United Airlines
self-service customer service voicebot
United Airlines connected the NLX platform to its contact center and automated 64% of wheelchair requests (90% CSAT) and 31% of flight cancellations through voice and chat.
Key points
- Self-service voicebot for cancellations, baggage, and accessibility requests.
- NLX platform in voice and chat, connected to the IVR and United's PNR.
- 64% of wheelchair requests automated (90% CSAT), 31% of cancellations.
- Evidence B, confirmed status, results measured over six months.
Objective
Absorb a call volume that had become unmanageable after post-covid traffic recovery, by letting travelers handle frequent actions themselves (cancellation, baggage, adding a wheelchair or a service animal) without waiting for an agent.
The deployment
United connected the NLX platform to its contact journey to automate specific requests through voice and chat. A traveler who calls or writes can cancel a flight, check a bag, or add an accessibility request to their reservation within minutes, guided by an assistant that understands the request in natural language and executes the action in United's systems. When a case falls outside the automated scope, the handoff goes to an agent. The rollout targets first the high-volume, low-variability requests, where automation holds without degrading the experience.
Results Proof B
A quantified NLX platform case study (automation rate and CSAT), corroborated by tech press (TechCrunch) that names United and describes the same rollout. The hard figures come from the vendor, the press confirms the scope and scale.
How it works
Documented architectureThe stack in detail
- plateforme NLX Multimodal conversational AI platform (voice and chat) that understands the request in natural language and executes the action in United's systems.
- infra IVR existant United Contact center phone channel to which the assistant is connected.
- infra Systemes de reservation United (PNR) Live access to the reservation record, a condition for executing cancellations, baggage, and accessibility requests.
How it runs, concretely
For ops teams-
1Capturing the request customer
The traveler calls or writes and states their request in natural language (cancel, check a bag, add a wheelchair).
-
2Understanding and qualification AI
The assistant identifies the intent and checks whether the case falls into an automated scenario.
-
3Execution AI
For a case within scope, the action is carried out in United's systems and confirmed to the traveler.
-
4Handoff to the agent AI
Outside scope or on failure, the record and the context are transferred to a human advisor.
The intent expressed by the traveler, matched against the reservation record live. Without reliable access to the PNR and the inventory systems, the assistant cannot execute the action and shifts everything to the agent, which cancels the gain.
How your customers perceive this type of use
Sourced studiesLes consommateurs n'acceptent pas les chatbots par defaut : 64% prefereraient que les entreprises n'utilisent pas d'IA dans leur service client (Gartner, 2024) et pres d'un utilisateur sur cinq du service client par IA n'en retire aucun benefice (Qualtrics, 2025). L'acceptation se construit sur trois conditions mesurees par Salesforce : savoir qu'on parle a une IA, pouvoir escalader vers un humain, comprendre la logique de l'agent.
Acceptance conditions
- Etre informe qu'on parle a une IA et non a un humain (pres de 75% le demandent, Salesforce 2024)
- Un chemin d'escalade clair vers un agent humain (45% plus enclins a utiliser l'agent IA, Salesforce 2024)
- Une logique de l'agent clairement expliquee (44% plus enclins, Salesforce 2024)
Red lines
- Rendre l'humain injoignable : c'est la premiere inquietude des consommateurs sur l'IA dans le service client (Gartner 2024) et 50% craignent que l'IA les coupe du contact humain (Qualtrics 2025)
- Remplacer le service client par l'IA sans alternative : 53% envisageraient de partir chez un concurrent (Gartner 2024)
Sources: Salesforce 2024 · Gartner 2024 · Qualtrics 2025
How to replicate
Inference, not sourcedData prerequisites
- API access to the reservation system (PNR)
- a map of the high-volume intents
- business rules per request type
Org prerequisites
- a contact center team ready to co-design the scenarios
- governance of the escalation to the agent
- a compliance framework for voice and sensitive data
Possible stack
- a voice conversational AI platform
- integration with the existing IVR
- connectors to the reservation systems
The plan, step by step
- Step 1Map the high-volume, low-variability intents (cancellation, baggage, accessibility) and prioritize them by volume and risk.Deliverable: A prioritized list of intents with annual volumes.
- Step 2Connect the conversational platform to the PNR and the IVR, write the business rules for the first request.Deliverable: A first end-to-end automated scenario in pre-production.
- Step 3Pilot on one intent with systematic escalation to the agent; measure automation rate and CSAT per scenario.Deliverable: A pilot review (automation, CSAT, failures and causes).
- Step 4Extend to the following intents and lock down compliance (voice, sensitive data such as accessibility requests).Deliverable: A catalog of scenarios in production and a permanent dashboard.
First step: Choose a high-volume, low-variability intent (e.g. cancellation) and automate it end to end before widening.
The same pattern, proven in other industries
ADT
genAI customer agent
AT&T
Governed enterprise genAI assistant: an LLM connected to internal knowledge, open to tens of thousands of employees and called at scale by automated systems
Bank of America
conversational financial assistant at scale
Sources
- S1 Empowering Travelers Through AI: United Airlines' Self-Service Revolution Interested party archive pending
- S2 How United Airlines uses AI to make flying the friendly skies a bit easier Established press archive pending
An error, newer info, a source?
This page lives on its accuracy. If a figure has moved, if the deployment has changed, or if you have a higher-quality source, tell us. Every sourced correction is verified before publication.