##### 14.04.2023
_In today’s globalized market, ensuring excellent customer service across languages is crucial for businesses, especially in the financial sector. In this article, we’ll explore how our client, Mastercard a global leader in payment innovation and technology successfully implemented Bilingual Customer Service Chatbot to cater to both English and Spanish speakers, resulting in improved call center efficiency, increased sales, and enhanced customer loyalty._

 ## The customer
Our client, **Mastercard’s Research and Development Labs**, is a global leader in the field of payment innovation and technology, offering a wide range of financial services.


## The context
Our client’s research and development labs wanted to develop a chatbot to improve customer service, they needed to ensure the chatbot could not only understand both Spanish and English but also understand and accurately respond to a range of different customer queries. Failure to do this would result in frustrated, isolated, and potentially angry customers.

Our client needed unique, multi-lingual text data covering 20 designated payment scenarios in English and Spanish to ensure their chatbot was smart, fluent, and not alienating customers. They also required that the data be provided in 3 weeks.

Obtaining this type of comprehensive data under highly specific parameters is a multi-step process that can prove time-consuming without experts involved. We worked closely with our client to ensure maximum efficiency and quality at every stage, from assisting with scenario development to drafting a comprehensive set of prompts to collect, validating, and annotating all collected text.

> Obtaining this type of comprehensive data under highly specific parameters is a multi-step process that can prove time-consuming

### The solution
#### Step 1 – Variant collection
Although the timing was tight, we could meet the deadline because of our well-tested workflows and global crowd. We sent original prompts – a question to which a model will respond – covering all **20 payment scenarios** to our crowd, who produced 2,000 variants for each language in response. We had each of those variants validated and corrected by different crowd members, and **for the 475 questions, we collected 1,500 answers** and had each one ranked for relevancy three times as part of our Bilingual Customer Service Chatbot development process.

#### Step 2 – Entity tagging
Finally, we had every variant annotated for entities by three separate crowd members. Ultimately, we collected over **6,000 entities split between English and Spanish**. We then removed repeats and delivered over **1,000 entities in each language** to Mastercard.

![Bilingual-Chatbot_1.png](https://prdstrapimediastorage.blob.core.windows.net/prdstrapimediastorage/assets/Bilingual_Chatbot_1_89179e3c12.png)

Our precise, industry-specific data was used to train chatbots that could understand numerous customer requests in English and Spanish. This improved call center efficiency, improved sales, and won customer loyalty. Although chatbots are an almost ubiquitous part of our lives, to truly enrich the customer experience, they need to be smart, fluent, and accurate, which can only be achieved with high-quality training data.

Want to learn more about how our tailored solutions can benefit the banking industry? Explore our **[range of services for the banking and insurance sectors](https://resources.defined.ai/finance-and-insurance/)** and discover how we can enhance your customer experience and streamline your operations!

thumbnail_Advancing-Payment-Technologies@2x-2048x515.png

Advancing-Payment-Technologies@2x-2048x515.png

Speech

Spanish Spontaneous Dialogue

Named Entity Recognition

Aspect-Based Sentiment Analysis

White Papers

thumbnail_whitepaper-crowdsourcing-2048x579.png

whitepaper-crowdsourcing-2048x579.png

Overcoming the Challenges of Crowdsourcing AI Training Data

You might also be interested in: