How ManpowerGroup and Other Companies Measure AI’s ROI

Last year, many companies’ AI experiments failed to make it into production, and as much as 95% of AI projects essentially failed to produce any value at all, according to the Massachusetts Institute of Technology’s Project NANDA research, detailed in “The Gen AI Divide: State of AI in Business 2025.”

But as organizations’ AI programs mature, more have found ways to move beyond the hype, with real-world use cases that accelerate decision-making, streamline operations and reduce manual efforts. 

Many early AI pilots were plagued by unclear strategy and poor data foundations, says Amy Machado, a senior research manager with IDC’s Content and Knowledge Discovery Strategies program. Yet many companies now prioritize clearly defined objectives and well-governed data, she says.

“As AI budgets grow, IT decision-makers are becoming more selective,” Machado says. “They don’t want AI for AI’s sake. They require proof of ROI, and to achieve that, technology vendors and partners must demonstrate tangible, use-case-specific value integrated directly into day-to-day business workflows.”

DISCOVER: Learn how to optimize your organization’s infrastructure for artificial intelligence.

How ManpowerGroup Uses AI to Make Sense of Chaotic Data

Sophie AI.Q started out as a way to unify ManpowerGroup’s labor market data, which was spread across its own systems and several thousand external sources. The platform relies on Snowflake as its database and Microsoft Azure for virtual machines, web hosting and other technical functions, and it runs thanks to a number of AI models.

Employees submit natural-language questions, in the same way they engage public large language models. For those users who are unfamiliar with the platform, Sophie AI.Q offers sample prompts as suggestions, such as “What are the top five jobs with the highest demand over the past six months?” and “Show me the hottest locations for AI-related jobs in the Southeast.”

“It’s for quite literally everyone in the company — legal, procurement, accounts payable, recruiters. We have something for everybody,” Leaming says.

Stijn Catteeuw, ManpowerGroup’s director for global innovation, says the company uses Sophie AI.Q alongside other AI tools across the organization, including Microsoft Copilot and agentic solutions, to assist with tasks such as talent matching, screening, candidate engagement and reskilling. For example, using conversational AI, candidates can offer additional context for their applications; providing details, say, on their years of experience as a forklift driver or prohibitions on heavy lifting due to a bad back.

ManpowerGroup relies on its AI screening capabilities within 10 countries where it operates, and Catteeuw readily cites the metrics that prove its value: “It saves us more than one hour a day for a recruiter. It gives us a 50% reduction in time-to-hire, and 60% of the screenings are done outside of business hours, which is also a big benefit for the candidates.”

Candidates also appreciate that the AI tools offer feedback on their applications, Catteeuw adds. While it is impossible for human recruiters to share specific recommendations with every rejected applicant, ManpowerGroup’s AI tools can point to missing skills, which may prompt candidates to upskill through a class or simply update their resumes to more accurately reflect their capabilities.

AI tools should always be seen as a way to assist humans in hiring, Catteeuw stresses, not to make the final call: “We will never allow AI to do the full end-to-end and to make the hiring decision. There is always a human review needed.”

CLICK HERE to learn more about how application modernization can help your organization increase ROI.

When Automating Accounts Payable Makes Sense

Graebel, a global workforce mobility and managed services company, relies on hundreds of suppliers around the world to provide talent movement services. Those suppliers invoice Graebel, which then pays the suppliers and packages the charges for clients whose employees are relocating. That workflow created a huge amount of manual work for Graebel’s accounts payable team before it started using AI tools, CIO Trent Krause says.

“Our accounts payable processes were incredibly fragmented and manual,” he says. “The only way we could scale was with hands and feet.”

Graebel already relied on Microsoft Dynamics 365 as its enterprise resource planning and financial system, and the company sought ways to add automation within that environment rather than acquiring new systems. Using AI features available in Microsoft Power Platform, Krause’s team automated much of Graebel’s invoicing processes. The system now reads invoice data (including dates, reference numbers, amounts and currencies), verifies data quality and loads all information into Dynamics 365. Invoices that require additional review are routed to employees for manual intervention, which are then loaded to Dynamics 365 through a secondary automated process. The new AI-fueled processes have improved the efficiency of Graebel’s accounts payable team by at least 25%, Krause says — in some cases, more.

How to Ensure AI Pilots Will Succeed

Organizations that can align AI with well-documented processes — how and where work actually gets done — typically meet with greater success, IDC’s Machado says.

“This requires early engagement, ongoing training and feedback loops, as well as visible leadership and line-of-business champions,” she advises. “Organizations that are proactive and can show people the benefits of an AI experience within the flow of their actual work tend to see far better adoption.”

Graebel now organizes its AI work around three main “pockets,” Krause says. The first involves so-called citizen developers, who are non-IT employees using commercial AI tools like Microsoft Copilot Studio to gain productivity. Another involves the company’s development team, which creates purpose-built AI solutions, such as a service order agent that automatically stages data from customer input forms in the correct enterprise systems.

DIVE DEEPER: Learn what you’ll need to build a foundation for scalable artificial intelligence. 

The third pocket is planned but has yet to go live: a customer-facing technology roadmap built on an Amazon Web Services stack, leveraging native capabilities to accelerate transferee resolution times and reduce manual intervention. The company expects to use specific AI functionality to support guided relocation recommendations, improve experience quality by adapting to individual context and relocation parameters and allow intelligent chatbot interactions to happen in real time.

The service order agent, in particular, has eliminated manual steps, resulting in smoother workflows and faster time-to-insight, Krause notes. When someone relocates, they provide Graebel with information related to home sales, shipping and storage, temporary housing and other services. The new agent interprets that information, checks data quality and reasons through which systems should receive each piece of data before it is then staged for human review, ensuring that employees have opportunities to catch any unclear information or obvious errors, such as a misplaced decimal point. 

“We still want a human in the loop,” Krause says. 

The value of AI, he says, often comes not in the form of one “silver bullet” solution that automates the work of an entire department, but rather from small efficiency gains that add up over time.

“If you’re saving 10% of somebody’s work every day, that doesn’t mean you get rid of that person,” Krause says, “but if I have 10 people doing the same thing, now I can move some of them to higher-value work.”