.
.
Wilson SilvaData is the new information currency for organizations, where improved and faster access to analytics can drive decision-making by HR, which sits at the intersection of some of the most critical data concerning talent acquisition, benefits administration, employee communications and performance tracking. Leveraging artificial intelligence (AI) can bring greater power to these areas and help derive real insights, predict trends and identify anomalies that will impact the bottom line over time. But the opportunity to improve quality and efficiency is significant.
Much has been written already about how AI can assist organizations in the recruiting process. Having historical data about employees’ performance combined with having detailed requirements around the skills, knowledge and competencies needed for a role can help recruiters and hiring managers pick the candidate that is likely to be the most successful in the role.
In payroll, teams have long provided quality assurance by running specific reports looking for potential errors in the payroll run. Those reports were based on cumulative payroll knowledge over time that indicated where errors are likely to occur. With that approach, payrolls were still at risk of an issue occurring that would not be detected by the specific quality assurance reports. Additionally, the reports in many instances would provide results that were false positives and require a lot of research to determine that the potential error was not in fact an error.
With the improvement in technology and AI, there are now ways to identify payroll anomalies using algorithms that rely upon historical data and not just the expertise and judgment of payroll analysts. For example, we have undertaken this at Alight Solutions, where our intelligent assistant, Eloise, assists payroll teams in performing quality assurance and preventing and predicting costly errors in payroll.
Utilizing previous payroll payment history, algorithms calculate the average of earnings, deductions and taxes by employee. The average of each earning, deduction and tax by employee can then be compared to the earnings, deductions and taxes by employee on the payroll currently being run. Any significant variation can be identified as an anomaly that requires research and resolution.
The resolution of these anomalies can be done through business rules and machine learning. Since much of payroll is defined by rules and limits, many anomalies can be explained simply by applying business rules to clarify the difference. For example, an employee hitting any tax or deduction that has a limit set by the business would typically show up as an anomaly. However, those types of differences are easy to explain and do not require the involvement of the payroll analyst.
The remaining differences require machine learning where payroll analysts research differences and provide responses to the anomalies identified. If the research reveals an error, adjustments are required to correct the anomaly prior to completing the payroll run. For those that are resolved as an acceptable difference, an explanation must be provided that can be utilized by the algorithm to explain similar anomalies in the future. Leveraging machine learning, these explanations and classifications of anomalies are returned to the database and then used to automatically explain subsequent anomalies.
As payrolls are run, the learning performed above minimizes the number of anomalies that require payroll analyst intervention. Dashboards are created that provide the analyst with an overview of the payroll run to show how many earning, deduction and tax opportunities were normal, how many resulted in an anomaly, how many were resolved via business rules, and how many were concluded to be okay due to machine learning.
Machine learning dramatically changes how the payroll analyst will process payroll in the future. The skillset required will demand more analytical ability as opposed to transactional. More importantly, through the right algorithm and supervised learning, the expectation is that fewer and fewer anomalies are presented to the payroll analyst for review, allowing efficiencies and improved quality in running payroll.
The example of anomaly detection described above applies to the quality of running the payroll gross-to-net calculation. However, the same approach can be applied to other aspects of payroll or other HR-related business processes. For payroll, moving the anomaly detection up in the processing cycle such that errors in inputs can be identified quickly and resolved prior to running payroll. Applying an algorithm to identify anomalies to inputs received via integration is another opportunity. Utilizing historical time data submission, similar applications can be applied to identify significant anomalies with employees’ time submission. Such anomalies can alert the timekeeper or the employee directly that time has not been entered or submitted. This could result in the reduction of off-cycle payroll runs and improve the employee experience by ensuring employees are getting paid timely and accurately.
Lastly, opportunities exist in processes related to outbound interfaces. For example, a file that goes from payroll to a tax provider typically is balanced to the payroll results to ensure that it is accurate. However, AI can also be applied to determine if what is on the file is consistent or normal given the previous transmissions of that same file. This gives the payroll team additional assurance that the quality of data going to the tax system is correct, which is something that could lead to large penalties and interest if incorrect.
These are some ideas of where the payroll business can take advantage of new technology and specifically cloud software and AI. The disruption in HR business processes is obvious, but it is also accelerating, so shifting from data entry to understanding deeper end-to-end processes will be critical and give rise to new value creators within HR.
