Retail's AI rebuild: why connected agents beat point solutions on returns and service

The National Retail Federation and Appriss Retail estimated US returns at $890 billion in 2024, 16.9% of annual sales ^[1]. Baymard Institute's aggregate of published studies puts online cart abandonment at about 70% ^[2]. Neither is solved by one tool: a return involves the customer conversation, a policy check, stock, a refund and a carrier.

One customer event starts a chain of agents, each with a narrow job, and a person sees only the cases that cross a risk threshold:

Klarna reported that in its first month its AI assistant handled 2.3 million conversations, two-thirds of its customer service chats, and cut average resolution time from 11 minutes to under 2 ^[3]. The company later said it would make sure customers can always reach a person, after concerns about quality. Both halves are the lesson: automation handles volume, and a clear route to people protects the customer relationship.

• Pick one journey with a measurable cost: returns, where-is-my-order or cart recovery.
• Connect the agent to the systems that resolve it: order management, inventory, payments and carriers.
• Set the policy thresholds above which a person approves, such as refund value or fraud signals.
• Track resolution time, contact rate per order, refund leakage and customer satisfaction.

Found this useful?

Share the key finding on X.