Customer support · Production

Airbnb transforms voice support with ML-powered IVR using ASR, intent detection, help article retrieval, and paraphrasing

The problem

Airbnb's traditional IVR system relied on rigid menu trees requiring callers to navigate pre-set paths, while generic ASR models produced a word error rate of 33% by misinterpreting Airbnb-specific terminology, undermining downstream intent detection and self-service resolution.

First attempt

Generic pre-trained ASR models misinterpreted Airbnb-specific terms, creating inaccuracies that impacted downstream intent detection and other processes.

Workflow diagram · grounded in source
1
Caller contacts support
trigger
“Call and greeting: IVR picks up and prompts, "In a few sentences, please tell us why you're calling today."”
2
Airbnb-specific ASR transcription
ai_action
“we transitioned from a generic high-quality pretrained model to one specifically adapted for noisy phone audio. Additionally, we introduced a domain-specific phrase list optimization that ensures Airbnb terms are properly recognized.”
3
Contact Reason detection
ai_action
“In production, we deploy the Issue Detection Service to host the intent detection models, running them in parallel to achieve optimal scalability, flexibility, and efficiency. Parallel computing ensures that intent detection latency rema…”
4
Self-service or agent routing
routing
“If self-service is possible, the system retrieves and sends a relevant help article or an intelligent workflow via SMS or app notification. If the caller explicitly requests agent support or the issue requires human intervention, the cal…”
5
Help article retrieval and ranking
ai_action
“We index Airbnb Help Article embeddings into a vector database, enabling efficient retrieval of up to 30 relevant articles per user query using cosine similarity, typically within 60ms. An LLM-based ranking model then re-ranks these retr…”
6
Paraphrasing for user clarity
ai_action
“A Paraphrasing model generates a summary of the user intent, which IVR shares with the user before delivering the solution.”
7
Resolution or escalation delivered
output
“The caller receives an SMS or app notification with a direct link to a relevant Airbnb Help Center article. If further assistance is needed, they can press 0 to connect with a customer service representative.”
Reported outcome

The ML-powered IVR reduced ASR word error rate from 33% to approximately 10%, improved customer NPS, reduced reliance on human agents, lowered customer service handling time, improved self-resolution rates, and achieved paraphrasing precision exceeding 90%.

Reported metrics
word error rate (WER)from 33% to approximately 10%
Intent detection latencyunder 50ms on average
Help article retrieval timewithin 60ms
Paraphrasing model precisionexceeding 90%
Show all 8 reported metrics
word error rate (WER)from 33% to approximately 10%
intent detection latencyunder 50ms on average
help article retrieval timewithin 60ms
paraphrasing model precisionexceeding 90%
customer NPSimproving customer NPS
reliance on human agentsreducing reliance on human agents
customer service handling timelowering customer service handling time
self-resolution rateimprovement in self-resolution rates
Reported stack
IVRASRLLM-based ranking modelvector databaseIssue Detection Service
Source
https://medium.com/airbnb-engineering/listening-learning-and-helping-at-scale-how-machine-learning-transforms-airbnbs-voice-support-b71f912d4760
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

The ML-powered IVR reduced ASR word error rate from 33% to approximately 10%, improved customer NPS, reduced reliance on human agents, lowered customer service handling time, improved self-resolution rates, and achiev…

What tools did this team use?

IVR, ASR, LLM-based ranking model, vector database, Issue Detection Service.

What results were reported?

word error rate (WER): from 33% to approximately 10%; Intent detection latency: under 50ms on average; Help article retrieval time: within 60ms; Paraphrasing model precision: exceeding 90% (source-reported, not independently verified).

What failed first in this deployment?

Generic pre-trained ASR models misinterpreted Airbnb-specific terms, creating inaccuracies that impacted downstream intent detection and other processes.

How is this customer support AI workflow structured?

Caller contacts support → Airbnb-specific ASR transcription → Contact Reason detection → Self-service or agent routing → Help article retrieval and ranking → Paraphrasing for user clarity → Resolution or escalation delivered.

Related customer support cases