Missing Insights

Our team was tasked with finding predictors of employee engagement driving company success where company success is defined as an engaged workforce.

We were given four datasets for analysis, each client representative of a specific industry: Restaurant, Retail, Banking, and Construction. In good news, each data set had the same structure; however, in not so good news, each data set we received consisted of over 1300 tables. This was a challenging task for our team but one we embraced.

The overall objective of this project was to:

1. Analyze and create new metrics that correlate and predict a company’s success based on their employee’s job satisfaction and engagement.
   

2. Develop relevant benchmarks and build out a sample dashboard to visualize the metrics and benchmarks.
   

3. Determine factors that could impact what metrics are relevant or where the benchmarks should fall based on client demographics, such as size or industry type, as well as identifying any key factors that could influence positive changes in job satisfaction and engagement.

Analyze & Development

We started our analysis by examining the tables to determine what data we had to work with but quickly realized this approach would need to be revisited for the following reasons: having over 1300 tables to analyze, finding most had limited records, and documentation was tough to follow.

Our next approach was to work backwards, starting with the 3 objectives Vibe HCM outlined. In doing so, we designed our dashboards and then went and found the data. There were 3 different workstreams that needed to be managed:

1. Data analysis in Oracle: We started by mocking up 3 dashboards: New Hires, Active, and Turnover in Figma, an online collaborative design tool. From there we could determine the categories of data we needed to find in the database. The main data would be hire dates, termination dates, promotion data, diversity, gender, zip codes, work location, and more.
   
   
   

2. Predicting churn in R: Several steps were taken here to account for data quality in R before modeling could begin. The following factors were being used in our model: salary, years of service, department, age, employee status, total roles, gender, marital status, location, distance to work, and employment type.

   • Null values:  Logistic modeling does not handle null values; thus, we drop any rows where factors have null values.
     

   • Outliers: Within numeric data, outliers are dealt with by looking at the 3rd quantile and filtering out anything that is 1.5 times more than the 3rd quantile.
     

   • Segmentation: Data is segmented by hourly and salary employees.
     
     

     We used basic logistical modeling using the factors that are most significant to the segment we modeled. There are three main segments per customer we are looking at.

     1. Overall Model, inclusive of hourly and salaried individuals together.

     2. Hourly Model, this is filtered on employment code = H

     3. Salaried Model, this is filtered on employment code = S

1. KPI’s in Tableau: KPI’s for “Roles by Diversity” and “Headcount by Month” filled the top of each dashboard.

   • New Hires

     • Top words used in Glassdoor reviews

     • Average days to Hire
       
       

   • Active

     • Top Salaries by Gender, Ethnicity, and Performance

     • Top Locations most and least likely to Churn
       

   • Turnover

     • Top Turnover by Supervisor and Location

     • Length of tenure
       

   • Employee Details

     • If a user clicked on any of the graphs within the dashboards, the employee details with name, title, salary type, supervisor, location, annual pay and more would display.
       
       

2. Extra content: We thought it would be a good idea to show the reviews on Glassdoor for each company Vibe HCM provided and this aggregated information was available on the landing page.

Insights & Dashboard

Below is our overall data flow completed for the project.

In addition to creating a functional data flow, there were several insights that were uncovered in the end. Below are the highlights from each dashboard and the R modeling.

New Hire Dashboard

• Current month (July) new hire is 178, increased by 58% from prior month. 

• Most new hires occurred in January and February this year, before the covid-19 shutdown, and there were nearly 0 new hires in April and May, but it started to climb slowly back up from June and July likely due to the reopening of public locations.

• Most new hires have been Male Hispanic cooks, Female White servers, and Male Hispanic dishwashers this year.

Active Dashboard

• There are zero women in C-Suite roles such as CEO, Executive Vice-President, and Chief Officer. 

• Top leadership positions are exclusively white.

• Managerial roles with poor performance have higher salary packages than their counterparts with better performance ratings, which is not a good practice because there is no monetary reward to incentivize hard-working employees. 

Turnover Dashboard

• The #1 reason why employees left is they accepted another job (45.3%).

• The #2 reason why employees left is they abandoned their jobs (19.2%).

• Most employees left during the 2–6-month time frame after they were hired.

• There were 7 supervisors that had more than 37 turnovers in the current year. It could be beneficial for the client to pay attention to them. 

R Model Insights

• For every 1-year increase in years of service, the probability a person leaves decreases by 42.51%. The client could then take this insight to heart and know that effort put into retention will eventually pay off. 

• A 1-mile increase in distance from work zip code to home zip code increases the probability an employee leaves the company by 1.7%.