Companies of all sizes and across industries are going all in on AI agents as a means to improve their businesses. The proof is in the numbers: Enterprise spending on generative AI is expected to hit $644 billion by the end of 2025, up 76% from 2024. What’s more, 96% of enterprise IT leaders say they plan to expand their use of AI agents in the coming year.
In many cases, these agents will replace humans as the first line of contact for users—whether they’re customers seeking goods and services or employees seeking help doing their work. For example: Gartner predicts that in replacing human labor, AI agents will reduce contact center spending by $80 billion in 2026.
But while AI agents are set to revolutionize how businesses operate, getting them right comes with unique challenges. Here are three hurdles companies will likely encounter on the journey to building and managing successful agents.
1. Your agents are only as effective as the data they’re trained on
IT and ops teams are building agents for everything from process automation to customer support to security monitoring. But no matter their specific purpose, for agents to serve users and businesses effectively, they need to operate based on how users actually behave. That means training agents on a holistic, comprehensive dataset of user behavior.
While the agentic age is still in its beginnings, we’ve already seen several notable examples of what happens when AI agents and bots operate on messy, incomplete, or otherwise unreliable data. In 2024, for instance, an Air Canada AI agent gave a customer incorrect information about the company’s bereavement policy, and the airline was found liable for the bot’s behavior and forced to reimburse the customer.
2. Your agents need to work in compliance with relevant policies and regulations
The Air Canada example also shines a light on another related challenge IT and ops teams face: Keeping agent behavior in compliance with company polices, regulatory guidelines, and ethical frameworks.
The speed and autonomy with which agents operate create complexities in governance and compliance. Without the right data governance frameworks in place, agents may inadvertently access or share sensitive data, breaching privacy regulations and creating reputational risk for businesses.
In order to head off these and other unwanted outcomes, IT teams need analytics to monitor and assess agentic behavior. Is the agent performing as intended? In the course of its training and continued operations, is it beginning to “hallucinate” or take aberrant actions? Having a real-time, end-to-end view into how agents perform is critical to optimizing their performance and the outcomes they drive.
3. Your agents need to generate meaningful ROI
Beyond monitoring compliance, businesses need analytical insights about agents in order to justify their investments. With the global AI market expected to reach $632 billion in three years (up from $250 billion today), IT teams can expect increased pressure to deliver concrete results in the face of growing AI budgets.
With the right agent analytics solution, teams can have a clear line of sight into whether and how agents are moving the needle for the business. They can answer key questions like:
- Are agents saving users time on key workflows or processes?
- Are they cutting costs and/or helping increase revenues?
- Do they free up teams to work on high-value tasks?
- Are they increasing outputs (task completions, conversions, etc.)?
Companies are understandably bullish about what AI portends for the future. With a clear line of sight into how agents behave and the ROI they generate, IT and ops teams can take the right actions to continuously improve them. This, more than anything, will be what differentiates the most successful agentic motions from the rest.
Learn about how the Pendo platform helps optimize all your software and the AI agents that go with it here.
