B2B Claims Platform: Operationalizing Machine Learning

A software company had launched a Minimum Viable Product (MVP) solution for one of its largest manufacturing customers. The solution processed claims using invoices or receipts but still required manual entry and intervention to verify claims for the builders and contractors purchasing products. This entire process often required two weeks and was dependent on a third-party provider. 

Trility’s client desired an automated workflow that used machine learning, accelerated processing speed for its customers and their buyers, and a scalable, customizable solution to roll out for additional existing and potential customers.

Another requirement was to maintain the in-production MVP solution as it was constantly and consistently processing claims.

The Trility team devised rolling two-week slots for production environments to ensure the software company maintained service to its customers. This approach allowed production environments to work in parallel – when all the claims were closed for one slot, that production environment was deprecated.

To automate and operationalize the software, Trility built a CI/CD "pipeline within a pipeline" using serverless capabilities with Azure Durable Functions. To automate the verification of claims, machine learning was utilized to extract and match data from invoices or receipts to the product catalog. The team rapidly generated reliable machine-learning results due to building CI/CD pipelines with Infrastructure as Code (IaC). This method allowed for a higher frequency of iterative testing to validate results sooner.

Trility also collected data and provided reporting for the software company to compare the machine learning results with manual ones before moving it to production. This data could continue to be monitored and tracked to show the ongoing improvements and ROI.

Trility’s client was able to roll out an improved platform that processed claims in real-time vs. the previous two-week timeframe. By operationalizing the machine-learning algorithms with CI/CD pipelines for data extraction and matching, the company reduced recurring costs and improved customer experience. 

The robust CI/CD framework automated workstreams, allowed for rapid, iterative machine-learning testing, and provided reusable patterns and modules to lower the cost of acquisition. The client was also positioned to support and mature the CI/CD framework and easily switch out other algorithms and tools based on the needs of additional customers. 

The client could easily onboard new customers with distinct, customizable components where features could be turned on or off - giving them opportunities to drive revenue.