Been a while since I have written a blog 🙂

I am taking the opportunity as I have some really fascinating insight to share.

At EVA.ai, we are constantly seeking routes to deploy automation solutions that transform how people work, within a process.

We believe that if it can be automated THEN it should be automated.

Saying that before employing automation capabilities – one should carefully explore, design and ensure the program improves the use-case in a lean, agile & user-centric way. 

Last week EVA.ai worked on a machine-learning powered, automation solution for an enterprise with tens of thousands of employees.

 

The Goal 

Our clients wanted to develop predictive outcomes that would better facilitate their business planning and strategic decision-making. The EVA solutions would work across 4 distinct use-cases:   

  1. Tenure: Predicted days in a role, before employees exited the company. This would cover the average time spent per position, the employee and the correlation between both variables.
  2. Efficiency: Measured output, specifically how efficient are employees at selling contracts (across different price points and product ranges)  
  3. Fraud: The count of fraudulent contracts signed within a given quarter 
  4. Contract Termination:  Whether the combination of either 1-3 prior use cases led to a contract termination (client or employee)

In essence, our clients are focused on reducing churn, driving hiring efficiency and retaining their brand value by plugging their existing historic data to power predictive ML via EVA to future-proof hiring decisions. 

 

Data Exploration

Our client gave us access to their HR data set, composed of 120 variables ranging from HR-specific contexts to employee information and performance output data. Some examples:

Employees: Division, Region, City, Position, Date hired, Date fired, Date leave, Status now, Efficiency, Disciplinary, Contract Fraud…

Candidates: Current Job position, Desired salary, Ready for a business trip, Email, Employment, Schedule, Work experience, Previous Job positions, Role Responsibilities, Education, Skills, Languages…

 

The Model

I would love to delve deeper into the exploration processes, modelling techniques and processes used to get to the output, but for blog purposes, I’ll keep to the good stuff  (if you are interested in the process, pls book some time with me for a conversation). 

We discovered 3 predictive models around the 4 business goals. The model for ‘Contract Termination’ presented the most limitations based on variables submitted  (Low R2 scores indicate little predictive value for the model as a whole. However 🙂 We were able to develop models across the 3 other business goals, specifically 

  1. Tenure: We uncovered 58 predictors with significant effects and an R2 value of .43
  2. Efficiency: We uncovered 27 predictors and an R2 of .92
  3. Fraud: We uncovered  18 variables and an R2 of .59.

 

Interesting Findings

As we worked through historic data here are some insights we were found based on our clients’ historic data

  • Candidates who have “Honesty” in their behavioural characteristics (within their profile) have 1.6x higher tenure within the company than people who don’t
  • People whose previous job titles included Supervisor or Account Manager and Sales manager were 7% more efficient than their counterparts.
  • Employees displaying skills in ‘Telephone Calls’ or who had the phrase “Customer focus” within their profile evidenced 4% fraudulent behaviour than people who didn’t

 

The Roadmap 

The ultimate goal as we begin to work with this enterprise company is to facilitate a predictive algorithm that scores people based on a subset of variables available but is reinforced continuously. Specifically speaking: 

  1. Talent clusters / Long lists: EVA allows hiring teams helping to make data-driven decisions by intelligently recommending lists of candidates to interview in priority based on our clients’ historical data and candidate attribute.
  2. Two Supervised learning models that encompass variables required to predict employees potential “Tenure & Performance” or “Fraud & Contract Termination” 
  3. Psychometric tests (5-minutes long) delivered by EVABot, iterated in phases to coincide with the accuracy (validity) of the supervised learning predictors.
  4. Reinforcement Learning: An ongoing process, where EVA monitors and takes into account every decision of the Talent Acquisition / HR team members related to ‘job-fit’ and its algorithm +  the effects these decisions have on placement and HR performance. The real-time data is used to drive even more hiring accuracy.

 

EVA and Machine Learning

Our approach to building ML HCM solutions depends on our clients’ use-case. We deploy the best-in-class methodology to build custom solutions that help our clients thrive. To learn more about EVA.ai and our ML process click here