![pattern](/_next/image?url=https%3A%2F%2Fedge.sitecorecloud.io%2Fsitecoresaa1efd-pointbdxp-production-37ec%2Fmedia%2FProject%2FPointB%2FPointB%2FImages%2FHome%2FInsights%2F2023%2F12%2FImplementing-a-Machine-Learning-Model-for-a-Global-Social-Media-Company_1.png%3Fh%3D485%26iar%3D0%26w%3D1440&w=3840&q=75)
Challenge
Peer-to-peer payments and e-commerce are becoming a normal part of the social media experience, and social platforms have expanded their financial capabilities to stay competitive. This includes allowing nonprofit organizations to solicit and collect donations from individual donors on their platforms – highly regulated transactions.
One global social media company found that it urgently needed a more thorough and scalable solution to assess its backlog of Know Your Customer (KYC) identity verification alerts from payment transactions. We worked closely with internal teams to create new procedures and bring the company into compliance. Part of that effort involved harnessing the power of machine learning natural language processing (NLP) to automate a portion of the process, setting the stage for efficient screening that can scale to accommodate increased volume without sacrificing accuracy.
We also helped this global social media establish a charitable Risk Rating Model in just 8 weeks. Learn all about it here.
![](https://edge.sitecorecloud.io/sitecoresaa1efd-pointbdxp-production-37ec/media/Project/PointB/PointB/Images/Home/iStock-Downloads/Data/iStock-1413502657.jpg?h=810&iar=0&w=1440)
Opportunity
During an internal review, the company identified that some donors were screened as Politically Exposed Persons (PEP), a part of the existing KYC process, creating a significant backlog of events to be reviewed. Our customer self-disclosed this gap to their international regulator and a resolution target date was mutually agreed upon. Without the resources to take immediate action, leadership looked to a trusted partner to develop a plan to expedite their four-month deadline.
We first put together a cross-functional team with the necessary skills and decision-making responsibilities, including the Deputy BSA (Bank Secrecy Act) Officer as executive sponsor, Head of Anti-Money Laundering (AML) as business owner, data scientist and data analyst, change management lead, and compliance subject matter experts. Then, we worked with that team to define work products, interdependencies, timing, and goals to ensure they aligned with regulations and business outcomes.
Our Approach
We engaged the Data Science team to determine the build status of their custom, in-house NLP model. The NLP model ingests the text-based alerts from the PEP screening service provider, parses information in the reports to meet auto-close criteria and passes the results to the case management tool. Â
The key to success was determining what was needed from the compliance SMEs to operationalize their model so it’d ingest PEP alert information and auto-close false positives. The team worked closely with the compliance SMEs to confirm:
1
Model Design.
2
Auto-close dismissal justifications.
3
Case management queue tags.
4
A/B testing for NLP model.
5
Organization and execution of go-live cutover activities.
At the same time, we helped the Data Insights team get Legal approval for the data definitions used to identify a donor’s geographic residence (which placed them in the international regulator’s jurisdiction). Using the new data definitions, we managed the reprocessing of over 6 million donors’ information with the company’s third-party PEP screening service provider.
Then, we engaged the change management team to update desktop procedures, create a job aid for compliance analysts, revise process error codes for quality control measurement, and schedule the go-live date.
Project Outcomes
After the cross-functional team implemented the new processes, our social media customer enjoyed the following benefits:
- A 37% reduction in false-positive politically exposed person (PEP) alerts from historic and ongoing donation transactions.
- A reduction of the compliance analyst workforce required to close the remaining alert backlog, freeing up their time to support the expansion to new international markets.
- An extension of the issue date from the regulator without financial penalty or increased oversight.
- Deployment of the model to other case management queues for additional Sanctions and AML-regulated products boosted the identification of false positives across other social media platforms.
- Clearly defined processes, roles, and communication channels across governance and data teams.
How Can We Help?
The experts at Point B help your teams realize their full potential. Get in touch here to learn how we can take your compliance function to the next level.
RELATED INDUSTRIES
RELATED SOLUTIONS