In 2016, the company I was working for was acquired by a large Fortune 500 company looking for adjacent technology. I was leading HR technology, operations and workforce analytics at that time. I had developed a strong interest in the analytics space over the course of my career and pitched a new People Analytics function to the CHRO and VP Total Rewards. They liked the work we were doing at the acquired company in the People Analytics space and the vision I’d created. Therefore, they asked me to build out a new people analytics function. It seemed like a good opportunity to build from scratch and seemed like fun to boot, so I said yes and happily began building a plan. We launched the team in January 2017.
"By taking a bottom up approach to building a People Analytics function we were able to achieve results"
I took the building strategy from the bottom up, going up the value stack in our corporate parlance. We started out with a data warehouse to pull together data from the different subsidiaries and different functional systems. It was important to be able to report employee data based on both a reporting hierarchy and a costing/ functional hierarchy. The warehouse allowed us to do this. Initially it took two weeks to pull a basic headcount report, but within two months we could deliver multiple cuts of a standard headcount report in near real-time, with push reports going out weekly, serving over 500 users with a complete set of standard workforce reports. We also built out capabilities for custom ad-hoc reporting.
The next step was to start analyzing the data to provide descriptive sliceable aggregate views that are effectively visualized to highlight trends and key insights. Working with IT we chose Tableau, our corporate standard software for data visualization and analysis. We got to work, cleansing the data and ensuring we could track not just point-in-time data, but also high value historical data and transactions data. We built some basic dashboards to visualize key measures such as turnover, spans and layers, and workforce visualization, and delivered the dashboards to a pilot group initially. The feedback was good and we launched to a broader audience including senior management and executives along with HR business partners. Security was set up based on automated directory groups and hierarchies. We also provided custom ad hoc dashboarding capabilities based on business needs.
Now that we had some good descriptive reporting and analysis of quantitative structured data to show what happened it was time to pull in qualitative structured and unstructured data to start looking at why it happened. This is where I started focusing on our survey capabilities. We set up an onboarding, new hire, quality of hire and exit survey with both structured multiple-choice questions and unstructured text based questions. We pulled some questions from our engagement survey for comparison and started building a data set for analysis including trend analysis and data segmentation by various demographics including diversity, job level, job type, function, hierarchy and location.
Once we had the “what” and the “why”, it was possible to pull together a predictive model. Our first model is for turnover analysis. It uses a combination of variables from data we have collected and created to help us understand employee segments at risk. Our model is relatively new. The accuracy rate is 70% and we are continuously improving it by add more data and diverse sources of data. We found an interesting correlation between employee recognition and retention.
As we continue building our data sets, data integrity and analytics capabilities we are exploring more advanced areas of analysis. We recently undertook a large organizational network analysis that combined ONA with employee sentiment to understand business group interaction after the acquisition integration. We are also using natural language processing to do combined topic modeling on unstructured data sets such as internal employee surveys and external data sets such as Glassdoor.
By taking a bottom up approach to building a People Analytics function we were able to achieve results by delivering basics early and building on each successive step.