Data entry ops · Production

Leading vacation rental company automates ML labeling pipelines with Labelbox to enrich unique property listings

The problem

A large vacation rental company faced the challenge of classifying and labeling vast volumes of unstructured listing image data across multiple business units with different formats, and needed to standardize ML infrastructure to run effective data science initiatives at scale while reducing human labeling costs.

Workflow diagram · grounded in source
1
Unstructured listing images submitted
trigger
“Many business units were handling unstructured data in different formats and figuring out the best way to classify, label and enrich these images of listings was a key priority”
2
Active learning prioritizes data
ai_action
“the company leveraged active learning and tightly integrated these workflows with their labeling operation in order to prioritize the right data to annotate”
3
Model generates labels
ai_action
“a majority of labels are model generated”
4
Subject matter experts review
human_review
“accepted by a team of subject matter experts”
5
Enriched unique listings published
output
“enables the team to create unique listings with rich metadata much faster than before”
Reported outcome

After three months, ML pipelines were fully automated with the majority of labels model-generated and accepted by subject matter experts.
Human labeling costs fell to a fraction of their original level, over nine million annotation tasks were completed, and the enriched unique listings collectively earned more than $300 million globally since the start of the pandemic.

Reported metrics
Unique listings earnings since pandemicmore than $300 million globally
Annotation tasks completedover nine million
Human labeling costsjust a fraction of what they were
time to fully automated ML pipelinesthree months
Reported stack
LabelboxAnnotateCatalogactive learning
Source
https://labelbox.com/customers/travel-customer-story
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

After three months, ML pipelines were fully automated with the majority of labels model-generated and accepted by subject matter experts.

What tools did this team use?

Labelbox, Annotate, Catalog, active learning.

What results were reported?

Unique listings earnings since pandemic: more than $300 million globally; Annotation tasks completed: over nine million; Human labeling costs: just a fraction of what they were; time to fully automated ML pipelines: three months (source-reported, not independently verified).

How is this data entry ops AI workflow structured?

Unstructured listing images submitted → Active learning prioritizes data → Model generates labels → Subject matter experts review → Enriched unique listings published.