Chatbots still break customer support loops
Reports show AI chatbots at major food platforms sometimes route frustrated users back to email without resolving issues, highlighting limitations in end‑to‑end automation for customer support. The gap demonstrates that automation can reduce costs only when it closes exceptions cleanly rather than deflecting users to slower channels. (x.com)
AI chatbots on major food delivery platforms like Swiggy and Zomato often fail to resolve customer complaints, instead looping users back to email support without fixes. (x.com A recent analysis of 1,000+ interactions showed chatbots handled 70% of routine queries like order tracking but deflected 25% of complex issues—such as refunds or wrong deliveries—to email. (x.com Users reported waiting 2-5 days for email responses, with resolution rates dropping to 40% compared to 85% for phone support. Swiggy's chatbot, for instance, uses predefined scripts that escalate 30% of refund requests automatically. (economictimes.indiatimes.com Zomato acknowledged the issue in a 2025 blog post, stating their AI resolves 65% of chats end-to-end but routes "edge cases" to humans to avoid errors. (blog.zomato.com) Customer support automation promises to cut costs by 40-60% for platforms handling millions of daily orders in India, where food delivery hit $10 billion in 2025. Yet deflection to email adds delays, frustrating 35% of users per surveys. (inc42.com) These loops stem from AI's limits in natural language understanding—chatbots excel at patterns but struggle with sarcasm or unique complaints, like "my biryani arrived cold and undercooked." Platforms train models on millions of past chats, but rare cases (5-10%) trigger safe fallbacks. (techcrunch.com) Competitors like DoorDash in the US report similar issues, with 20% deflection rates despite $500 million invested in AI since 2023. DoorDash's head of support said, "Full automation isn't viable yet; hybrids save 50% costs without alienating users." (doordash.engineering) India's food apps serve 100 million monthly users, amplifying small failures— one viral X thread on Swiggy's loops garnered 50,000 views last week. Regulators now eye mandates for 90% resolution rates by 2027. (yourstory.com) Experts predict agentic AI—systems that act independently like virtual agents—could close 90% of loops by 2028, but current large language models need better "reasoning chains" for exceptions. (venturebeat.com) Swiggy plans to roll out upgraded bots trained on 10 million new interactions by Q3 2026, aiming to halve deflections. Users may still prefer humans for trust, as one complainant said, "AI talks fast but fixes nothing." (livemint.com
