AI is everywhere — in chatbots, pilot projects, and a growing number of apps — but more often than not, ROI from AI has proven elusive.
Some of this shortfall is understandable. Any time a major new technology emerges, several years of experimentation are needed for IT vendors and businesses to hammer out the features that matter, the glitzy demos that don’t, and the combination of capabilities and user behaviors that end up really delivering value to an organization.
There’s no question that generative AI can help with tasks like generating email content for sales teams and indication of compromise (IoC) descriptions for security analysts. But overall, the game-changing financial benefits that were trumpeted in 2023 and early 2024 have often turned out to be the modest tooting of a dime-store kazoo.
Nothing hammers this point home like a now-famous study by MIT NANDA, The Gen AI Divide: State of AI in Business 2025, which found that 95% of generative AI projects have achieved zero return. Zero. Zilch. Nada.
The details in this report matter, both for understanding the problem and recognizing the solution. It turns out the issue with poor results isn’t AI per se, but which AI technologies companies are investing in and what infrastructure they have — or don’t have — in place to support them.
The authors write:
“Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95% of organizations are getting zero return. … enterprise users reported consistently positive experiences with consumer-grade tools like ChatGPT and Copilot. These systems were praised for flexibility, familiarity, and immediate utility. Yet the same users were overwhelmingly skeptical of custom or vendor-pitched AI tools, describing them as brittle, overengineered, or misaligned with actual workflows.”
Some analysts have noted that the time frame for this study was short. “What enterprise technology could possibly deliver an eye-popping return after only six months?” they ask. But the length of deployment wasn’t the biggest issue.
Here’s the key finding from the paper, which too often is overlooked in industry commentary:
“ChatGPT’s very limitations reveal the core issue behind the GenAI Divide: it forgets context, doesn’t learn, and can’t evolve. For mission-critical work, 90% of users prefer humans. The gap is structural, GenAI lacks memory and adaptability. Agentic AI, the class of systems that embeds persistent memory and iterative learning by design, directly addresses the learning gap that defines the GenAI Divide. Unlike current systems that require full context each time, agentic systems maintain persistent memory, learn from interactions, and can autonomously orchestrate complex workflows. Early enterprise experiments with customer service agents that handle complete inquiries end-to-end, financial processing agents that monitor and approve routine transactions, and sales pipeline agents that track engagement across channels demonstrate how autonomy and memory address the core gaps enterprises identify.”
Generating text and images is helpful, but it’s only so helpful if the GenAI technology can’t learn from its environment, understand the history of business transactions and interactions, and act on the detailed context in which humans and useful digital processes need to operate.
So, the issue with ROI isn’t AI per se. It’s that GenAI — as awesomely impressive as it can be — isn’t enough to deliver the benefits businesses are looking for. To win big with AI, companies need to make the leap to AI agents, which can learn over time and operate meaningfully in complex business contexts.
Beyond GenAI: Realizing ROI with AI Agents
What are AI agents? They’re autonomous or semi-autonomous AI programs that use instructions, context, data models, and operational guardrails to make decisions and take actions. Agents can think on the fly. They can operate in uncertain conditions and can apply probabilistic reasoning — just like humans can — in ambiguous circumstances. And they can automate workflows that traditional digital transformation couldn’t address, because those workflows were too variable or complex for rigid, deterministic programming.
Another important difference with agents: They can work with tools, such as APIs, databases, other AI agents, and, yes, generative AI. They call on tools as needed to do their work. And they learn as they go, becoming smarter about their business context the longer they work in it.
Because agents can maintain long-term memory, they can solve problems that GenAI tools simply haven’t been able to solve on their own. That enables them to deliver ROI that GenAI tools simply haven’t been able to achieve.
Like GenAI tools, though, AI agents need support from other IT resources to be successful. An AI agent analyzing sales activity, for example, will underperform if it doesn’t have access to all the sales data it needs, along with business rules, access to finance and logistics systems, operational guardrails and security controls, and so on.
AI activation is the work of putting the data, workflows, and automation in place to make AI agents successful. It encompasses:
- Universal Enterprise Integration: To give agents the data they depend on for doing their jobs right, IT teams need a fast, easy way to connect every application, data source, and system using low-code/no-code development and pre-built connectors. They also need the ability to process mission-critical workflows with enterprise-grade reliability across hundreds of systems.
- Trusted Data Foundation: IT teams also need a way to validate and match data against “golden records” with built-in data management capabilities, so that they can ensure agents reason with accurate, contextual information from verified enterprise sources.
- Governed API Management: IT, security, and compliance teams need a way to secure and control all APIs with role-based access, security policies, and compliance enforcement. Teams also need an easy way to define and constrain each agent’s operations, access, and behavior through secure, governed APIs.
- Low-Code/No-Code AI Agent Design: To bring the power of agentic AI into business units everywhere, employees need a way to design agents using natural-language prompts, leveraging agentic workflow templates and proven use cases. A low-code/no-code development platform makes it easy to convert business requirements into compliant and connected agents, eliminating the need for complex coding.
- Centralized Agent Governance: IT teams should be able to monitor and control all their AI agents from a single platform, no matter where the agents are built. With a centralized agent registry, anomaly detection, full observability, telemetry logs, and compliance monitoring, teams can gain organization-wide visibility and prevent security breaches.
- Agentic Workflow Orchestration: To achieve the greatest possible ROI, teams should be able to build end-to-end automated processes where AI agents autonomously collect data, reason through contextual options, enforce policies, and take secure actions across enterprise systems.
- Security and Trust: In every industry, organizations need to secure and govern agents from Day One to scale with confidence and control. They should demand audit trails, ISO/IEC 42001 compliance, built-in guardrails, and full observability into all agent activities.
Building and operating agents without data activation is a recipe for failure. Agents will lack the data to make proper decisions, or they’ll operate without adequate security controls, putting critical data at risk.
With AI activation, agents promise to deliver the business transformation and stupendous ROI that AI pundits have long been promising. AI activation helps companies realize all the benefits that agentic AI is offering.
Learn more about building, managing, and securing AI agents today.