Thought Leaders in Artificial Intelligence: Stuart Nisbet, Chief Data Scientist, Cadient Talent (Part 3)

Sramana Mitra: Going back to my question about the dataset upon which you are applying your algorithm, are you working on résumés? Where are you getting all these data from? 

Stuart Nisbet: If we went out to apply for a job at a local retailer, they use an online system where they create a profile, enter their name, address, email, and other information.

On that profile, they give information regarding their work experience and educational experience. There are also assessments that you take during the process. All of that data is being captured. To put it simply, it’s an Oracle table. Oracle is our backend store for that, but there are dozens of tables that are pulled in that have information that we need.

These are data that are collected in the application process and if the manager does an in-person interview, there is additional information that is brought in. That’s what is delivered to the managers for them to be able to sift through.

In some jobs, especially for specific positions or requisition jobs, a lot of the hiring that we do is focused on hourly hiring. For example, we have a cashier position and we’ve got 40 of them. We are always hiring for a cashier, so it’s not for one specific requisition; it’s for an hourly job where we assume that there’s going to be a turnover.

We assume that we always need cashiers for retailers, gas stations, or fast-food restaurants. That is the nature of the data that is gathered and that is what’s moved into models to predict the next best hire.

Sramana Mitra: What did you learn based on applying this AI algorithm? Give me examples of the types of insights that you have managed to gather about what makes a good hourly cashier fit.

Stuart Nisbet: That is a great question and that is the core of what it is that we want to do. What’s interesting is, it’s not one variable. There’s not one thing that you can choose in terms of educational achievements, work experience, or where they have been.

In many cases, what we are finding through these models is that it is a combination of a certain set of variables. One that we are very interested in. If I am applying for a job at Sears, for example, and my work zip code has a certain value, we geocode alongside your home zip code and get a commute distance.

It is fascinating to see what impact it has. If I am applying to a Sears, Target, Kmart or Walmart and so forth and the one I’m going to for a $10 to $15-hour job is twenty miles away, we can see that you are going to drive by a half dozen retailers that pay similarly and offer similar benefits.

Your likelihood of staying at a job at a $12 to $15 rate at a distance of 20 miles is far less, and that varies by the region that you are in. Two miles in San Francisco is too far to commute for a job like that and 20 miles in Abilene, Texas may not be.

There are similar insights that we are seeing between education, experience, distance to work, and assessment score. All of these things build a profile that we use to determine which applicants are the most likely to be successful on the job.

Once we’ve trained the model, we test it. We hold back the 2020 data and train from 2015 to 2019, and then we show mathematically the ones that we would have chosen.

Then we compare that with the ones that have lasted long based on what we predicted. We can bring analytics, deep learning, and artificial intelligence into something that just hasn’t had it in the past.

This segment is part 3 in the series : Thought Leaders in Artificial Intelligence: Stuart Nisbet, Chief Data Scientist, Cadient Talent
1 2 3 4 5 6