In August 2020, employees at Citigroup executed what should have been a normal operation for the bank: sending Revlon creditors a routine interest payment. Instead, the bank accidentally paid out more than $900 million—the full value of the principal—to lenders.
Ironically, it was a simple mistake that could have been easily avoided, what CEO Jane Fraser would go on to call a “massive unforced error.” But the bank was using antiquated software, an employee in India checked the wrong box and the employees who were supposed to review the transaction missed the error.
Citigroup spent two years in court clawing back $504 million that creditors initially refused to return. Schadenfreude aside, Citigroup’s blunder is a cautionary tale—and an argument for the growing role of automation and AI. Observers have noted it could have been avoided had the software been more advanced and opportunities for human error minimized.
Companies win or lose based on the talents, creativity and productivity of their employees, yet CHROs are increasingly being challenged to bring AI and automation to bare to improve efficiency, reduce errors and improve employee outcomes. Getting these implementations right, and ensuring they are done in a way that supports employees, is critical. “Digital transformations are happening across every type of work, in every industry, in every company,” says IBM CHRO Nickle LaMoreaux. “You need to make sure that your employees are keeping their skills up to date and that they’re evolving. This is an application around AI that is really underutilized.”
Naturally, it’s one thing to say that AI and automation are important tools for CHROs, and another thing entirely to implement them in a way that achieves its intended goals. A headlong rush to automate processes or implement AI risks wasting resources and creating unintended consequences. The poor performance of many employee tracking programs—intended to draw conclusions about employee productivity by monitoring activities such as mouse movement but in reality often fail to measure and adequately account for real world working conditions—have drawn nearly endless negative coverage in recent months.
Even a cursory look at many of these failures reveal companies that were collecting the wrong data—or assuming the data was more meaningful than it was—and then drawing poorly informed conclusions based on it. When implementing AI and automation, CHROs must “align their measurements with what they really expect of people,” says Diane Gherson former IBM CHRO and now a senior lecturer in organizational behavior at Harvard Business School. Likewise, “just because you can, doesn’t mean you should.”
Understanding what AI and automation can and cannot achieve, how to implement them correctly and what the potential consequences are, is vital for CHROs. “AI is only applicable to 5 or 10 percent of companies as it stands today,” says Joseph Quan, founder and CEO of HR data firm Knoetic. “With AI, we build predictive models and machine learning models. A common pitfall is that you cannot get a large enough sample size. For a lot of mid-market companies, it’s not practical. You’re not getting real data and outcomes because you’re running off of too small a sample size and too small a period of time.” The largest companies likely will benefit from AI, while smaller companies are often better served by focusing on data analytics and automation instead.
Nevertheless, the prevalence of AI is accelerating rapidly. “Before the pandemic we saw a lot of clients coming to us saying, ‘I want an AI solution’ but they didn’t have any of the digital infrastructure or data they needed at all,” says Dominic Richmond, network manager at Brainpool, a UK-based AI solution provider with more than 500 freelance AI engineers and data scientists in its network. “Over the course of the pandemic, we’ve seen just absolute rapid digitization. People are now in a position where they’re able to adopt AI.”
The Power of Automation
To be successful, CHROs and their organizations must be clear about what they’re trying to achieve. AI may be more powerful than what they need in many cases. “HR has a lot of very repetitive jobs,” notes Gherson. “For instance, in payroll people are reconciling one spreadsheet to another spreadsheet. It’s just the most amazingly boring job, but, you know, someone has to do it because people need to be paid correctly. Automation clearly helps with that and enables you to be so much more efficient.”
Implemented correctly, automation schemes will free people up so they can be upskilled, according to Gherson. Ideally, the employee who previously spent a large amount of time reconciling spreadsheets will instead be trained “to start seeing patterns so that then you can start fixing your policies.” (Instead of checking boxes in outdated software, they can be on the watch for mistakes in transactions, for instance.) In simple terms, automation allows employees to do more interesting work and improves the quality of operations because people are more focused on spotting patterns and resolving problems.
Beyond improving efficiency and paving the way for employee education and more fulfilling jobs, automation of rote tasks can also help ease hiring pressure. Jobs that require a great deal of repetition also tend to be high turnover, so much so that “actually you never lay anyone off,” Gherson notes. “It’s just you don’t fill all those open jobs.” Once those tasks are automated, hiring managers no longer have to worry about filling the high turnover positions.
Better Outcomes
While management teams typically look to AI for gains in efficiency or productivity, within the HR sphere it can help bring about better employee outcomes and strategic gains for the company as well. “One place where we are really significantly using artificial intelligence to help create better outcomes is in the place of skills development and learning,” says LaMoreaux.
IBM has created an internal learning platform, which has a myriad of courses on it. These courses cover an enormous range of topics and include the company’s own curriculum, outside materials and employee-generated content. “It’s really focused around not only just technical skills, but also soft skills and leadership skills,” LaMoreaux says. A Netflix-style algorithm helps the best content rise to the top. But the company’s AI takes this a step further.
Because IBM has data on the types of projects that its employees work on, in addition to more common data such as their education and other skills they have reported, it can build a more complete picture of the actual skills its individual employees have and where they either have deficits or may want to learn more. “Using AI, we’re replacing the old annual skills taxonomy assessments with what we call skills inference,” LaMoreaux says. The algorithm can then recommend a learning path, effectively creating unique lesson plans for each and every employee.
The ability to tailor on-demand training to fit individual employees’ needs not only helps the company develop the skills it needs internally, it also gives employees greater control over their own growth at the company. As priorities shift or new technologies emerge, they are empowered to learn new skills. “A lot of companies are getting better at being transparent about ‘hey, these are the areas that we’re investing in,’” says LaMoreaux, who notes that the AI-driven learning platform allows employees to seriously look at what they would need to learn to become, for instance, a blockchain engineer. At the same time, the algorithm can help ensure that employees are taking training on topics they already know.
Employees and Managers Benefit
For decades, employees—and HR itself—have had to contend with systems and databases that don’t talk to each other. Consider a new employee who is being onboarded. They must get an ID, a laptop, log-in credentials, payroll, benefits, parking—the list goes on and on. “There’s nothing more annoying than having to look at multiple sites and try to stitch it together,” Gherson says. Companies that have good data integrity and are able to create a common identity for employees can be much more agile, and it benefits the worker too.
Historically, HR has been organized around processes, and while some individual processes may have worked great, they were not necessarily well-integrated, which often created a lot of pain points for managers, forcing them to reconcile data or leaving them without access to crucial information about their reports. “Managers’ jobs have become harder and harder and harder as we pile more and more on them,” Gherson notes, with the pandemic presenting an entirely new set of challenges and rapid changes, such as the shift to the remote work.
Managing a team now often requires “many more check-ins and a great demand to spend time to really understand employees and show empathy in a way that maybe they hadn’t before,” Gherson says. Reminders such as automated nudges to check-in on employees or award recognition bonuses can ease managers’ loads while also improving retention, productivity and satisfaction within a workforce. On the flip side, managers can also be nudged to check on employees who might struggling, such as those who haven’t received recognition from their peers for collaboration. Or a system may be able to suggest employees who are at risk of leaving for a competitor and should be given raises. AI is uniquely good at noticing these sorts of patterns that people might otherwise overlook. At the same time, this capacity to recognize patterns is not the same as being able to understand the context for those patterns.
Ethical Considerations
Implementing AI isn’t like setting up a typical tech platform. While it can be very powerful, it can also have unintended consequences or take companies down the wrong path. CHROs must make sure they not only “understand what the technology is but also what they are being given by a solutions provider,” says Brainpool’s Richmond. Not only that, leaders must “have an understanding of what the ethical issues are going to be.” Instances of AI absorbing society’s ills, such as racism and sexism, from data sets are highly publicized reminders of the challenges inherent in this.
The most important thing CHROs can do when implementing AI is ensure there is “always a quite strong human element which advocates for augmentation rather than replacement,” Richmond argues. In other words, AI should be used to enhance human decision-making and efficiency, not replace it. When designing and implementing automations and AI, CHROs should assign an “ethics owner throughout the entire process.” And while some larger companies have opted for “more of an ethics committee,” Richmond believes the approach is too broad and the ethics owner should be someone involved in the day-to-day processes.
IBM’s LaMoreaux identifies five pillars of AI ethics, which organizations should try to adhere to:
- Explainability: CHROs need to be able to explain what the AI is doing and why.
- Transparency: The end user needs to know and understand why the AI is doing what it is doing.
- Robustness: AI shouldn’t be used on a small dataset. Has it been tested enough?
- Fairness: Are there tools in place to ensure bias doesn’t creep in?
- Privacy: Do you have the right to use the data that’s being used?
Ideally, people should try to think about “augmented intelligence,” LaMoreaux says. “Particularly in the HR space, it is important that companies use AI as decision-making support, not the decision-maker.”
Four Steps for AI Implementation
Knoetic’s Quan breaks down a successful AI implementation into four consecutive steps.
1. Get your data in order. While it’s not sexy, it’s impossible to successfully implement automation tools or AI without good data integrity to start with. Typically, companies “have data scattered across all these systems,” Quan says. CHROs and their AI and automation teams need to focus first on understanding and mapping all the data in the company and “unifying it on a common schema.” That way, when an employee shows up in multiple systems, they’re identified across all of them. “There an enormous amount of data engineering in connecting them together in one place,” Quan adds. (One of Knoetic’s core products is a software suite which integrates and analyzes HR data from across different programs and systems.) The end goal is to achieve data integrity and a common identity for every employee. Skipping this step and going straight to data analytics can lead to larger problems down the line.
2. Reverse engineer your problems. All too often, companies begin by saying they need an AI or analytics initiative without defining the problems they expect it to resolve. Instead, before building and integrating AI and automation tools, CHROs must engage in an internal audit with stakeholders “who know what the biggest problems are in the company which they want to solve for,” Quan says. That way, when a tool is built or implemented, it’s targeting a real problem in the right way.
3. Build up to AI. “There’s a ladder to get to AI,” Quan says. Once companies have identified which perceived problems they are hoping to solve, they should begin with descriptive analytics first, which will help establish a baseline of how the company is really doing. This step can often be surprisingly enlightening, especially in HR departments which have lagged in terms of data collection and analysis because of poor data integrity. “Many CPOs don’t know what their attrition was in the last quarter or how many employees they have or who their best hiring managers are,” Quan adds. “It sounds so boring, but it’s so important.” Even CHROs who believe they have these fundamentals often don’t, or at least not with a high degree of accuracy.
4. Full AI implementation. If the company is large enough and has a high level of data integrity and good baselines, AI implementation can yield impressive results, helping to guide CHROs and their teams in everything from hiring to providing raises and educational opportunities. However, AI will only be successful if it’s working from a strong foundation, and it’s important to build that foundation slowly and carefully. “Do not sign up for an 18-month data integrity project,” Quan says. “Sign up for a three-month project with a small section of data where you see some milestones and analytics come out of it.” Starting small will help prove the team can notch monthly and quarterly wins rather than getting bogged down in a massive implementation.