What We've Learned from Agentic AI Customer Service Deployments

What We’ve Learned

Manhattan Active® Maven is our purpose-built AI chatbot and email agent, designed to help retailers automate and optimize their customer service capabilities. It’s been over a year since our first Maven deployment went live, and we want to share what we’ve learned:

Most customers who chat with a retail business have a question about their order status or shipment. In fact, 50% of the chat volumes Maven handles are about order status, ETAs, and shipments. Some of the most common questions include:

• Can I get my tracking link?
• Is my order ready for pickup yet? When will it be ready?
• What address is this order shipping to?

When we were building Maven, we had a theory: We predicted that most customers who chat with a retail business would be asking something about their most recent order. This is why Maven proactively displays the customer’s latest order when the chat starts. Our theory has proven true; 1 in 5 customers see their most recent order – with the status, ETA, and available actions like return or cancel – and they don’t need to ask any further questions.

There are still (and will always be) customers who want to speak to an agent. When we launched Maven, we let customers connect to a human agent immediately if they wanted to. The problem was that customers were skipping the AI without understanding the type of questions or issues it could resolve for them.

We made changes to:

1. Greet the customer with a few examples of what Maven can do for them.
2. Require the customer send a minimum of 2 messages to Maven before we let them connect to an agent.

After these changes, we’ve seen Maven achieve over a 50% deflection rate.

While these changes encourage customers to interact with Maven’s AI before connecting to an agent, we fully understand there will always be customers who need to talk to a human agent to receive the service they need.

Customers expect support to be there when and where they need it most. Service should adapt to the customer’s schedule, not the other way around. While we see the most chat activity in the U.S. evening hours, we see customers chat with Maven around the clock, including the wee hours of the morning. Traditional chat coverage by human agents just can’t cover that type of 24/7 schedule without hiring global service teams.

The possibilities for automation in Customer Service are endless. Manhattan Active Maven’s agentic capabilities are already robust—including replacing damaged products, adjusting prices when items go on sale, or refunding customers for lost packages—and these capabilities will continue to grow and improve. Now that we have seen AI agents successfully perform these intricate tasks for real customers, we believe the future of AI agents to provide faster, scalable customer service is bright.  Learn more about Manhattan Active Maven here.