The use of artificial intelligence and machine learning (AI/ML) has exploded just as managers and executives are adapting to hybrid work environments and struggling to oversee remote employees.
A recent report from research firm IDC predicts that by 2024, 80 per cent of Global 2000 companies will use AI/ML-enabled "Digital Managers" to hire, fire, and train workers in jobs measured by continuous improvement — but only one in five companies will get any real value from the move without human engagement.
The ongoing worldwide COVID-19 pandemic has forced many companies to adopt new work models, ranging from fully remote to "hybrid" approaches where individuals and teams work at or across multiple digital and physical locations.
This shift to more distributed workforces has required companies to devise new methods for managing, leading, and organising enterprises, according to IDC.
In fact, about 41 per cent of companies see the ability to manage a remote and hybrid workforce as a critical skill to hire or develop in-house, according to IDC's April 2021 Future Enterprise Resiliency and Spending Survey.
Today, digital-management software based on AI/ML is used to scan resumes and cull applicants, determine daily employee performance, recommend additional training, and determine when and how many employees are needed for a job – especially for shift-type work.
Amy Loomis, a research director for IDC's worldwide Future of Work market research service, said that while the use of AI/ML to hire and fire workers might seem remarkable, “it’s rather widely used in [human resources] circles today to greater or lesser degree.
“Algorithms are often used to stack-rank employees offering recommendations on who would be best fit to hire or targeted to fire,” Loomis said.
Stack-ranking, also known as forced ranking or forced distribution, uses a statistics-based approach to rate employees on job performance in comparison to other team members.
Stack-ranking software can be used to suggest that employees take additional training, push managers to perform employee remediation, or in some cases prompt the firing of a percentage of employees who fall below performance thresholds. A company, for instance, could choose to fire all employees who fall into the bottom 10 per cent of performers.
Amazon comes under fire
For example, widespread media reports during the past year claimed Amazon uses software algorithms or “bots” to hire and rate employees, “firing millions of people with little or no human oversight.”
Overall, a large percentage of the Amazon workers are terminated for job abandonment. Only a small percentage are terminated for performance issues, according to Kelly Nantel, an Amazon spokesperson.
The company, which employs more than 1.4 million workers, denied its algorithms are used solely to fire workers. The company's workforce management technology supports and enhances the experience of job candidates and employees. It’s not meant to replace managers, but to aid their decision-making with data and information, according to Nantel.
"There’s a distinct difference between a personnel management system flagging someone who has abandoned their jobs — and as a result they’re automatically terminated — versus our performance systems that help give feedback to our managers on where and how our employees are performing and stacking up against one another and giving recommendations and feedback to those who may be struggling," Nantel said.
"Contextually, it’s easy to say thousands or hundreds of thousands are fired by robots. Well, in some cases that’s true in job abandonment cases, but they’re not fired for performance issues ever," Nantel continued. "They’re not coached, fired or disciplined through any technology."
Shannon Kalvar, research manager for IDC's IT Service Management and Client Virtualisation Program, said that while companies may not rely entirely on software bots to fire employees, recommendations based on AI/ML weigh heavily in decision making.
“We are human beings who are overworked and over stressed. What is the likelihood you’re going to disagree with a suggestion when it comes through — especially if you’re remote managing somebody?” Kalvar said.
Digital management software was already in use before the pandemic, when it mainly helped manage trucking fleets, retail workers, service workers, and other “task oriented” jobs. For example, the gig economy enabled flexible hours for delivery services, which enabled same-day delivery for retail products and groceries. Delivery trucks were no longer pre-packed days in advance.
In 2015, for example, Amazon started its gig-style Flex delivery service using contract drivers instead of full-time employees. Contract workers’ performance is closely monitored by software algorithms that track their routes and delivery times.
“A frighteningly large number of organisations have digital managers,” Kalvar said. “We’ve seen a huge uptake in interest in that and it’s already starting to roll out for office workers in addition to everyone else. Today, it’s really a problem in task-oriented jobs, but you have to realise we’re all moving into task-oriented jobs.
“There’s plenty of software that detects problems with process, which is another way of saying, 'Where are people screwing up and do they need to be remediated?'” Kalvar said.
The issue has become a hot-button one in Europe, where the European Commission is eyeing rules that could force companies to be more transparent about their use of algorithmic management.
One major flaw with algorithmic employee management is the disparate nature of applications. Some tools are embedded in ERP system software, others are standalone applications and services. In a large enterprise, there can be many different personnel management and training applications, and many of them do not talk with each other.
That's a problem at Amazon, which uses various types of software and algorithms. Some track employee time and attendance, others oversee worker performance, while still others keep a record of employee disability leave.
A manual patch the company deployed to enable communications between its time and attendance monitoring algorithm and its employee-leave system failed to integrate the two systems.
"In some cases there have been issues where an individual might have been out on leave and two systems were not talking to each other and the system generated a form email or letter being sent out to an employee saying they’d abandoned their job when, in fact, they were out on leave," Nantel said. "We’re in the process right now of fully implementing a patch that connects those two systems together.
"We’re not unique to some of those challenges, and when you’re a company as big as Amazon and you're scaling and growing as fast as we are, we certainly have found some situations where our technology and our systems haven’t kept pace," she said.
Over the next several years, the use of AI/ML-based management software is only expected to grow. Investors and other analysts have projected that the AI software market will more than double from $150 billion to more than $500 billion in the next five years or so.
For example, IDC predicts the worldwide AI market, including software, hardware, and services, will grow from $327.5 B in 2021 to $554.3B in 2024 with a five-year compound annual growth rate (CAGR) of 17.5 per cent.
Forrester Research has taken a more conservative view, projecting the market will grow to $37 billion by 2025. Forrester explained its numbers in a report supplied to Computerworld, saying most applications add AI functions without monetising them — and the custom-built AI apps that businesses create don’t generate market revenues.
“AI is fast becoming as fundamental to software as software has become to business. As a result, AI software will increasingly be embedded into existing software products by existing software vendors,” Forrester said in a white paper published last year.
“Companies will find that it makes the most sense to acquire AI functions through these software vendors. At the end of the day, AI will be everywhere in software products, just as analytics, workflow, and data are part of those same software products."
Forrester draws a distinction between software “Build Platforms” that are general and specialise in enabling users and vendors to use AI to develop AI-infused applications, and “Buy Applications.” The latter are AI-infused software tools designed to help users improve business outcomes.
The prevalence of AI now in various platforms and apps means corporate execs leading a dispersed workforce need more than just new skills — they also need new “mental models” for understanding productivity, leadership, and the relationships between employees, managers, and enterprises, according to IDC’s Kalvar.
In other words, organisations using automated employee management software need to reevaluate their relationship with their workers.
“Today, we’re still very much stuck in an industrial-era mindset. The concept of an office as a factory is not a useful tool, though,” Kalvar said.
Autocratic styles of leadership, which emphasise the benefits of employee work to the managers — and the corporate bottom line — have to give way to human skills of oversight, IDC’s study said. Without human engagement, employees lack a sense of corporate community and don't feel invested in the outcome of their work.
And without human oversight, companies risk losing out on qualified candidates because resume-scanning algorithms are often not set up properly and, thus, cull resumes of potential hires. Additionally, employees who've been fired by a bot — even if they were terminated without a good reason — are rarely rehired, Kalvar said.
“This happens in lower-wage jobs, especially,” Kalvar said. “If you’ve fired all of the qualified people because you didn’t want to keep them around, you’re done, because most companies will not rehire fired employees. The people are still there to hire, but you can’t find them. They’re invisible to you.”
If, for example, a company requires applicants with a college education, even if their work experience qualifies them for the job, algorithms will automatically cut them from a prospective list of applicants.
“If you have a region with low college attainment and you put all your jobs to require college through your filtering software, you’re going to blow through the existing candidates pretty quick,” Kalvar said. “There may be 30 per cent of the population who could be considered [for an opening], but you’re not going to see them.
“That creates a perception of talent imbalance and shortage,” he added.
Companies are already beginning to take note and change their management culture. For example, at SoftBank (a Japanese financial institution) humans review resumes rejected by AI/ML to ensure promising candidates are not overlooked.
“Honestly, there aren’t any best practices yet. I’d argue figuring this out is the big challenge for humans who manage,” Kalvar said. “We need to figure this out fast. Those who figure this out will have highly engaged, highly loyal communities working together. Those who fail will run highly lean organisations. That’s going to look really good for a couple of quarters.”
But without human intervention, those putative gains might fade away.