Back to case studies
Jul 2020
15 Min read

Improving customer service interaction using NLP

Company Background
Orange is a leading network operator for mobile, broadband internet, and fixed line telecommunications in 27 countries, with enterprise IT and communications services extending to 220 countries and territories. Orange Silicon Valley (OSV) is a US subsidiary of Orange, centered around business development, fostering innovation, technology disruption and helping international companies get access to a wide range of services such as: logistics, retail, 5G Customer support, IOT, and AI.
Once we gave our brief with sample data sets, the Propel project management team was quick to develop an annotation
guideline which helped in setting a structure through which the task could better be delivered, it was truly a smooth sailing for us

Sarah Luger, Senior Director of AI, NLP, & ML at Orange Silicon Valley.

In 2019, Orange Silicon Valley’s Natural Language Processing (NLP) team, in its desire to improve customer service interaction with Orange (Sonatel) customers in Senegal (a multilingual country with customers interacting across French, English and Wolof), needed to collect, source and analyse multiple data sets from the different languages spoken in the country in order to deal with customers in different languages when they called or messaged, seamlessly.

For OSV, this was an opportunity to pilot a new hypothesis, and use the learnings from the pilot to push for a larger roll-out budget across multiple markets - which seemed impossible before, considering the bureaucracy characteristic of large corporations, and the risk involved in testing such assumptions.

Natural Language Processing for Customer Service

To streamline efforts and ensure her team focused on the core mandate of the NLP department at Orange Silicon Valley (OSV), they turned to Propel for help in using Natural Learning Processes to improve systems based on identifiable characteristics and features of translated sets. Propel commissioned a selection of verified talent through the platform who are English school teachers based in Senegal - this selection gave us translators who were fluent in Wolof as native speakers, fluent in English as a profession and fluent in French as their national official language.After 4 months of extensive collaborative work, we were able to translate thousands of data sets from various sources and give the OSV team extensive insights about the data analysis performed. The Propel team successfully modelled systems that could accurately predict the category of a translation (or its source) within accuracy levels to the 90th percentile. In all, we assembled a team of data scientists, web crawlers and translators to work on the different aspects of the project.
Saved the NLP department at OSV over 730 hours during the 4-month engagement period.
Successfully tested out the team’s hypothesis at low cost with minimal risk, and ready for scale.
Established a long-term partnership much more than a one-off project engagement - fast access to Propel talent pool ready to hit the ground running.

More Case Studies

Oct 2020
15 Min read

Building a diverse remote team with propel

Read More