EY’s ChatGPT-infused payroll managed services target low-cost EX gains

EY is leveraging its partnership with Microsoft to apply ChatGPT in the “last frontier of back-office disruption”—payroll. The plan is to use it to bring customer experience (CX) levels of satisfaction to employees’ payroll experience. Human resources leaders and CFOs should take notice of this new solution as the potential for bringing innovation to a core part of their back-office services.

Payroll is complex, making it ripe for generative AI disruption

Sheri Sullivan, Partner Global Payroll Operations Leader at EY, sees payroll as a significant managed services opportunity, bringing product and service innovation to enterprise clients rather than simply enabling traditional labor arbitrage by outsourcing to third parties.

Payroll is complex; there are employment laws, disclosures, unique employee contracts, and tax implications to consider. Moreover, complexity varies by location, industry, and business model. Further complications have arisen with an increasingly virtual workforce, often making payroll a hot mess for many firms.

The solution for many firms has been to outsource payroll to a combination of third parties to manage, transact, and triage. Unfortunately, this can introduce as much complexity as it solves. In addition, employee, legal, financial, and support data is now further distributed. The result is poor employee engagement, low innovation, and few improvements over time.

These shortcomings are an opportunity for EY. With a payroll managed services business in 159 countries, EY has leveraged its relationship with Microsoft to apply Azure OpenAI and ChatGPT to bring innovation beyond rote bot services to improve the employee experience (EX). EY’s improvements harness a mix of technical capabilities, partnerships, and domain knowledge. They focus on creating a more “human” experience by leveraging the natural language search and response popularized by ChatGPT while “educating” the AI with trusted company knowledge, policies, and records.

A ChatGPT-infused solution may be a game changer in managed services for corporate payroll

EY’s experience suggests there are three typical ways of handling payroll questions. These include

• Ignoring requests,
• Redirecting requests to local HR for human intervention, or
• Pushing support requests to the firm’s either captive or BPO shared services center.

Often, they found employees didn’t get the answer they needed in a timely or complete manner. A lack of resolution results in a low level of employee satisfaction. EY took this as a challenge to solve.

It leverages its partnership with Microsoft’s Azure OpenAI team to make EX work like CX. Together, they collaborated on how they could use OpenAI’s ChatGPT solution to answer employee questions. Sheri Sullivan explains, “We used our proprietary information, global payroll regulation data, the anatomy of a pay slip (what the abbreviations mean, links to supporting documentation, etc.) and policies, then fed the large language model.” EY populated a generative AI solution with existing trusted data; it can learn and communicate with a natural language flow. Benefits include

• Reduced overall cost-to-service: Fewer people are needed to search multiple systems to find answers.
• Improved employee experiences: Employees get to the correct answer immediately, driving resolution, not escalation.
• Reduced training time: HR delivers internal training and support resources and solves high-value problems, then feeds those back into the model.

For the proof of concept, EY used redacted policy documents to validate whether the bot could accurately answer multi-lingual inquiries.

EY found ChatGPT was fast at ingesting and excellent at triangulating between these items. Using confidence scores and local payroll specialists to check if answers were accurate, EY found that the solution achieved 63% accuracy in the first week. By the second week, accuracy was up to 90%!

Based on these results, EY believes it can set up prompt engineering custom pipelines for multiple scenarios, which may be its killer differentiator in this market. EY will be able to use lessons learned from applying LLMs to a corporation’s unstructured data, structured data, and feedback loops to make it easier for employees and leaders to use more natural language to search vast amounts of information.

The Bottom Line: Improved employee satisfaction is just the first step in harnessing ChatGPT to benefit a wide range of enterprise functions.

While this example focuses on payroll service innovation, EY is developing the muscle for a “generative bot-as-a-service” capability. For enterprises looking into what generative AI can do for them, this example shows how services partners are quickly coming up with new ways AI will quickly benefit additional services in marketing, sales, operations, finance, and more.

Firms can expect generative capabilities of a large language model, focused on trusted internal information, to boost employee productivity and satisfaction, allowing for trusted responses to be promptly actioned. Bringing a partner approach to solving common problems frees resources to focus on greater value-creating opportunities.