Retail AI Failures Stem From Broken Journeys, Not Algorithms

Retail brands racing to deploy AI in their contact centers are solving the wrong problem, according to Mark Rohan, co-founder and COO of telecom testing platform Klearcom.

While the industry focuses on data ethics and algorithmic transparency, the real failures happen at the journey level—when customers get routed incorrectly, trapped in loops, or bounced between departments despite technically successful system performance. No governance framework fixes a broken call flow, Rohan argues in a recent commentary first published on Retail Customer Experience.

The metrics gap

AI decouples system performance from customer experience in dangerous ways. A contact center can report clean operational metrics—calls connected, prompts delivered, routing executed—while prompt changes or model updates quietly degrade what customers actually reach. Standard quality assurance wasn't built for systems that evolve between tests.

The most revealing failures are calls that connect but route incorrectly. Every internal metric looks fine because the system completed its task. But the customer calling about an undelivered order who encounters an unhelpful chatbot and gets transferred multiple times leaves with an impression of organizational incompetence, regardless of backend performance.

Rohan points to high-profile examples. When UK delivery company DPD's chatbot was updated in a way that stripped guardrails, it wrote poems criticizing the company and swore at customers—problems that went undetected until a viral social media post exposed them. Swedish fintech Klarna deployed AI it claimed handled work equivalent to 700 agents, only for CEO Sebastian Siemiatkowski to later admit the transition hurt service quality, prompting the company to resume human hiring.

Pre-launch testing proves nothing

A successful launch test is a snapshot confirming the system worked once under controlled conditions. It reveals nothing about behavior six weeks later when prompts are updated, carriers reroute traffic, or speech recognition struggles with regional accents.

Retail brands need to test journeys as customers experience them: from dial tone through IVR prompts, routing decisions, hold times, and final resolution. With AI, the bar rises further—tests must verify not just connection but correct destination, appropriate solution, and reasonable timeframes, running more frequently than traditional QA cycles allow.

Why it matters

Enterprise contact center outages can cost over $540,000 per hour, but operational costs can be contained. When AI fails customers, the damage appears as lasting reputational harm. In an environment where customer tolerance for friction has reached bottom and switching costs have disappeared, the gap between internal performance metrics and actual customer experience becomes existential. Forty-seven percent of consumers will abandon brands over poor data practices, according to Sogolytics research cited by Rohan, and some believe AI actively worsens their experience.

Airbnb recently reported its AI customer service bot handles 40 percent of issues independently, demonstrating the technology's potential. But internal performance and customer experience require different measurement approaches, and AI investment often widens that gap rather than closing it.

In retail, trust builds across the full customer journey, not just during AI interactions. Monitoring needs to become as central to contact center operations as the automation layer itself. The winners won't be brands that automate fastest, Rohan concludes—they'll be the ones that can prove every customer journey still works.

The analysis was originally published by Retail Customer Experience, where Rohan detailed how AI complexity demands new approaches to quality assurance in customer-facing systems.