61% of CEOs polled in a survey commissioned by IBM said their organization is actively adopting AI agents and preparing to deploy them at scale. In its recent paper, Start realizing ROI: A practical guide to agentic AI for tech leaders, IBM highlights important considerations for enterprises that want to maximize ROI from AI agents.
Here are six key takeaways:
1. Data is still a major barrier
The beauty of AI agents lies in their ability to automate processes by making decisions autonomously. But to do this, they need context, and that means having access to relevant business data.
Large language models are trained on vast amounts of data, which makes them very powerful. But that generic training data can only do so much on its own. Insights from generic data cannot provide the company-specific nuances that agents need to truly understand a business's processes, customers and challenges.
The major problem for many organizations is that important company data is not easily accessible to AI. It is unstructured and fragmented, scattered across the organization and siloed in disparate systems.
If they want to maximize ROI from agentic AI, businesses must first focus on making their data available to AI models. Without this, agents will lack the context to make accurate decisions, leading to mistakes, missed opportunities and inefficient workflows.
2. Governance does not slow things down, it adds value
As businesses scale AI across processes and departments, it is essential to have structured systems to manage, control and track what it does, and to ensure it is accountable.
AI agents can make autonomous decisions, completing multiple actions in workflows that span several business functions, adapting plans in real time to achieve the goals they have been set.
No wonder that the paper's authors say governance is the "critical foundation of great AI" not just something you tick off for compliance purposes.
IBM's research suggests good governance can help enterprises develop AI faster, launch initiatives sooner and reduce costly rework. And companies polled by the company attribute more than a quarter (27%) of their AI efficiency gains to strong governance.
Forward-thinking leaders are using governance as a strategic tool, addressing risks before they take root and enhancing innovation. Importantly, with AI evolving continuously at pace, companies need 'adaptive governance' - systems that can continuously evolve to keep up with rapidly changing AI capabilities and risks.
3. Don't just automate tasks - transform processes
While it is easy to use AI to automate individual tasks, the biggest returns come from finding ways to transform end-to-end processes and workflows.
Agents are not meant to operate in silos. Focusing on using agents to streamline one off tasks in separate parts of the enterprise, spreads AI resources too thinly. It will likely produce marginal gains at best.
For agentic AI to be truly transformative, agents need to be working together and directed at core workflows - where they can make the greatest impact.
In practice, this means teams of agents, pulling together - often operating in hierarchical systems - to tackle complex multi-step processes that span multiple business functions. This is where the real opportunity to maximize ROI lies.
4. Define what ROI means to your business
Defining clear ROI goals and KPIs is key to reaping the full rewards of agentic AI. Without this clarity and focus, it is difficult to direct AI in a meaningful way.
For many organizations, productivity improvements are the obvious measurable outcome they look for. IBM, for example, set an expectation to achieve $4.5m in productivity gains in 2025 by automating workflows and reducing manual effort.
Revenue growth is another key metric. IBM's research suggests that nearly half of AI-first organizations have achieved sustained revenue and operating profit growth attributed to AI since 2022.
There are also significant intangible benefits that AI can deliver. It can improve the employee experience by automating routine and resource demanding tasks and increasing customer satisfaction through AI-enabled self-service technology. In one IBM survey, 44% of executives pointed to improved employee experience and talent retention as a key benefit of AI.
Finally, cost reduction is oftent a key reason for deploying agentic AI. By automating core workflows and streamlining operations it can reduce overheads and help businesses do more with less.
5. Employees can make or break your AI agent success story
To make a success out of AI and agentic AI, enterprises must take employees with them in the transformation journey. If employees mistrust AI, fear it, or view it as a competitor, then even the best AI strategies will struggle to deliver.
In fact a Forbes magazine study found that 31% of employees admitted to deliberately trying to sabotage their company's AI intiatives.
This underscores why businesses must be transparent about their AI plans, involve employees in the process and act on their feedback. AI agents must be positioned as tools to empower, not replace, workers.
If organizations want to incentivize their people to embrace AI agents, they need to train them to use AI to upskill their roles and enhance the employee experience.
6. Orchestration matters
Orchestrating how AI agents communicate, share data and collaborate across workflows is essential to generating business value from AI. Without this, you end up with poor performance, fragmented automation and inefficiencies.
To avoid this, businesses need a strategic orchestration layer to coordinate AI agents and their tools across end-to-end processes, ensuring they work together effectively.
This central control hub should monitor workflows, prioritize which tasks should happen when and dynamically allocate resources in line with business goals. Importantly, it also needs to enable AI agents to share feedback and outputs so they can learn from each other and refine their decisions autonomously. This is essential if AI is going to transform processes and maximize ROI.
Agentic AI is quickly emerging out of the experimentation phase. Enterprises now want it to start deliverintg tangible value. They need to integrate agents into their core workflows, orchestrated to work in sync with one another and supported by relevant contextual data and governance that keeps pace with AI.