Multilingual Synthetic Data

Case Study
Retail chatbot

Deploying a bot which is able to engage in successful conversations with customers worldwide.

Category

Retail

overview

Over the past couple months, we have been working with one of the world's Top 3 fashion retailers to help them improve their customer service bot to answer FAQ’s on their website. We are proud to help retailers innovate to enhance their customer experience.

Defining the problem

Artificial Intelligence/ Machine Learning is giving a new push to customer service automation. As enterprises know through experience, customer service automation is a highly ambitious goal, at its highest level, it involves a sensible conversation between a person and a machine. Customer service software lacks the expected levels of language understanding and this is halting the adoption process.

 

As one of the world´s top fashion retailers, they receive thousands of queries in multiple languages every day through their online chat. Being able to answer customer inquiries quickly is extremely timely and laborious, a cost most companies can´t afford. Their existing training data was created manually and was not enough to build an effective conversational agent.

Proposed solution

Multilingual Synthetic Training Data for customer service automation. This approach offers consistency between intents, verticals and languages. It also enables fast error correction and speed of retraining.

Reduced Time to Market

1

Scalability

f

Modularity

Permanent Accuracy Improvement

~

Full Data Protection

Multilingual

Easy Integration

Figures highlights

Generation of training data is reduced from months to days

Roll out in multiple languages in parallel

Reduction of overhead costs from day 1

  • Out of the box accuracy - 65%
  • Accuracy - 90%*

* Accuracy rate in up to 6 months

the results

Bitext employs a scalable and data-driven linguistic-in-the-loop methodology. This approach provides a measurable improvement to Natural Language Understanding performance and intent detection. Therefore, we can dramatically increase virtual agents´ accuracy, thus making them more human-like.