Orange could not stay away from the artificial intelligence revolution. Through its business division, Orange Business, the telecom operator advises its customers on new use cases enabled by new machine learning or generative AI technologies. The French group also intends to capitalize on the contributions of AI for its own needs.
It all starts with the acculturation of its employees. 27,000 employees have already followed the program called Data AI Starter. This training aims to explain the contributions of AI in their daily work while addressing the ethical dimension linked to its use. It is also about giving them advice on “prompting”, or the art of making themselves understood by the machine, by refining the wording of requests.
On the art of “prompting”
Standard profiles (personae) have been defined so that the training course adapts to the specific needs of a developer or marketer. A library lists ready-to-use prompts and communities allow the exchange of best practices. “ Working with generative AI allows employees to improve their skills while reducing their apprehensions regarding the tool,” advances Joachim Fléchaire, VP AI Tools & Technology at Orange.
To move from theory to practice, Orange has made available to its employees an internal and secure ChatGPT for the generation of text and images. This tool is based on high-end and general reference models offered by OpenAI or Anthropic and open source models designed by Meta, Mistral or Hugging Face.
“This mix allows us to respond more precisely to our needs », Estimates Joachim Fléchaire. The technique known as retrieval augmented generation (RAG) also refines the relevance of responses based only on internal knowledge bases.
Decision support for telephone advisors
While the use cases are almost endless, Orange focuses on two major themes: improving the customer experience and optimizing its networks. In the first client, the aim is to equip call center representatives. ” LAI can analyze conversations with customers in real time and dynamically push the right technical sheet or offer onto the agent’s screen, continues Joachim Fléchaire. This saves him from having to search for information in multiple knowledge bases. »
Dynamic scripting techniques can detect keywords like “failure”, “access problem”, “Android” or “iPhone”. The tone of the conversation, whether cordial or not, also provides valuable information. “ AI-generated content is not exposed directly to the customer, with the possible risk of hallucination, specifies Joachim Fléchaire. It is a decision-making aid for the telephone advisor who always retains control. »
Likewise, AI makes it possible to recover context information that has been entered by the customer in a chatbot. The latter does not have to re-explain everything. At the end of the exchange, the AI will automatically populate the CRM. To support this change, telephone advisors will undergo training. A bot will in particular simulate an exchange with a customer. When scaling up to AI, store salespeople or on-site technicians will also be equipped.
Site location selection and predictive maintenance
The other large family of use cases concerns the optimization of telecom networks. First of all, AI helps shed light on the most judicious investments. “ What is the most relevant location to install a base station or deploy fiber in relation to the needs of our customers? », asks Olivier Simon, VP, Data Technology, Models & Analytics at Orange.
Predictive analysis is based on the evolution of network traffic over the next four or years. To do this, the AI relies on town planning data which allows us to know that such a business campus will be set up in this area, such and such a shopping center will expand. As in France, network coverage is already extensive, the establishment of a new site aims above all to increase network capacity.
Another use case: predictive maintenance. Sensors measuring temperature or electrical voltage provide real-time data on the state of health of the operator’s equipment. ” LTheir analysis makes it possible to detect abnormal behavior and anticipate intervention on a power supply, a battery or an air conditioning system, without waiting for their replacement every X years », specifies Olivier Simon.
Real-time anomaly detection
The optical assessments of the network also provide valuable information. When the optical fiber is twisted, moved or incorrectly mounted, the signal loses intensity and returns to its starting point. Beyond predictive maintenance, monitoring of technical data allows real-time detection of anomalies. “ This allows you to move from a reactive mode – calls to customer service generate a trouble ticket – to a proactive mode, rejoices Olivier Simon. VSThis is a quality of service issue. »
This monitoring makes it possible to locate the problem – an impacted site or several – and to know its origin. “ Most remediations are done without on-site intervention by changing, for example, network settings. A football stadium fills up and the mobile network saturates, it needs to be resized”illustrates Olivier Simon.
And if an on-site intervention is nevertheless necessary, the technician has a summary of this pre-investigation information. “IHe knows which equipment is causing the problem and what the most likely remedial action is, for example restarting it. This allows him to get straight to the point,” judge Olivier Simon.
AI projects applied to networks only really started four or five years ago with a scale-up going back to 2023. The next stage, over the next two or three years, will focus on the constitution of the network’s digital twin. By showing in this 3D modeling the different equipment but also the network layers (IP, fiber, MPLS), the detection of anomalies will be even more relevant.
AI-generated visual, Microsoft Designer