How to know a business process is ripe for agentic AI

This could be the year agentic AI hits the big time, with many enterprises looking to find value-added use cases. A key question: Which business processes are actually suitable for agentic AI?

Business consulting firm Deloitte predicts that in 2025, 25% of companies that use generative AI will launch agentic AI pilots or proofs of concept, growing to 50% in 2027. The firm says some agentic AI applications, in some industries and for some use cases, could see actual adoption into existing workflows this year.

But not all business processes are good candidates for agentic AI and therefore are not worth the AI investment. Some market observers see an alternative — deterministic automation — continuing to dominate automation in production this year.

Here’s what CIOs should keep in mind for ensuring agentic AI pays off for a given workflow.

Business alignment, value, and risk

How can an enterprise know whether a business process is ripe for agentic AI? One factor to consider, as with any IT investment, is whether adopting agentic AI will add actual value to a process.

“A successful agentic AI strategy starts with a clear definition of what the AI agents are meant to achieve,” says Prashant Kelker, chief strategy officer and a partner at global technology research and IT advisory firm ISG. “It’s essential to align the AI’s objectives with the broader business goals.

Agentic AI needs a mission. Without a clearly defined purpose, it’s like sending a ship to sea without a destination.”

Initially, a decision on whether a business process is a good candidate to switch to agentic AI would follow the same internal processes applicable to an evaluation of whether to use any new business or technology solution, says Reiko Feaver, a partner at law firm Culhane Meadows whose practice focuses on AI.

If a cost/benefit analysis shows that agentic AI will provide what’s missing in current processes, and deliver a return on investment (ROI), then a company should move ahead with the necessary resources, including money, people, and time. This equation is further complicated by the fact that pricing for vendor agentic AI offerings may be complex and not yet fully clear.

The level of autonomy, increased resources, and complexity of agentic AI for a given process or processes also create challenges that must be considered, Feaver says. How well an enterprise can address these challenges can help determine whether processes are ready for agentic AI.

“Does the business have the initial and ongoing resources to support and continually improve the agentic AI technology, including for the infrastructure and necessary data?” Feaver says. “Does [it] have in place the compliance review and monitoring structure to initially evaluate the risks of the specific agentic AI; monitor and correct where issues arise; measure success; remain up to date on applicable law and regulation?”

In addition, can the business afford an agentic AI failure in a process, in terms of performance and compliance? “What would be the business impact if the agentic AI failed and had to be replaced with another solution to handle the relevant business process?” Feaver asks. “Can that business process be backed out easily to another solution?”

Data and actionable frameworks

Another key attribute of a good agentic AI use case is the quality of the data being used to support a process.

“To derive tangible value and ROI on agentic AI, companies need to ensure they have high-quality data,” says Saket Srivastava, CIO at work management platform provider Asana. “If the data [that agents] are acting on is outdated, isn’t meaningful, or doesn’t align with company goals, organizations aren’t going to find valuable output from these AI agents.”

AI agents also need information on who is responsible for specific tasks, what the objectives are, when actions need to happen, and how the process unfolds. “Without this actionable framework, even the most advanced AI systems will struggle to provide meaningful value,” Srivastava says.

For Asana, agentic AI plays a pivotal role in the company’s efforts to transform work management internally and for its customers. Last year, it introduced AI agents that advise on priorities, drive workflows, and take action on work, all while adapting to the unique ways individuals and teams work, Srivastava says.

Asana also recently launched Asana AI Studio, which uses agentic AI to enable teams to create no-code, AI-powered workflows. “These workflows [allow] AI agents to handle repetitive, manual tasks like triaging project requests, drafting briefs, or assigning work, significantly reducing the time teams spend on busy work,” Srivastava says.

“We’ve enabled all of our employees to leverage AI Studio for specific tasks like researching and drafting plans, ensuring that accurate translations of content or assets meet brand guidelines,” Srivastava says.

For example, Asana’s cybersecurity team has used AI Studio to help reduce alert fatigue and free up the amount of busy work the team had previously spent on triaging alerts and vulnerabilities. The IT department uses Asana AI Studio for vendor management, to support help-desk requests, and to ensure it’s meeting software and compliance management requirements.

Customer service: A target agentic AI use case

One area that might be ideal for agentic AI is customer service. Enterprises have used interactive voice response (IVR) systems and early customer service chatbots for some time to automate customer interactions, says Sheldon Montiero, chief product officer and head of generative AI at technology consulting firm Publicis Sapient. But they are rule-based and operate within fixed, predefined workflows, he says.

“IVRs rely on rigid decision trees, meaning they struggle with complex or unexpected queries, often frustrating customers who get stuck in endless loops or are forced to repeat themselves,” Montiero says.

Legacy chatbots operate on keyword matching and pre-scripted responses, Montiero says. They work well for simple, structured inquiries, such as

checking an account balance, but fail when customers phrase questions in unexpected ways, introduce multiple topics, or require contextual understanding, he says.

“Both approaches lack true adaptability and dynamic problem-solving — leading to frequent escalations to human agents and poor customer experiences,” Montiero says. Agentic AI introduces a new paradigm, shifting from rule-based automation to context-aware, self-improving, and autonomous customer service agents, he says.

“Customer service is a powerful use case because it requires resolving customer issues that can be complex and multi-step with dependencies, involving contextual understanding, understanding nuance and reasoning through customer issues, and adjustments based on changing conditions, Montiero says.

Customer service interactions involve unstructured data such as text, images, and voice, and operate in dynamic environments, requiring constant learning and real-time adaptation, Montiero says.

“Competitive advantage can be maximized by immediate, autonomous resolution, and using feedback to improve over time,” he says. “Legacy chatbots and IVRs were about automating tasks. Agentic AI is about solving problems and delivering real-time, adaptive, and personalized customer experiences.”

Agentic AI sweet spots

In addition to customer service workflows, experts see four generic process scenarios that could be provide meaningful use cases for agentic AI.

1. Elevating hybrid business processes

One scenario where agentic AI can have an impact is with business processes that already blend automated and human decision-based tasks, says Priya Iragavarapu, vice president of data science and analytics at global management and technology consulting firm AArete.

“Agentic AI is best applicable for business processes that have dual approaches of programmatic and manual tasks intertwined within a single process,” Iragavarapu says.

An example of this would be when an insurance claims processing workflow involves automated validation of structured data — such as verifying policy numbers and coverage dates — combined with manual review of unstructured documents such as medical reports or exception cases that require human interpretation.

2. Bridging and orchestrating silo-spanning workflows

Another scenario where agentic AI can be leveraged for value is when a business process spans several, siloed teams, where each team does not have visibility or access to other teams’ data or systems.

“Then it is best to build an AI agent that can be cross-trained for this cross-functional expertise and knowledge,” Iragavarapu says. An example of this is an order-to-cash process in a large organization, where the sales, finance, and logistics teams each operate in separate systems.

“The AI agent can integrate and aggregate data from all these systems, providing a unified view to identify bottlenecks, send proactive alerts about delays, and assist in reconciliation tasks,” Iragavarapu says. The agent acts as a bridge across teams to ensure smoother workflows and decision-making, she says.

When processes span multiple teams or departments and require significant coordination, they can benefit from AI’s ability to act as an orchestrator, Srivastava says. Asana’s agents can suggest optimal workflows and ensure accountability by tracking team progress. “This ensures that work aligns with goals and reduces the risk of miscommunication or missed deadlines,” he says.

3. Aggregating automation through multiple, repetitive steps

“Processes that involve routine, repetitive actions, such as data entry, task allocation, or report generation, are ripe for [agentic] AI,” Srivastava says. “These tasks often consume significant employee time but do not require deep creative or strategic thinking. Agentic AI can automate these workflows, allowing employees to focus on higher-value activities.”

ISG is using agentic AI for some components of its ISG Tango tool. “We are upgrading [the tool] to incorporate AI elements and piloting use cases in procurement, sourcing, and supplier management, which forms the bulk of our consulting services,” Kelker says. “We are fast tracking those use cases where we can go beyond traditional machine learning to acting autonomously to complete tasks and make decisions.”

Steps that are highly repetitive and follow well-defined rules are prime candidates for agentic AI, Kelker says. “For example, matching invoice processing in supplier management led to our invoice forensics offering, which ensures companies do not double-pay for services,” he says. “The rules should be clear and repetitiveness should be high. This is the ideal combination.”

4. Supplanting costly manual tasks

Yet another scenario that is ripe for agentic AI is when a business process involves a manual approach and it would be too expensive to hire workers to handle the tasks.

“An example of this is customer support operations in a rapidly growing business,” Iragavarapu says. Instead of hiring a large team to handle routine customer inquiries such as order status updates, account issues, or basic troubleshooting, an AI agent could handle a significant portion of these interactions autonomously.

“It can resolve common issues, escalate complex cases to human agents, and learn over time to improve its responses,” Iragavarapu says. “This approach reduces operational costs, improves response times, and allows human agents to focus on higher-value interactions, such as handling disputes or building customer relationships.”