Fraud Detection at FICO with MongoDB and Microservices
FICO is more than just the FICO credit score. Founded in 1956, FICO also offers analytics applications for customer acquisition, service, and security, plus tools for decision management.
One of those applications is the Falcon Assurance Navigator (FAN), a fraud detection system that monitors purchasing and expenses through the full procurement to pay cycle. Consider an expense report: the entities involved include the reporter, the approver, the vendor, the department or business unit, the expense line items, and more. A single report has multiple line items, where each line may be broken into different expense codes, different budget sources, and so on. This translates into a complicated data model that can be nested 6 or 7 layers deep – a great match for MongoDB’s document model, but quite hard to represent in the tabular model of relational databases.
FAN Architecture Overview
The fraud detection engine consists of a series of microservices that operate on transactions in queues that are persisted in MongoDB:
- Each transaction arrives in a receiver service, which places it into a queue.
- An attachment processor service checks for an attachment; if one exists, it sends it to an OCR service and stores the transaction enriched with the OCR data.
- A context creator service analyzes it and associates it with any past transactions that are related to it.
- A decision execution engine runs the rules that have been set up by the client and identifies violations.
- One or more analytics engines review transactions and flag outliers.
- Now decorated with a score, the transaction goes to a case manager service, which decides whether to create a case for human follow-up based on any identified issues.
- At the same time, a notification manager passes updates on the processing of each transaction back to the client’s expense/procurement system.
To learn more, watch FICO’s presentation at MongoDB World 2018.
Leaf in the Wild: SilkRoute Chooses MongoDB Over SQL Server for Critical Quality Assurance Platform
Leaf in the Wild posts highlight real world MongoDB deployments. Read other stories about how companies are using MongoDB for their mission-critical projects.
MongoDB chosen for development productivity, operational efficiency with Cloud Manager, and “truly outstanding” professional services
From manufacturing to retail, every part of the supply chain is starting to see the value of data. Whether it’s developing IoT quality assurance applications in manufacturing to ensure your products are defect-free or building data-driven customer loyalty programs so that brands can connect with and reward their fans, the top companies are working to improve their approach to data.
SilkRoute Global is a software-as-a-service company focused on this industry. Its analytics products automate processes and present consumable, useful information to its customers. To understand the benefits they get from MongoDB, I spoke with Devin Duden, CTO of OmniSky (a division of SilkRoute) & Senior Software Engineer, and Amjad Hussain, CEO & Chief Data Scientist.
Tell us a little bit about SilkRoute.
SilkRoute is a passionate team of designers, machine learning scientists, and software engineers with tremendous industry knowledge of manufacturing, distribution, and retail. We live for solving big problems. Our industry-specific predictive and prescriptive analytics platform creates immense operational and strategic value for our customers. Our customer footprint is global and growing. Applied machine learning, business process automation, and mobility are woven into the fabric of everything we build. We offer a unique risk-free rapid implementation and integration approach for our customers to enjoy our solutions.
Please describe how you’re using MongoDB.
The application SilkRoute is building is a mobile application performing RFID inspections on industrial manufactured products. The application provides a centralized data store of customers’ products and the inspections associated with a product, and allows those customers to easily share the inspection records with others. MongoDB was chosen for this application based on:
- Simplified schema design
- Increased flexibility for modeling complex relationships (e.g., using MongoDB eliminated recursive relationships necessary in a SQL-based solution)
- Easier capture of user generated data
- Reduced development timeline
- Durability, scalability, and disaster recovery
What were you using before MongoDB? Was this a new project or did you migrate from a different database?
The current version of the application is a client-server implementation using SQL Server as a cloud sharing data store and Windows CE on the mobile device. The application is a rewrite.
How did you hear about MongoDB? Did you consider other alternatives, like relational or NoSQL databases?
I was introduced to MongoDB three years ago when I started working at SilkRoute. We were working on a social network at the time, which was using MongoDB as its primary data store. The RFID mobile application’s technical requirements were originally to use MS SQL Server. This technical requirement was provided by the client. During our working Joint Application Design session with the client, we suggested using MongoDB, but didn’t make headway. When we attended MongoDB World 2015, we gathered enough details about MongoDB’s capabilities, along with real-world examples of high-volume, transaction-based applications being developed on MongoDB, that we were able to persuade the client to switch from SQL Server to MongoDB.
Please describe your MongoDB deployment, technology stack, and the version of MongoDB that you are running.
The MongoDB deployment is a 5 node replica set using Cloud Manager for operational management and deployment.
The replica set is deployed in the US East AWS region across all availability zones. At this point, we have not implemented sharding. The MongoDB replica set has been deployed in AWS following MongoDB’s best practices using Amazon Linux AMIs. Each production node will be running on EC2 instances with 16 GB memory and 4 core CPUs, with three 100GB provisioned IOPS EBS volumes. Each volume is XFS format. One volume is mapped for “data”, one volume is mapped for “log”, and one volume is mapped for “journal”. The API stack is written in .NET 5 using C# MVC/Web API framework. We are using the MongoDB .NET driver version 2.0.
Are you using any tools to monitor, manage and backup your MongoDB deployment? If so what? Do you use Ops Manager / Cloud Manager?
The replica set has been deployed and managed using Cloud Manager. Cloud Manager simplified and streamlined replica set deployment and operations. This solution is the first time the majority of team members used MongoDB.
To reduce time spent with MongoDB replica set deployment and configuration, Cloud Manager was a great fit. Following Cloud Manager’s directions to create AWS EC2 instances made it very easy for us to create images, and build/tear down replica sets quickly.
Streamlining manual tasks allowed the team to focus more time on development than deploying a fully managed MongoDB replica set. In addition to Cloud Manager, the team just started using MongoDB Compass to analyze collections and document sizes.
Are you integrating MongoDB with other data analytics, BI or visualization tools? If so, can you share any details?
At this point we have not integrated any BI. One of our objectives is to connect with the client’s BI system using the MongoDB Connector for BI and/or extract data from a tagged node to hydrate a SQL-based BI system. We’re planning to perform a POC on the Connector for BI, now that it has been released.
How are you measuring the impact of MongoDB on your business?
SilkRoute measures MongoDB’s impact by many factors, including ease of use with deployments, a code first approach, increased agile development model, reduced total cost of ownership, and reduced time to market. The ease of deployments reduces or eliminates maintenance windows when spinning up a replica set or upgrading database versions, which means higher uptime for customers and less productive time eaten up for developers. A code first approach adds to increased savings by eliminating daunting DDL script management and aids with better agile development. These factors result in reduced total cost of ownership and faster time to market.
Do you use any commercial subscriptions or services to support your MongoDB deployment?
SilkRoute is a MongoDB OEM partner. For the RFID application we will be embedding MongoDB Enterprise Server 3.2 and managing the deployment with Cloud Manager.
We allocated a budget for MongoDB’s professional services in the early stages of the project. The professional services were tailored to the team’s skill set and agenda. With two separate onsite sessions, we covered topics from deployment, management, and recovery using Cloud Manager, to schema modeling and scaling. The value gained working hands-on directly with a MongoDB consulting engineer was twice the investment. During one session, we encountered a disaster recovery situation in a non-production environment. Unexpected though the situation was, I personally gained the most from the experience of working through the issue with a MongoDB expert in a very collaborative fashion. The professionalism and knowledge of our MongoDB consulting engineer was truly outstanding.
Do you have plans to use MongoDB for other applications? If so, which ones?
Yes, both internal initiatives and client initiatives. These include BI, a Warehouse Manager SaaS solution, a customer loyalty/couponing app, and client SaaS solutions, which we are not at liberty to disclose at this point. We would prefer to use MongoDB for all application and system development projects. Our preference to use MongoDB for development is based on ease of use, an emphasis on a code first approach for projects going forward, and built-in scalability and durability.
Have you upgraded to MongoDB 3.2? What most excites you about this release?
We’ve been developing the solution using MongoDB 3.0.x. We are actively migrating the database to version 3.2.1, and the production deployment will use 3.2.1. The most exciting features of MongoDB 3.2 for us are the BI connector, document validation, $lookup, and WiredTiger's in-memory option. We feel the biggest value add to our clients are the BI connector and the in-memory storage engine.
The BI connector will allow our clients’ BI environments to integrate directly with the solution we are building, eliminating the need to write ETL processes from MongoDB to a BI environment.
The in-memory storage engine will increase performance with read operations, which will reduce latency with API requests. Anything to increase overall performance is a plus.
What advice would you give someone who is considering using MongoDB for their next project?
I would highly recommend allocating a budget for MongoDB’s professional services to help with operations, deployment, and schema modeling. The value gained with their best practices approach really reduces learning curves and POC time. Coming from a SQL world, prepare ERDs and break the ERDs into schema designs. This approach will help bridge team members from a relational to a non-relational data store. Take a top-down development approach as it will uncover access patterns that may help with schema modeling.
Thank you for sharing your MongoDB experiences with us!
If you’re comparing MongoDB with relational databases, read our RDBMS to MongoDB Migration Guide to learn more.
About the Author - Eric Holzhauer
Eric is a Product Marketing Manager at MongoDB.
How We Brought the MongoDB University App to Life
Recently we announced the availability of a new mobile app for MongoDB University, available for free in the App Store for iPhone and iPad. I sat down with Shannon Bradshaw, the Director of Education at MongoDB, and Sacha Servan-Schreiber, the engineer who designed and built the application, to get some more details