I recently visited the iCIMS campus in Holmdel, N.J. It’s the former home of Bell Labs, birthplace of the transistor and where the foundations of the digital age were laid. I was on site for iCIMS’ annual conference, which blends analysts and customers, and I spoke on a panel about what people need to know and do to be smart about using AI in HR tech.
The company prefers not to be seen as a technical organization. It would rather let others take the early-stage risks while it harvests the market of things that work well–it doesn’t want to be seen as the inventor of new, shiny objects, but rather be known for the deep value it delivers to customers.
Read more from John Sumser here.
That’s a foundational difference. iCIMS sees itself as the buffer between innovation and the customer. The idea that innovation is inherently good passes the risk of failure directly to the customer. iCIMS prefers to watch the market closely and improve ideas that have demonstrated utility.
There’s a bundle of irony embedded in being a customer-centric operation that occupies an office building that was the heart of pre-Silicon Valley tech. iCIMS is not trying to pretend that it isn’t a software company. It’s that they’re rejecting the kind of innovation spurred by venture-capital hedge betting–the company doesn’t believe in innovation for the sake of innovation.
The ongoing market debate about bias in AI systems stems from a pretty simple set of facts. AI tools (we’re talking machine learning and natural-language processing) are always built by teams for a very specific purpose. When applied to multifaceted problems like recruiting or HR, they tend to be constrained by the limits of the team that built the tool and its original purpose.
Intelligent tools all have strengths and weaknesses. All AI tools can be better in some settings and not so good in others. They are purpose built and rarely fit a particular customer’s use case perfectly. This is one of the ways that bias creeps in.
For instance, there are a ton of search tools that claim to be able to match resumes and job descriptions; it’s an industry unto itself. Each individual search tool does some things well and other things less well. A tool that does a fantastic job making sense out of job titles, however, may not be effective at understanding the detailed skills required for a job.
Given iCIMS’ tendency to wait until a technology is proven and its cautious approach to new and shiny objects, when it does decide to put energy into filing a patent in AI, it’s worth paying attention.
iCIMS recently did just that for Ensemble Matching in job search and resume-matching processes. By knitting together three different AI-driven search tools with a voting methodology, the company believes it can:
- improve the quality of results a job-hunter sees;
- solve a greater number of job and resume-search use cases;
- improve the explainability of results; and
- manage the use of search tools cost effectively.
At the heart of the idea is a voting system that overcomes a number of the fundamental problems in the use of intelligent tools in recruiting. In particular, there is a sense in the market that a single AI search engine is not a significant improvement over older search methods.
The voting engine uses historical success, ongoing performance evaluation, job/skill category variations and usage patterns. It takes input from the search tools and discovers a single view of the best candidate or the best job.
Hang on to your chairs: 2020 is going to be dominated by the intelligent-tools story. The research is going deeper, and new techniques are being invented daily. Meanwhile, the output of all that R&D is being used to build higher-order functionality like this tool from iCIMS.
You can bet that this next layer of the adventure will be the topic of my upcoming master class in intelligent tools at Select HRTech.