“Intelligent technology is being applied to a variety of problems across the HR spectrum. Since it appears to be a fundamental capability of any application, the diversity of solutions continues to grow, and that growth is accelerating.”
– John Sumser

The Technology that Powers Intelligent Software

Intelligent technology is being applied to a variety of problems across the HR spectrum. Since it appears to be a fundamental capability of any application, the diversity of solutions continues to grow, and that growth is accelerating.

It is easy to feel overwhelmed by emerging technology of any kind. The data scientists in our industry have PhDs from schools like Stanford, Berkeley, Carnegie Mellon, MIT, and the University of Toronto. A conversation with them about the operational details of their specialty is likely to make your head swim. But, a surface level understanding should be enough to evaluate a given solution as long as the fundamental question remains: “How does this create business value?”

The following is intended to give you a simplified foundation for understanding the new technologies that many in the market still call AI, but is better labeled Intelligent Software.

  1. An algorithm is a set of rules and calculations.
    A non-automated procedural decision-making process (like levels of approval based on the dollars involved) can be considered algorithms. They can get extremely complex. For instance, the model that describes the interaction of every word in a resume with every other word in the resume and all of the words that may be related to each of those words is essentially beyond comprehension at a granular level.
    Still, it is an algorithm. Hyper-complex algorithms are often coupled with other expressions of data to form models of departments, companies, talent pools, and so on.

  3. Big Data (BD)
    Companies like Amazon, Google, Apple, Facebook, Microsoft, eBay and the rest of the top 500 Websites have unimaginably huge sets of data. BD is a broad reference to the many tools used to analyze those data and discover underlying patterns. The evolution of Big Data over the past decade produced many of the tools that are currently being used to address HR problems.
    There are only a few data sets in the HR universe that qualify as Big Data: all of job postings; all resumes in the word; the total volume of employee communications at companies with more than 100K employees; employee data at organizations with more than 250,000 employees (such as the US Department of Defense–3.2M, the Chinese Military–2.3M, Walmart–2.1M, McDonald’s–1.9M, UK National Health Service–1.7M).
    In the rest of HR, the problems are somewhat different because the data sets are small by comparison. Some easy to execute techniques become slow and incomplete when there isn’t enough data.
    The larger HR Tech firms (Workday, Ultimate Software, Smart Recruiters, Kronos, IBM, Cornerstone On Demand, Ceridian) may examine data across their client base to maximize the utility of Big Data tools. They often refer to this as benchmarking, but the process is more like a search for pattern repetition.
    Big Data gathered across company lines in a client base always raises issues of confidentiality and privacy. At this point, the large players are all well versed in the nuances of these problems and have little difficulty ensuring that individual and company data is adequately isolated.

  5. Machine Learning (ML)
    At its very essence, computing is composed of loops (repeated processes/algorithms) and counters. A counter tallies the number of times the loop completes itself. Sophisticated counters and loops handle variations and report the count as a distribution.
    Once a primary analysis of a data set is complete, machine learning can go to work. The primary analysis unearths critical patterns, often prompted by a data scientist’s hypotheses. From there, machine learning tools watch the evolution of patterns by counting and measuring their distributions.
    For example, a flight risk analysis involves building a model of the patterns that are related to the likelihood that a given employee will leave (or that a specific department will experience attrition and how much). The various factors (time in grade, promotability, performance ratings, attendance, workload, overtime hours, and a host of others) are combined into a formula that scores the employee or department.
    Each time an employee leaves, the math distributions are revised by the machine to reflect what happened. Each employee who stays also exerts an influence on the machine’s view. Each new piece of data makes the machine recalculate the probability of what could happen next.
    When ML systems are installed, they have some type of baseline estimates that are revised as the system gains specific experience. All ML tools that are installed in a unique environment have a learning curve. (Be sure to read the short chapter on latency.)

  7. Natural Language Processing (NLP)
    NLP is the sophisticated processing of text (and a significant subset of machine learning). It is used in a variety of ways to try to determine different kinds of meaning in documents. Sentiment Analysis identifies and categorizes opinions and emotions expressed in a piece of text.
    NLP is also used to create significant alternatives to standard search. The use of word vectors to represent concepts allows you to submit the resume of your best engineer and have the search engine discover the best candidates like that engineer.
    NLP is the primary tool used to create machine understanding of nuance and inference. Many words have multiple meanings. By building a library of associations with other words, machines begin to learn how to understand the meaning of a given phrase in a particular context.

  9. Bots/Chatbots
    A bot (short for robot) is an application that runs a series of automated scripts. These tools are used to collect data from around the internet or within a company’s systems. Common types include spiders and web crawlers.Chatbots are the most current variation of this technology. They are used to automate highly repeated process with observable answers. Chatbots generally have text interfaces (although voice-based tools are coming rapidly).
    The narrower the task, the better these tools perform. HireMya uses chatbots to sort and process the screening and hiring of low skilled hourly workers. The interview can be reduced to a decision tree so there is no misunderstanding. More complex levels of nuance lead to higher error rates.
    Chatbots are one area where HR Technology can suffer from a shortage of transaction volume. It takes work to understand the various inferences in a dialog about HR issues. For now, complex chatbot installations in HR require a training and supervisory staff to handle high levels of error.

  11. Robotic Process Automation (RPA)
    Tools like IFTT (ifttt.com) are simple, readily available script writing technologies that allow a user to link events in a process based on If-Then statements. That’s the foundation of RPA stringing together events in a process leading to a conclusion. It is desktop based processes where there is a clear decision tree.

It’s worth saying that these offerings are much more like laboratories working on an experiment rather than typical software offerings. The data scientists who lead these projects employ these technologies in a state of permanent innovation and the end result of their work may even be a discovery that happens while they are answering a completely different question from the one they started with.

Read previous post:
Could we have made the first trip to the moon in 2020?

“The question is, could we do it all today starting from scratch? Assuming the technical solutions are all in place,...