Case Study

Uber uses TAUS Estimate API to measure and improve the quality of its global customer care platform

Support Content is the Face of the Customer
The quality of content significantly influences the support experience of Uber’s customers. For instance consider an Eater who requests help canceling a severely delayed order. The same resolution, such as a refund, can be accompanied by either a robotic-sounding machine-translated message, or a message with a style and tone that expresses genuine empathy and acknowledges the user’s disappointing experience on our platform.
Support content, as a natural representation of the support experience and the brand’s promise, significantly influences how customers feel and perceive the company brand. Furthermore, such content plays a crucial role in educating users about product behavior and company policies while encouraging them to action. Additionally, support content (such as knowledge base articles) helps deflect commonly asked questions, reducing the number of contacts handled by support agents. In summary, support content is typically the first point of contact for a dissatisfied customer. As such, it is crucial that this content not only placates and reassures the customer, but also addresses the primary issue, transforming customer dissatisfaction into delight.
Ready to get started?
Uber’s Customer Care Platform
Uber’s Customer Care platform (Support platform) presently supports content across multiple business sectors, including Uber Mobility (Rider, Driver), Uber Delivery (Eater, Courier, and Merchants), Uber For Business (Organizations and Employees), Uber Freight (Carrier, Shipper). Support content is translated with MT engines into and out of more than fifty languages. The platform facilitates the experience for hundreds of content creators, strategists and specialists who produce the support content. Furthermore, it is used by thousands of agents who incorporate the authored content into their daily handling of support tickets, and by millions of users who access it as an essential part of their support experience.
Uber’s challenges
Human review at a global content scale is unthinkable, costly, and time-consuming.

Uber is seeking to address several challenges with the Support Content platform, focusing on  improving content quality and effectiveness through personalization, relevance, empathy, and user sentiment. These challenges include:

1- Can we filter out bad machine translated support content before it reaches our customers?
2- Can we measure and enhance the “robotic machine translated score” of the support content?
3- Can we measure and improve the user sentiment when interacting with support content?
4- Can we personalize the tone and voice of the content (e.g., transitioning from informal to a more formal structure) to suit the customer’s conversation patterns?
5- Can we identify and highlight content clean-up actions for content strategists and admins?
6- Can we automatically suggest the creation of new content based on data from our customer tickets?
7- Can we enhance content authoring with auto-suggestions, flag inconsistencies in grammar/tone/voice, recommending similar content, etc.?
The TAUS Estimate API solution
In October 2023, Uber’s localization team integrated the TAUS Estimate API into the Bliss Content Platform. TAUS’s team has customized QE models for multiple languages utilizing its vast repository of in-domain data. Every chat outbound message and support article are now evaluated using a TAUS QE score, achieving 85% accuracy compared to human evaluation. The TAUS Estimate API processes approximately half-a-billion characters every month returning scores in real-time. With this, the first objectives have already been realized: poor machine translations are filtered out, and all support content is scored, providing Uber’s team with a strategic tool to begin enhancing the customer experience.

In the 2024 roadmap, the focus is on expanding the language set with QE scores, and analyzing lower-quality graded support content by utilizing data and models to personalize voice and tone. The goal is to identify and clean up poor content, such as offensive language, improve the source language before it is machine-translated, and retrieve new content from language models. The TAUS NLP team continues to work closely with the Uber globalization team to optimize results from Large Language Models and to further expand on use cases and opportunities offered through AI

About TAUS Estimate API
The TAUS Estimate API provides real-time insights into your content quality, reducing the need for human review and enabling rapid assessment and deployment of accurate language models. The Estimate API seamlessly connects to your content workflows and provides quality scores customized with your data so the results match your organization for accuracy, tone, and authenticity.
Let's connect

Talk to our NLP Experts to gain insights into the quality of your MT output and make decisions based on that, saving cost and avoiding catastrophic errors with a customized Quality Estimation model.

Discover more Case Studies

Enabling 15% Increase in Number of Perfect Translations for ING Hubs Poland

ING Hubs Poland found out that training with TAUS datasets improves the number of perfect translations by 15% and with 95% precision.

Domain-Specific Training Data Generation for SYSTRAN

After the training with TAUS datasets in the pandemic domain, the SYSTRAN engines improved on average by 18% across all twelve language pairs compared to the baseline engines.

Customization of Amazon Active Custom Translate with TAUS Data

The customization of Amazon Translate with TAUS Data always improved the BLEU score measured on the test sets by more than 6 BLEU points on average and 2 BLEU points at a minimum