FedEx's Chatbot Sent Dillon Thompson in Circles. That Was a Design Choice, Not a Malfunction.

FedEx's Chatbot Sent Dillon Thompson in Circles. That Was a Design Choice, Not a Malfunction.

Customer Experience
A contact center that ships a chatbot ships an objective along with it. Dillon Thompson spent four months finding out which one FedEx chose.
By Shashi Bellamkonda · July 24, 2026
59%
Consumers who find AI customer service agents frustrating, up from 54% a year earlier (AnswerConnect, 2026)
85%
Consumers who would rather talk to a person, up from 83% (AnswerConnect, 2026)
31%
Consumers who'd hang up on an AI agent rather than continue, up from 29% (AnswerConnect, 2026)
Key Takeaway
Dillon Thompson's missing e-bike turned four months of chatbot loops into a viral proxy for AI customer service failure. Four organizations, four chatbots, and not one of them handed him to a person with the authority to open an investigation. Thompson hit a system built to contain his call rather than resolve it.

FedEx told Dillon Thompson his e-bike had arrived and been signed for by someone with the initials M.M., a name that matched nobody in his building. He was standing in his kitchen with an air fryer full of chicken thighs and no bicycle in sight (WIRED).

The bike had cost him nearly $2,000.

The ordeal that followed became a widely shared account of contact-center automation. A week after WIRED ran it, Meghan McCarty Carino interviewed Thompson for Marketplace Tech, carrying the story from a single reader's account to a national radio audience. Thompson called FedEx to track the bike. He called his bank and card issuer to dispute the charge. He called his local police department to file a missing-property report. Every line put a chatbot between him and a person who could act.

Every Line Ran the Same Script

FedEx's bot ignored his requests to speak with a person. The bike company's shipper opened a claim and closed it by email, telling him the loss was FedEx's to resolve. His bank and card issuer routed him through chatbot queues that both ended with a human explaining the package was lost on FedEx's watch, so the dispute wasn't theirs either. The police department's non-emergency line took his report through a chatbot and promised a callback that arrived once, during a work meeting, with no voicemail.

Four organizations ran four different chatbots, and each one could explain why the problem belonged to someone else.

A Researcher Calls It Sludge With a New Face

Ryan Hamilton, a marketing professor at Emory University who studies consumer psychology, has a name for the friction companies build into service channels on purpose: sludge. It predates chatbots. Automation gave it a faster interface.

"AI ... has just sort of ramped up the dystopian nature of it."

Hamilton's point is not that companies deployed AI by accident. Some know exactly what they traded away. A chatbot that never escalates keeps contact volume off the phone lines, and contact volume is the line item a customer service budget gets measured against.

The Gap Was Authority, Not Intelligence

Thompson's bike claim was an exception at four different companies, and at every one of them, the routing logic treated it as routine. That distinction matters more than whether the underlying model was capable enough.

This publication made a version of this argument in June, when an airline's own mobile app caught a TSA PreCheck lapse that its chatbot could not ("The Airline App Knew. The Chatbot Didn't."). Access to operational data decided the outcome. FedEx's phone bot carried a knowledge base of shipping policy, and nothing in its design gave it the authority to hand Thompson to someone who could open an investigation on a signature that matched no one at his address.

A Smaller Vendor Is Betting the Objective Can Change

Somewhere between deflection and resolution sits the design choice every contact center leader is making, whether or not the org chart calls it that. Riptide, a San Francisco startup led by chief executive Doug Marinaro, has built its pitch around naming that choice directly. Its product, Ripley, is built to escalate by default rather than contain, pulling a customer, a dispatcher, and a human agent into one running conversation instead of a queue.

AAA - The Auto Club Group is Riptide's flagship account, running Ripley across roadside assistance for 13 million members. Riptide reports response times ten times faster and handle time cut in half against its own baseline, alongside a customer satisfaction lift and improved agent retention. Those figures come from the vendor and one customer relationship, not an independent audit, and belong in a buyer's evaluation with that context attached.

One Flagship Account Is Not a Track Record

Riptide has raised a modest seed round and employs fewer than 20 people by available counts. Its case for resolution-first design is real. Its evidence that the model holds up across an enterprise portfolio, the way Zoom's contact center business has started to show at scale, is not yet built.

Zoom crossed $100 million in contact center revenue by making the same argument Riptide is making now, that resolution rate beats handle time as the metric worth optimizing ("Zoom Crossed $100 Million in Contact Center Revenue by Selling Resolution, Not Speed"). The idea is not new. It hasn't yet survived contact with a portfolio of Fortune 500 accounts the way it has with one auto club.

Key Takeaway
The chatbot that failed Dillon Thompson was untasked with resolution, not undertrained. Four organizations had staff on payroll who could have opened an investigation. None of their systems routed him to one on the first call.
CIO/CTO Viability Question
Ask your customer service AI vendor for first-contact resolution rate broken out separately from deflection rate, by escalation type, before you sign the renewal. If the only number the vendor can produce is deflection, you already know which objective the model was tuned against, and it is not the one your customers are calling about.
Sources:
Thompson, Dillon. "My Ebike Delivery Went Missing. When I Tried to Recover It, I Ended Up in Chatbot Hell." WIRED, 15 July 2026, www.wired.com.
McCarty Carino, Meghan. "A Modern-Day Odyssey Through AI Chatbot Hellscape." Marketplace Tech, 23 July 2026, www.marketplace.org/shows/marketplace-tech.
AnswerConnect. "2026 AI Customer Service Attitudes Report." AnswerConnect, 13 May 2026, www.answerconnect.com.
Riptide. "Don't Escalate. Advocate." Riptide, 27 Mar. 2026, www.riptidehq.com.
Disclaimer: This blog reflects my personal views only. Content does not represent the views of my employer, Info-Tech Research Group. AI tools may have been used for brevity, structure, or research support. Please independently verify any information before relying on it.