been made previously for an identical alert. Tara 3 simply uses the prior decision and narrative to support her adjudication decision for the new alert. 

(2) Higher automation rates and improved performance. Tara continues to learn more each day as she adjudicates millions of alerts. Add to this Tara 3’s automated application of historical decisions, and Tara 3 delivers automation rates in the 60-80% range – a substantial uplift from just a year ago when Tara 2 was the standard. 

On the performance front, Tara 3 can process 10,000 records in just 46 minutes, or approximately one-quarter of a second per record. This represents a 30% improvement over Tara 2’s already impressive performance.  

  Tara 2  Tara 3 
Volume  10k  10k 
Time  66 minutes  46 minutes 
Per record  0.39 sec/record  0.27 sec/record 

“The combination of making ever-more efficient algorithms and using the Python programming language has delivered substantial performance improvements in Tara 3,” said Peter Cousins. 

(3) Optimized decisioning. Tara 3 provides optimal decisioning by incorporating three new technology innovations: 

  1. The name matcher 
  1. The new rules engine 
  1. Context recognition 

Tara 3’s name matcher was designed with more robust training data which WorkFusion gleaned from handling extremely large volumes of transaction sanctions alerts. Tara 3 also benefits from the use of state-of-the-art algorithms. The name matcher model employs a staged preprocessing approach, effectively filtering out irrelevant information. Once the process starts, the model uses dedicated logic for name type, separating matching logic for personal and company names. It also incorporates semantics, enabling the model to weigh the significance of matches, improving match relevance and reducing the likelihood of mismatches. 

The new rules engine within Tara 3 introduces a significant enhancement in how alert disposition decisions are made. This new approach utilizes multiple calculated data points generated by model execution. Leveraging greater detail within the data enhances the logic behind alert adjudication decisions to a more granular level. There is also a starting set of rules based on best practices shared across all Tara users, and it is tailorable to each customer. 

Finally, Tara 3 has been enhanced with the ability to recognize the context in composite fields and establish linkages between multiple elements, such as account number, name, address, CUSIP, ISIN, IBAN, BIC). She then arranges all the data into a single composite field that understands the context. 

To take advantage of the exciting innovations available in Evelyn 3 or Tara 3, contact your WorkFusion account executive today. 

To learn more about our new releases or how Digital Workers can help your business, please request a demo

Share this article