Orange: Building a Sustainable
Data Practice

More than ever, the global health crisis of 2020 has illustrated the need for telecommunications companies to lean into Everyday AI, bringing data science, machine learning, and AI out of the realm of experimentation and into the day-to-day operations of the business. With increased demand, use cases like monitoring network usage, building 360 degree customer knowledge, and ensuring network resiliency are more important than ever, while concerns about customer retention and — like all industries — agility in forecasting come to the forefront.

Orange, one of the largest operators of mobile and internet services in Europe and Africa and a global leader in corporate telecommunication services, has been working to ramp up its efforts in data science and machine learning for the past several years, increasing its capacity to leverage data in all areas of the business. By choosing the right technology and empowering more and more people to work with data (both via hiring and upskilling), Orange has been able to overcome the challenges of being a non-digital native business to improve the overall level of data competency at the organization.

Challenges

The client services department at Orange has a data science team which, until two years ago, was performing mostly ad-hoc analysis for the business and had limited ability to work on more complex machine learning-based projects. In order to scale out the team and expand their scope, they had to overcome several challenges:

1. Tooling: Only those who knew the tool and its proprietary language could work with data, which limited the use of data to statisticians or data scientists. Even then, the data was difficult to access,making projects difficult to get off the ground. At the same time, the tool was geared toward BI and wasn’t capable of supporting machine learning-based data projects. 
2. Hiring: The data team at Orange was struggling to hire talented data scientists fresh out of university and with lots of ambition as well as creative ideas (traits they were seeking to enliven their data science practice). Unfortunately, this was largely a function of the tooling challenge, as young data scientists were largely looking for jobs where they could work with open-source tools (such as Python or R). Anyone they did bring on board had to learn the legacy tool and took months to get up to speed and start being productive.

It’s worth noting that these challenges are not unique to Orange — they likely sound familiar to any telecommunications or even non-digital native business looking to pivot and develop a more robust data practice. How did the data team at Orange overcome them?

How Orange Sparked Change

Change at Orange came from the bottom-up and began with the client services data team wanting to work on more cutting-edge machine learning projects.

People

In order to spend more time on machine learning, the team realized they needed to empower people (like analysts) to work on their own simple data analysis projects, freeing up the data team’s time for more advanced data science work with potentially bigger impact.

By enabling analysts and business people to work on data analysis themselves, data practices are more infused throughout the client services organization and not siloed to just one team. Today, there are more than 100 analysts and other business users across Orange who are empowered — with Dataiku — to work with data.

Technology

Orange needed to choose the right tool that would:

• Allow the data team to work on machine learning projects.
• Allow new data scientists to get up to speed and start being productive quickly, working with tools they wanted to use.
• Enable analysts to work independently.
• Not alienate veteran employees with lots of valuable experience but who didn’t necessarily want to learn a new tool or system.

Dataiku, the platform for Everyday AI, was the solution that helped facilitate the change at Orange. They were drawn to Dataiku’s flexibility, openness, and dynamism. 

Machine Learning Use Cases

Once the team was armed with Dataiku, they were able to start transitioning smaller BI projects to the business and work on machine learning use cases. A few examples include:

• Call load detection + triage: When a client calls Orange customer service, all agents might be busy with other callers. So the team set out to create a machine learning-powered system that anticipates this volume and offers to call the customer back at a certain (less busy) time while still respecting the business’ service-level agreements (SLAs). In addition to call center data, the model also takes into account the reason for the customer’s call to be able to more accurately determine priority as well as how long the call might take. Note that there are more than 400 skills in which a particular agent may be specialized, so this was a non-trivial task. Overall, the model took less than a month for the team to build using Dataiku.
• Preventing unwanted charges: The client services team’s main goal is to improve customer net promoter score (NPS), which means increasing overall satisfaction. One of the biggest sources of frustration for clients can be excessive charges, so the data team wanted to address this problem with machine learning (specifically clustering) by looking at different variables that might point to clients incurring charges — e.g., overseas charges or going over call/data volume. Dataiku allows the team to perform clustering quickly to understand what certain portions of the client population have in common that the business side might want to take action on.

In addition, analysts and business people are more empowered than ever thanks to Dataiku. Key performance indicator (KPI) dashboards that previously took up to a month or more to build and update now take one week, maximum. The success in use cases on the machine learning side has allowed the team to continue to grow, going from six data scientists three years ago to more than 25 people both on the data science/analytics and on the data tooling side.

“Before, it was up to us to prove value to the business. Now it’s the business that approaches the client services data team for projects around machine learning. Two years ago we’d get maybe one request per year; now we get several per month.”