"Drop rate of customers between prescription upload and order confirmation was very high."
— Business Head, Prescription medicine
The older ordering method required the customers to wait hours before their orders could be processed. This resulted in many customers dropping out between prescription upload and ordering (80-90% of the customers). The business identified that automating the following processes could result in faster turnaround time and improved customer conversion rate:
The business head approached the machine learning team to build a machine learning pipeline that could solve these problems. They wanted a solution fast since it directly affected their revenues.
"We were spending a lot of time doing things that were not our expertise."
— Lead Data Scientist
The machine learning team was getting delayed in the project's delivery. The team did lots of back and forth with the DevOps team to set up infra for new experiments, create demos, deploy model APIs, etc. They faced the challenge of making sophisticated models like OCR (Optical Character Recognition) and blur detection on prescription data. This data was noisy but required the model to be accurate and hence needed multiple experiments and iterations with state-of-the-art model architectures.
The machine learning team needed help concentrating on solving the complex machine learning problem because they were busy trying to get the model production ready. This meant an extended period of delay in realizing the business impact.
The company wanted an MLOps tool that the machine learning team could use to set up the machine learning pipeline without needing DevOps help to build, test, demo, productionize, and monitor their models.
TrueFoundry enabled the Data science team to become independent regarding their MLOps requirements. The team could act independently on things that typically required back-and-forth with the DevOps team.
"TrueFoundry has acted as a partner for the Data Science team and often went beyond their scope to ensure our team's success."
— Senior Data Scientist
The Machine Learning team uses TrueFoundry for the following:
In the development phase, the team used the TrueFoundry platform to
The team could independently, with the platform, deploy production models within an hour:
The team regularly needed feedback from the product managers and business heads, so they used the platform to:
After deploying the model, the machine learning team used the TrueFoundry platform to set up a pipeline to monitor its performance and ensure that it delivers business impact by:
Given the sensitive nature of the data and model predictions, the Machine learning team used the TrueFoundry platform to:
The team had workloads running in AWS and GCP and needed to move some models from one cloud to another. They used the multi-cloud control plane of TrueFoundry to:
"We took just 6 days instead of the expected 4 months to move our ML pipelines from AWS to GCP migration with TrueFoundry, which was amazing. We have been an early partner with TrueFoundry and have seen the product improve significantly."
— DevOps Lead
Using the ML models deployed on the TrueFoundry platform, the team was able to offer a much smoother customer experience. They automated the manual processes hence freeing up the pharmacist team's time. The project decreased customers' prescription to checkout time from 2 hours to 5 minutes.
These changes improved the conversion percentage of the customers from uploading the prescriptions by ~1 percentage point, which would have a $ 1.5 Mn topline impact for the company in the first year and potentially more going forward.