Like many, I love good advice. But sometimes, I need help to get something done.
The next rev of AI — agentic AI — will move us from advice to getting stuff done. It will enable businesses that harness it to take a transformative leap forward.
But leap to what? And transform how?
Agentic AI can reduce the cost of customer support by 25-50% while dramatically improving quality and customer satisfaction because it goes beyond simple task execution. It can also autonomously resolve complex workflows and customer interactions. When applied to customer support, for example, agents don’t just respond to queries but comprehensively resolve inquiries from start to finish, reducing human intervention and increasing efficiency.
As with all new technologies, adopting agentic AI presents challenges. A company must have its workflows well-documented and deeply understood and possess a robust knowledge base on which the agentic AI can draw. And just as with generative AI, data privacy and security concerns require companies to understand the large language models (LLMs) they tap into and how information is stored and passed by them.
However, the right adoption strategy for intelligent automation can ensure success. To reap the most benefits, companies will need to do three things:
- Start in the right place
- Balance agentic AI with human expertise
- Tap into a network of agentic expertise
While it’s still early days, here’s what we’re learning as we work with clients in various industries to integrate agentic AI into their workflows and operations.
Don’t start small — start smart
Perhaps counterintuitively, the best place to start is with your highest-volume use cases. Isn’t that risky? Not if done properly. In fact, although starting with low-volume use cases might appear to reduce risk, it actually increases the risk of not seeing sufficient impact to justify the investment.
Starting with high-volume use cases offers the greatest potential return on investment (ROI), enabling a company to quickly realize significant impact, maximize efficiency gains, and demonstrate the clear value of using AI agents.
How do you mitigate the risk of starting too big? By initially implementing the agents with just 1% of the biggest use case volumes. This approach allows you to identify and fix potential issues while preparing for broader automation.
For a retail company, this might mean automating “where’s my order?” or return-processing workflows. In addition to monitoring shipments across the company’s fulfillment network, an AI agent could verify a customer’s identity, check real-time status and update the customer — even offer options if the order has been unexpectedly delayed.
For returns, an agent could check the company’s return policies, gather customer information about the return, suggest next steps, and complete appropriate associated tasks, like printing a return label, scheduling a pickup, issuing a refund, etc. The return agent could also watch for patterns of abuse and, if warranted, adjust its decisions and next steps accordingly.
After a company deploys an AI agent on a sample portion of a high-volume workflow, it must monitor workflow activity to identify where it might need adjustments. When the agent functions smoothly, the company can expand its use in pre-defined amounts until it eventually handles the entire workflow volume.
Of course, not all tasks and workflows lend themselves to total automation with agentic AI. in fact, keeping human experts connected to the overall workings of AI agents will yield the best results.
Balance AI with human expertise
As a company examines its workflows and processes for automation candidates, it will find instances best suited to human oversight or direct action. Agentic AI is an incredible, highly capable innovation, but it has limitations.
Three in particular:
AI agents, like the LLMs that support them, don’t currently possess general intelligence. They function best in narrow, well-defined areas. So, while humans might learn how to perform a particular task and abstract from that knowledge principles they then apply to different, unrelated tasks, AI currently cannot.
Then, there are workflows with extremely complex decision matrices that demand significant experience and experience-based judgment. For example, a retail company might need content for a straightforward marketing campaign. An agent can handle that — and execute the campaign.
But want to revisit a brand’s expression and promise across multiple markets? An agent wouldn’t be up to the task. It would require insight into market trends, brand perception, cultural differences across markets, and insight into how brands evoke emotions.
Finally, workflows dependent on typically “messy” human communication and emotional nuance that require distinctly human elements such as compassion best remain with humans. Think of customer service issues involving irate customers or healthcare interactions where a patient’s emotional or mental state may be at risk.
But I’m not describing a binary decision process: give this to the AI agents; everything else goes to humans. In practice, a hybrid model works best.
While there needs to be a clear delineation between AI and human roles, even when tasks need to be handled by human experts, AI should still be on hand to extend their abilities and make the most of their expertise.
Generally speaking, companies should use agentic AI for transactional, repeatable tasks and tap human expertise for high-stakes interactions, emotionally complex scenarios, and situations requiring nuanced judgment. A $50 warranty claim might be fully automated, while a $5,000 claim would most likely benefit from human emotional intelligence and brand-sensitive handling.
Tap into an agentic network
Perhaps most important, don’t try to dive into agentic AI solo. Establish a network of expert partners. Emerging agentic AI platforms can supply the technology across digital and voice channels. A systems integrator and advisor that understands customer operating environments can train agentic models for specific customer needs and then integrate them into a company’s operations.
Integrating these models into enterprise systems requires deep expertise in complex workflows and industry-specific challenges. It also requires an intricate understanding of workflow decision points and where human interaction is most needed – or beneficial, so that agentic AI is a boon to workers and team productivity.
Agentic AI offers businesses a powerful way to improve efficiency, enhance customer experiences, and drive innovation. But success isn’t about rushing in. It’s about making smart, informed choices: Starting in the right place, applying a hybrid human/AI model, and tapping into the right network.
Because with the world of AI changing so quickly, you can’t afford to go it alone.
