As AI and AI agents face an ROI test, some companies are focusing on specific business areas instead of enterprise-wide AI adoption.

GFT Technology’s new report ‘ The Future of AI, Unfiltered ’ found that 90% of respondents are concerned that global investment in AI may be growing faster than the business value it can realistically deliver,

Focusing on specific business areas allows companies to better understand the risks of an AI or AI agent project, where it can be deployed, how it impacts the workforce and help to better build governance and ownership while allowing companies to measure ROI and other benefits more effectively.

This report looks into four distinct business areas for AI and AI agent deployment: legacy tech, payments and payroll, HR coaching and real estate and CPG retail operations.

How Companies Deploy AI and AI Agents to Modernize Legacy Tech

Legacy technology found across a wide range of industries, from banking to biotech to critical services like food, water and energy, limits AI and agentic deployment. GFT’s survey found that almost all of those surveyed (95%) said legacy systems cause a delay in their organization’s ability to deploy and scale AI. More than half (56%) described that delay as moderate or major.

More importantly, 84% reported canceling at least one AI pilot or project due to limitations in their legacy systems, and almost all 93% believe running AI on their organization’s legacy tech stacks would eventually trigger an enterprise-wide security crisis.

“When we talk about legacy systems, we’re not simply referring to old technology,” Marco Santos, global CEO, GFT Technologies, told me. “We’re talking about technology that has become difficult to evolve, integrate, secure, or scale at the speed that business now requires,” said Santos.

“Legacy is rarely one single system,” said Santos. “In large enterprises, it’s usually an interconnected estate across core applications, integration layers, data environments and infrastructure.”

As AI becomes more autonomous demanding conditions to run effectively, the organizations making the most progress with AI are creating more modular, composable architectures around their critical systems, rather than simply adding another AI layer on top of them, Santos said.

“The goal isn’t to modernize everything,” said Santos. “The goal is to modernize what prevents the business from scaling AI, in a secure, impactful and sustainable way.”

An effective modernization path, according to Santos, begins by identifying the business-critical use cases that are being constrained today, then mapping all the systems, dependencies and data behind them. With that map, businesses can modernize selectively, rather than trying to replace everything at once.

Regarding legacy tech cybersecurity, the fix is to modernize the architecture where necessary, simplify integrations, improve data controls and introduce security and governance directly into the way AI is deployed, Santos explained.

“The more autonomous AI becomes, the more important the architecture and governance around it become.”

Modernizing Payments and Payrolls with AI and AI Agents

Another business area where leading companies are turning to deploy AI and AI agents is payments and payrolls. While about half (46%) of workers are uncomfortable with AI in payroll, according to PayrollOrg, many companies believe AI can help solve important payroll pain points.

The 2026 global ADP survey found that 35% of 1,816 senior payroll stakeholders across 20 countries said a lack of automated processes is the leading cause of payroll inaccuracies and 29% said AI adoption is key to transforming payroll operations.

“Payments is fragmented and fragmentation is why so much of it is still done by hand,” Maria Camila Ramirez, co-founder and COO of Ontop , a financial infrastructure for global workforce management, told me.

“There is no single system to automate,” said Ramirez.

In global payments, every country, every rail and every provider is its own platform with its own login, its own format and its own rules, and none of them will standardize for a company, Ramirez explained.

The work between systems has always fallen to people but now it is exactly the work AI agents are good at, said Ramirez. “They operate across systems that do not connect, using the same screens a person uses, with no integration to build first,” said Ramirez.

To deploy agentic solutions, Ramirez said that companies should start by looking for the recurring manual work someone does every day. That person, rather than engineering teams, should build an agent to automate these tasks, said Ramirez.

“Automating a payment process means encoding its exceptions, and the exceptions are not written down anywhere, they are in the head of whoever handles them,” said Ramirez.

To safely deploy agents in payroll and agents, Ramirez said a line should be drawn at irreversibility, allowing agents to read, extract and recommend, while a person approves anything that moves money and makes every number it produces reconcile against its source. “The agent is never the system of record, only a faster way to read one,” said Ramirez.

Other considerations in agentic payments and payrolls include cybersecurity. Credentials should be kept in a vault and referenced, never written into the agent’s instructions; permissions set to read-only wherever the job allows and 2FA kept by a human, Ramirez explained.

How AI and AI Agents Drive Engagement, Wellbeing and Productivity

Employees are not just using AI to work. They are also using it for self-work-coaching. A new study from Cloverleaf, a team performance and AI coaching platform, found that when workers complain about their bosses, all AI public models tell them to quit their jobs.

What advice an AI gives workers, as they increasingly turn to AI to speak of their work-related conflicts, is something organizations really need to understand, Kirsten Moorefield, head of research at Cloverleaf Labs, told me.

Cloverleaf’s study also found that only three times, when giving 638 distinct pieces of advice, AI models coached workers towards conflict repair and towards genuinely investing in work relationships. Most of the advice the models gave in these evaluation scenarios was about winning, surviving, or managing the situation.

Wellbeing is linked to engagement, which is linked to productivity, causing $10 trillion in global losses, according to the recent Gallup State of the Global Workplace 2026 report.

Moorefield from Cloverleaf Labs, who believes that AI can be used efficiently for AI coaching and extend the capabilities of HR teams, told me companies should put guardrails inside the master prompt of every AI that they are rolling out.

“Workers need personalized, individualized support every single day at work,” said Moorfield. “AI and technology can provide that,” said Moorfield.

“HR needs to come up with new ways to meet their employees in the flow of work with individualized support for them, rather than frameworks and concepts and classes they need to go and take,” said Moorfield.

Deploying AI and AI Agents in CPG Retail Operations and Real Estate

According to the 2026 Promotion Optimization Institute survey, 61% of retail professionals surveyed agree that their companies face difficulties in executing promotions as planned, and about half (47%) said headquarters support teams do not have the necessary capabilities to support pricing, trade allocations and go-to-market strategies,

*Agents can bridge the gap between corporate trade marketing strategies and real-world physical retail execution,” said Ruslan Okhrimovych, CEO of Effie AI , an agentic retail execution platform for consumer packaged goods (CPG) brands working with brands like Pepsico, Shell and Coca-Cola, told me.

Highlighting that physical stores still dominated gross CPG sales volumes against online retail, Okhrimovych said most current AI development focuses on back-office tasks, leaving a massive untapped opportunity for field workers in physical environments.

Agentic retail moves beyond ‘systems of record’ that just gather data to ‘systems of action’ that tell workers what to do and when, said Okhrimovych.

Effie AI’s retail agent allows merchandisers to record video of a shelf while on-device AI analyzes products, prices and promotions without needing internet. The agent compares the shelf context against complex ‘playbooks’ from CPG brands to generate specific action plans.

“The system provides immediate verification, acting as an ‘AI supervisor’ to ensure tasks are completed correctly at the shelf,” said Okhrimovych. Companies like Nestle, which are using Effie AI have seen significant efficiency gains and financial benefits through early planogram adoption, including a 56% optimization in time spent by field workers, said Okhrimovych.

“The technology also drives KPI boosts in planogram and promotion compliance, often reaching 95% to 99% accuracy,” Okhrimovych added.

Okhrimovych said that deploying Effie.ai typically takes up to four weeks. Speaking of how it impacts workers, Okhrimovych said that rather than total replacement, the technology shifts human roles.

“Low-skill roles may move toward gig-economy/crowdsourced operators, while high-level employees move toward relationship building and negotiation,” Okhrimovych explained.

Another example of AI solving a specific industry problem comes from Revyse , a Bend, Oregon-based company led by CEO Bobbi Steward, which developed an AI-powered vendor intelligence platform that streamlines compliance.

By centralizing contracts, vendor credentials and spend data in one system, the company is helping operators catch costly problems like missed renewals, expired insurance and vendors paid without an active contract. The same approach translates naturally to retail, where companies managing hundreds of suppliers and store locations face the same challenges of contract sprawl, compliance tracking and hidden spend.

AI and AI Agents Business Focus Areas: Final Thoughts

“I think this is where the AI conversation has fundamentally changed,” Santos from GFT told me. “Business leaders need to stop evaluating their AI investment based on how many pilots they launch or how many of their employees have access to a model”.

“Instead, they need to measure what changes in the business,” said Santos.

Ultimately, the test is simple: If you can’t connect an AI initiative to a measurable business outcome, you should question whether it ought to be scaled, Santos said.

Focusing on specific business operations when deploying AI and AI agents allows companies to better modernize under greater control.