Amazon may have grabbed the attention of compensation managers across the country with its announcement that it would double the minimum wage it pays warehouse workers to $15 (albeit while eliminating incentive bonuses and stock grants).
But every day across the country, individual compensation decisions are made by human resources professionals. What kind of salary offer should you make to that new developer you want to hire, or that data engineer you have been courting? Your human resources department probably does some research about the going rate before they make an offer. But in today's world of analytics and algorithms, is looking up last year's salary rate for that job going to give you the granular, up-to-date data that you need?
Best known as the main company that has provided outsourced payroll services for decades now, human capital management service company ADP sees what 740,000 different companies in 140 countries pay their workers. Marc Rind, ADP's chief data scientist and vice president of product development has been leading a team that's been working to harness that insight for the good of its customers. For the last 6 years at ADP, Rind has been working with the company's DataCloud, developing analytics and dashboards from key metrics for HR professionals.
"It's not just what your overtime rate is," he told InformationWeek in an interview. "It's also about what your overtime rate looks like against peers in your industry."
Most recently, Rind said, the company has gathered enough data to begin working with predictive models and AI to leverage those technologies in its DataCloud.
ADP has done this by gaining its customers' consent to use their data to create benchmarks. But with salary and financial data of companies at stake, great care is taken to protect its security, Rind told me.
"We take it extremely seriously from the get-go," he said. Among the protections, ADP makes sure it has enough data from enough companies so that the salary information is truly anonymous. There's enough data so that no one can reverse engineer it to find out what companies reported it.
"If we don't have that minimum in there, we don't share the number," he said. So if you are an HR manager and type in some obscure job title in a small town, chances are you won't be able to get a salary figure or range for annual compensation.
Still, this data is very sensitive, and companies today are realizing just how valuable their data is. Does anyone ever hesitate to share their company compensation data?
"When we started this project, I have to admit I was worried that many organizations were not going to participate," Rind admitted to me. "But only 7% have opted out of participating, and that number is decreasing." The reason it's decreasing? If you opt out you can't get access to the benchmark data, and companies are increasingly seeing the value in getting those insights. Rind noted that in the past companies might rely on surveys to get such data, but by the time the insights became available, they were already a year old or older. ADP's data is closer to real time -- about a fiscal quarter old.
For those who have not opted out, ADP is making it increasingly easy to consume the insights.
"It's my philosophy that it's no longer about looking at a section of the software or having a feature," he said. "Intelligence no longer should be a feature. It should be in everything that we do."
ADP is doing that through insights embedded in the platform it already offers to customers, as well as insights embedded into the mobile app. Executives can access insights via the app, for instance, if they don't use the platform software every day.
If an HR manager is about to make a salary offer to a prospective employee, ADP's platform can generate an alert if the offer is too low for that job title and that geographic market. The platform may also surface insights such as identifying managerial positions that are likely to experience high levels of turnover.
"It finds the most interesting pieces of data and pushes those out to you," Rind said. "As you interact with the application, we pick up on what interests you most, and it gets smarter and smarter."
But surfacing the insights, the platform provides managers with useful data without requiring them to dig deep for it, especially if they don't have a lot of time, Rind said. And when they see something that interests them, they can dig deeper when they have time.
How They Do It
ADP's DataCloud is built with a Cloudera implementation of Hadoop in an on-premises implementation. ADP uses H2O.ai as a framework because of its strength running distributed and at scale, according to Rind. ADP also uses Spark, Scala, and Python leveraging Spark. The company has developed the front end for its application in-house and built the java-based visual analytics itself. Rind said the stack also relies on the RESTful APIs to enable embedding of the product anywhere in the product or workflow.
"Because it's not about having to have people find insights," he said. "It's about the insights served up and recommendations made depending on who you are and what you are doing."