Case Study

Medical Manufacturer Gets Better Predictive Insights With Machine Learning

The Customer’s Challenge

The human capital management team at a major Midwest medical device manufacturer needed a more robust data analytics solution as they looked for ways to accurately predict manager performance. Once they landed on Dataiku, they needed a dedicated partner to both prove out the value and train their team on how to build machine learning algorithms on the platform.

phData’s Solution

As a Dataiku partner, phData (and its expert data engineering team) was well-positioned to get the client up and running on Dataiku, but also train the team on Dataiku for a specific data science project. The project used production data that had been anonymized so that any resulting insights or models could be directly applied to the business.  A two-week setup process followed by weekly training sessions equipped the client’s data scientists with what they needed for machine learning-backed decisions.

The Full Story

While the client knew that they would need a machine learning (ML) platform like Dataiku to mature their predictive and analytical capabilities, they also wanted to show the value of the project and get their team up to speed on how to use the platform. 

With that in mind, we launched this pilot project to prove that Dataiku will be beneficial for the organization as a whole and to train a small cohort on using ML for performance management. 

What better way to show ROI than to build out and train on a useful model right off the bat?

phData delivered a complete Dataiku project demonstrating end-to-end implementation of a machine learning solution used for forecasting manager performance while training their data scientists and enabling them to independently execute similar projects. 

Why phData?

phData is a Dataiku partner with an experienced team of data scientists that have a proven track record of training companies on the platform and machine learning best practices. 

Getting Dataiku up and Running

While the phData team and the client were eager to jump into machine learning models, we first needed to get them on Dataiku itself. 

For the first two weeks, our team focused on understanding the client’s current setup, their needs within Dataiku, and getting the Dataiku instance set up. We acquired a trial license for the client (again, based on our Dataiku partnership). 

With the trial license, we set up data preparation, data visualizations, ML models, and a dashboard for displaying key insights to executives. 

To ensure clean, complete data sets for building ML models, we created a modular data workflow consisting of independent components like this: 

Using modules within the workflow allowed phData to develop intuitive, reusable components to process the company’s performance data.

This was a critical first step, as we wanted to give the client the easiest way to work on ML models. Instead of having to code things in several independent, siloed tools, they can now bring it into the Dataiku platform and build standardized, replicable models there. 

Getting the Team up and Running with Machine Learning

Once the team had the basics in place, we were able to provide weekly interactive Dataiku training sessions with the client to enable learning and provide a context for evaluating Dataiku’s effectiveness.

This was ‘real-world’ training while we built an ML model for predicting management performance. Using live, anonymized data, the team aligned their training with project milestones—in other words, they used actual developments from the prior week as source material for each session.

The workflow developed for this engagement demonstrated the process of manipulating and combining the live, anonymized data that had previously been uploaded to Dataiku; by using Dataiku’s workflow organization features, we produced a number of intuitive building blocks for processing each data source which were then combined into a comprehensive workflow similar to this one:
 

By using structured workflows on (anonymized) production data, we delivered a project directly applicable to the customer’s specific use case.

Our data science team’s weekly training sessions included: 

Results

In this pilot project, we not only set the client up on Dataiku but also set them up for success with their ML models for predicting performance. Additionally, the project that was developed and used for training fully executed an actual business use case by identifying managers that currently perform at a high level and are predicted to do so in the future. The model has the potential to save $1.8 – 3.6 million over its lifetime. Because Dataiku allows one to export and import models, the model that was built in a trial environment can be directly imported into their production environment as soon as it is ready.

We built a fully functioning model that delivers measurable business value, complete with visualizations and executive dashboards. At the same time, the training sessions for the client enabled their data scientists to complete additional end-to-end projects independently. All in all, we helped the client’s human capital management team discover the power of Dataiku—and just how easy it is to use. 

Take the next step
with phData.

Learn how phData can help solve your most challenging data analytics and machine learning problems.

Dependable data products, delivered faster.

Snowflake Onboarding Accelerator

Infrastructure-as-code Accelerator

Snowflake Account Visualization and Auditing

Operational Monitoring and Observability Accelerator

SaaS SQL Translator