It’s easy for intelligent software that learns from decisions made by the company to fall behind what employees are experiencing and learning. John Sumser walks us through the important issues.

Latency is the difference between the real world and what your intelligent software thinks the real world is. More formally, it is the time between stimulation and response. In software that learns, it is the amount of time it takes a machine to understand the current environment.

Most systems that tailor themselves to a customer’s culture or procedures have an embedded learning curve. The vendors who provide these services often speak of a learning period that ranges from 60 days to a year. The real distance, however, is some number of transactions.

The number of transactions (or amount of learning) required to have a system effectively reflect a customer’s present day values, processes, and policies is a little hard to pin down.

The following kinds of major happenings shift organizational behavior and create latency in your intelligent systems:

  • CEO transitions
  • leadership changes
  • mergers & acquisitions
  • stock price volatility
  • layoffs
  • rapid growth
  • life cycle transitions
  • new product launches
  • disruption
  • market pressure
  • external political changes
  • reorganizations
  • attrition rate variations
  • external economic circumstances
  • capital structure of the firm
  • new technology
  • redesigned processes
  • financial performance

Each of these factors has a significant influence on how people fit into the organization and how the organization responds to various levels of effort. Today, we are just learning how to measure and understand the impact of these variables on operations, productivity, human capital utilization, employment branding, engagement, and employee experience.

It is easy for a machine that learns from decisions made by the company to fall behind. The human beings are learning the new terrain and have their own learning curve. The machine’s learning inherently follows the path that the humans take. At each of these inflection points, the machine falls behind and takes some time to catch up. The process is anything but instantaneous.

Worse yet, the pace of change is accelerating making it impossible to really understand where the machine is on its learning curve. There are no currently observable tools for delivering a full assessment of the state of the machine. Currently, no vendor promises more than 80% or 90% accuracy in the performance of their tool.

Talla, the bot company includes an error correction process in order to guarantee performance. Google offers Chatbase (chatbase.com), a chatbot analytics tool that allows you to see the depths of the latency problem in a chatbot. There are currently no tools for examining these sorts of alignment issues in assessment, resume matching, cultural measures, or the performance of machine recommendations.

The lag between recommendations and results is not a new problem. HR has always suffered from a feedback loop that stretches over the course of a year or more. Automating that process will definitely make it more consistent. We simply don’t know whether that additional consistency is useful and have no data to prove the case either way.

Jobs are not static things. They evolve as the company learns and the technology changes. They shift when they are outsourced. They can be declared irrelevant. They get merged. They obsolesce. The people doing the jobs get smarter and do different things for better results. Any tool that views the job as a static thing is subject to significant latency problems.

The hardest to fill jobs are constantly evolving. What makes a software developer an asset today makes her a liability in a year. The requirements for the job are constantly changing. A resume matching process that utilizes last year’s job description will inherently have a latency problem.

The same logic applies anywhere the machine is covering something that changes from employee policies to benefits choices from managerial decision making to recruiting automation. Managers who take on digital employees that learn will have to be able to measure, articulate and repair the latency problems those tools exhibit.