.png)
T-Mobile US
Telecom · US
75% resolved without a live agent
T-Mobile's IntentCX resolves three out of four customer interactions autonomously, saving $80M per year
Built jointly with OpenAI, IntentCX recognises customer intent in real time across voice, chat, and app, autonomously executing billing adjustments, plan changes, and issue resolutions while briefing human agents for any escalation.
Friction
T-Mobile's call centres had average call times of 8–10 minutes, low first-call resolution, and high agent turnover. Customer experience is central to T-Mobile's positioning, making this a strategic as well as operational problem.
Breakthrough
IntentCX built on GPT-4o and GPT-4-turbo with real-time retrieval over T-Mobile's customer, billing, and network systems. Real-time speech recognition, intent classification, autonomous action execution, and at escalation a dynamic briefing for the human agent. One of the largest publicly disclosed enterprise OpenAI implementations in telecom.
Impact
75% of customer interactions resolved without a live agent (Q4 2024). Average call time down 30%. $80M annual savings in customer care operations. NPS in customer service segments up 9 points. 100% of customer service employees supported by AI during every call.
Unlock the full analysis with breakthrough, impact, what made it smart and its technical approach below!
Problem
T-Mobile's brand promise is built on a fundamentally different customer experience. Yet the call centre operation looked like any other telecom: long waits, multiple transfers, unresolved first contacts. The gap between brand positioning and operational reality was a strategic risk.
What made it smart
IntentCX does not just answer questions, it executes. It can make billing adjustments, change plans, and resolve issues in real time during the interaction. This transforms the AI from a routing assistant into an autonomous agent with transaction authority, within defined guardrails.
Technical approach
GPT-4o and GPT-4-turbo with real-time retrieval over customer, billing, and network systems. Real-time speech recognition and intent classification trigger automated actions where possible. At escalation, a dynamic briefing is generated for the human agent, providing full context so they can continue the conversation without the customer repeating themselves.
Strategic lesson
Autonomous customer AI creates maximum value when it can execute, not just recommend. The key design question is not what the AI can understand, but what it is authorised to do.
Reflection question
In your customer service operation, what percentage of interactions require a human decision versus a human execution of a standard action, and how much of the second category could be safely automated?