According to consulting firm McKinsey, employees of large corporations spend an average of nine hours per week searching for data and answers to questions regarding their work, which impacts productivity. Researchers at Orange are exploring this issue, focused on a specific area of innovation: enhanced customer relations, or how to facilitate call center agents’ work.
Making Life Simpler for Call Center Agents
Searching through internal databases for the right information, among hundreds of documents that can be dozens of pages long, is a time-consuming undertaking — and one that can be stressful when there is a customer on the line waiting for an answer. During a hackathon, teams specializing in data and AI launched the idea of FIND. In just a short amount of time, they developed a prototype for an advanced search engine able to browse, collect and retrieve information from multiple data sources. They are testing it at Orange Luxembourg, where it is quickly making strides in terms of both capacity and performance.
The solution finds relevant information with sources in response to a query written out in keywords or natural language, and generates a suggested answer to be sent to the customer.
AI to the Rescue
Requests are multiplying within the Orange Group, the project is expanding and, most importantly, the tool is adding a new string to its bow. “We superimposed a layer of generative AI, in this case consisting of GPT technology, onto the search engine in order to further streamline call center agents’ work,” explains Jean-François Pellet, Product Owner of the FIND project. “Thanks to this addition, not only can the solution find the relevant information and provide links to the relevant documents in response to a query written out in keywords or in natural language — it can also to generate a suggested response which can be sent directly to the customer (with human oversight, of course).” FIND stands out for its efficiency and versatility. While the solution is primarily aimed at call center agents, it may also be useful for other purposes. A scaled-down version of FIND is available to all Orange employees through the in-house generative AI acculturation tool, where everyone can enter documents in a range of formats to perform a search and receive a generated answer.
Constant Improvements in Capabilities
In terms of technology, FIND is the result of the combination of an Elasticsearch (an open-source search engine) component adapted to internal needs and GPT, OpenAI/Microsoft’s (LLM). Orange teams have also developed specific algorithms, which give FIND significant added value. “These algorithms are intended to improve the search function as the ways it is used are analyzed: the more people use the tool, the more it will learn and will be able to refine its capabilities, particularly when it comes to classifying/reclassifying data. Our research teams monitor the system to measure its performance, and ultimately direct its evolution toward more optimized components.” Thanks to its architecture, FIND can incorporate new technologies that provide better value for money.
Save Time, Save Money
The high level of performance the tool has already achieved generates financial savings. Initially trialed in Luxembourg, the system is gradually expanding its reach within the Group—notably at Orange France and Orange Business—and will be tested through co-innovation with external customers.
Large Language Models are neural networks capable of mastering the complexity of language from training based on vast amounts of data. BERT (Google) and GPT (OpenAI) are among the most well-known examples.
Jean-François Pellet
