Team members:
David Rickheim, Sr HR Data Scientist, with:
- Rena Rasch
- Larry Sundberg
- Melissa Sharpe
Country: United States
Organization: Medtronic
Medtronic is a healthcare technology and manufacturing company with 100,000 employees around the globe. Some of our products include cardiovascular pacemakers for those with brady- or tachycardia, implantable defibrillators, portable insulin pumps for those with diabetes, and many more.
Awards Categories:
- Best Acceleration Use Case
- Best Approach for Building Trust in AI
- Best ROI Story
Business Challenge:
In the fallout of the great resignation and with the increasingly competitive Talent Acquisition landscape, our Predictive and Advanced People Analytics (PAPA) team has increased its focus on predicting employee retention in a variety of applications.
Cost of turnover, referred to as employee replacement costs, can vary significantly by position and geography but has been estimated to be between 50% and 200% of someone's salary (Gallup, Payactive, Gartner). Additionally, departing employees often cause contagion effects within their group or, when highly valued, within their greater organization. Being able to predict turnover before it occurs and notifying relevant Talent Management specialists so that they may respond to it could save significant costs and boost the employee experience at Medtronic on the whole.
Medtronic is a healthcare technology and manufacturing company with 100,000 employees around the globe. On an annual basis, as a part of our Organizational Talent Planning process, managers assign risk retention ratings to their employees (low, medium, high) with the assistance of Talent Management consultants and specialists. This is performed at the manager level up through the CEO.
By doing so, key talent, one of the most important assets our company has, can be supported through career growth and developmental opportunities. In some cases, this may also involve retention bonuses or workload changes. Historically, managers are far from perfect in their identification of high risk employees, creating an opportunity for PAPA to assist with this retention risk assignment.
We decided to create aggregate level insights regarding the main drivers of turnover for US-based employees at Medtronic, along with individualized lists of US-based employee retention risk ratings *to assist* managers with retention risk assignments. To ensure proper understanding of what was being delivered and the nuance behind how the insights were created, many trainings were held with our business partners and discussion guides and FAQ sheets were distributed. To pare back the scope of the project, which had the potential to become very broad, we decided to focus our tool for employees at the Senior Manager level and above.
Business Solution:
Dataiku, with its seamless integration with our Human Resources Data Warehouse located in Snowflake, made the process of importing and munging data much easier. Many different tables were used as data sources, including Workday history (human capital management software), salaries, one-time bonuses and payouts, performance ratings, and aggregated Organization Health Survey (OHS) results (grouped by a minimum of 5 respondents for data privacy and ethical reasons).
Many features could be created with this data, including performance trends, OHS response trends, and manager history. Over 100 different recipes were then created to combine and transform the data into the final data table. Models were trained and evaluated for the performance when predicting employee attrition within the following 12 months - another simple task with Dataiku.
With Dataiku’s intuitive GUI, commenting functionality, and timeline history, quickly editing, deciphering, and explaining steps within the sequence of data manipulations to other PAPA team members and others outside of PAPA for review was easy. Compared to the time required to format SQL scripts and Python Notebooks into easily sharable documents, Dataiku required less than half the time to format the workflow into a sharable structure. Fairness metrics also came into play when needing to evaluate and demonstrate to ethics exports that the final models used were objective and generally without bias towards protected classes.
In the end, thousands of predictions of risk retention ratings were delivered to 20 HR business partners and talent management specialists, augmenting and assisting the talent planning work that they conduct on a semi-annual basis. Roughly 200 US-based employees were identified by the model as a much higher risk rating than what they were previously assigned by their manager, potentially saving Medtronic millions of dollars in avoided turnover.
Day-to-day Change:
While our insights are planned to be provided on a set cadence rather than a continuous basis, improving the employee experience by properly identifying teams, business units, and individuals that are at high risk of turnover through our Organization Talent Process is of the utmost important within our group.
We have the ability to make teams work more effectively, for managers to be more confident in their retention and recognition efforts, and for employees to be more satisfied with their jobs.
Some praise that we have received in response to the delivery of our work includes:
“I believe in this tool, and I think it’s important and I’d like us to be able to use it more widely and more successfully”
“The power behind [the HCI tool is] identifying people that may have been higher retention risk than we thought they were, and others that might have been lower that what the leader may have interpreted”
“There were some things that made you think differently or… expanded our horizons on what might trigger [turnover] as well as just knowing that things are going to continue to be dynamic and change. And it reiterated the importance of staying close to your employee”
We plan to continue to improve our data workflow and model performance in the future as we continue to support our Talent Management teams throughout their talent planning work.
Business Area Enhanced: Human Resources
Use Case Stage: Built & Functional
Value Generated:
By identifying roughly 200 US-based employees who have much higher retention risk assignments than what their respective managers have issued, we have the potential to save millions of dollars this fiscal year. While it is too early to evaluate the impact of our work through statistical means, anecdotal feedback has shown than our insights have been proving valuable in the assignment of employee retention risk ratings.
Value Brought by Dataiku:
I think that I'm at the point that I've used every feature that Dataiku has to offer, from quick dashboard creation to showcase some statistical analysis, to fairness metrics to test for potential biases regarding protected classes. None of this work would've been accomplished in the time that we were able to do it if we weren't working with Dataiku.
Other important features:
- Quickly training machine learning models (likely 2x-3x as fast as traditional Pythonic methods)
- Evaluating model performance using a variety of metrics
- Retaining model performance history across a project (something that would be incredibly difficult with Python at the scale that we are using Dataiku)
- Integrating with Snowflake databases (computation time roughly 4x+ faster than running on local machines)
- Interacting on the Dataiku web forums when assistance is needed
Additionally, working collaboratively with my fellow Data Scientists for review and feedback purposes and with my fellow Data Engineer for constructing the data workstream were made easier with Dataiku’s easily navigated UI, timeline history, and comment functionality.
Value Type: Improve customer/employee satisfaction
Value Range: Tens of Millions of $
