“This survey doesn’t have the data I need for my city.”
“I’ve gone through a thorough process to come up with the range for this role, but the hiring manager is telling me that the range is way too low.”
“This is a brand new position, I am not sure which surveys I should be using to price this job.”
“I’m tired of spending all this time manipulating Excel spreadsheets”.
If you’ve been doing comp for a while, you’ve probably run into all of these issues. While the work of getting pay right is so important, there are certain aspects of comp work that can be downright frustrating. Things like merging datasets in Excel, poring through a big list of surveys to try to match the right ones to your job titles, aging data, weighing data, tweaking your job pricing model can take days on end. There’s just so much friction in the process. And even when you’ve gone through each step with care, there’s still the possibility that hiring managers don’t see eye to eye on your work.
The truth is that job evaluation and job pricing are complicated problems. And in the past, there just weren’t great solutions out there.
Fortunately, the technology that goes into compensation management platforms has gotten so much better over the years, to the point where software can automate up to 90 percent of manual work while dramatically improving how job evaluation and job pricing are done.
How Artificial Intelligence Is Being Applied to Job Evaluation and Job Pricing
Photo credit: Alex Knight
Job evaluation (also called “job matching”) is a complicated problem, and getting it right is critical for making good pay decisions. Each job has multiple compensable factors, including required skills, experiences, and other attributes. It’s tricky to determine which of these elements should impact pay the most and how much a specific factor (e.g. a skill) should impact pay.
Here at PayScale, we’ve applied artificial intelligence, machine learning and data science under the hood in our compensation and survey data management platform to tackle these problems. Using the millions of rich salary profiles we’ve acquired over the course of 10+ years, we’ve built a dynamic data model to help you price your jobs. Here’s what makes this data model useful:
- You don’t have to build it; it’s auto-generated based on your search criteria for each pricing exercise.
- This model is fitted around the job you’re trying to price, meaning it automatically takes into account data collected for attributes associated with that job.
- The model learns over time which factors most significantly impact pay and optimizes for these factors.
- The model takes the freshness of the data into consideration; it first tries to find more recent data points, but will reach back further to ensure a statistically significant number of similar data points to generate a result.
All in all, it’s a model that incorporates prior knowledge, both from the data and from compensation professionals about what factors impact pay in a job.
What’s Going on “Under the Hood”
Behind the scenes, we’ve developed a bundle of technology we’re calling “Helix”. Helix is a combination of artificial intelligence, machine learning, and data science/algorithms working in the background to help with tasks including matching your jobs to survey jobs, recommending surveys and recommending pricing for each role.
The way our system works is similar to the way that Google surfaces relevant information to you when you type in a search query. In the same way that Google learns from what millions before you have typed into the search box, and then surfaces predictive results based on what you typed and infers your intent, PayScale has learned from millions of job title searches executed on our platform to predict what job you’re looking for when you’re benchmarking a job.
Similar to Google, we use a technique known as natural language processing to deliver better matches by poring through hundreds of thousands of job descriptions, and identify the specific attributes in job descriptions that affect pay.
Here’s how we’ve applied AI to solve problems in the compensation domain.
Artificial Intelligence is simply teaching a machine to function like a human. When trying to solve complex problems such as job pricing and job matching, PayScale has discovered ways to take the human intelligence both from our customers and our staff and teach it to computers.
Machine Learning is a subset of Artificial Intelligence that allows the computer to learn from human behavior. Rather than giving explicit commands to a computer in every scenario, the computer takes in the data about the human behavior and learns to “think” and “act” like humans would in these scenarios. One application is that the software can automatically figure out which jobs are similar and then recommend to you which surveys to use in a pricing exercise.
Artificial Neural Networks is one technique we use to better predict and match against big data sets. Inspired by biology, these computational tools are suited to matching and learning.
Natural Language Processing is the idea that software can understand the intent of the user and to use natural language to predict the job someone is searching for (think of when you use “auto-complete” in Google Search). We also use natural language processing to deliver better matches by poring through hundreds of thousands of job descriptions, and identify the specific attributes in job descriptions that affect pay.
All in all, the benefits we deliver are two-fold: first, the technology automates the routine, repetitive tasks associated with comp so that you can focus your time on the more strategic elements of comp; second, the tech helps you get the right inputs to get to more accurate results.
If you would like to learn more about how PayScale’s compensation management platform can make your life easier, schedule a demo with us online.