Customer files checked by virtual agents in Senegal

How can artificial intelligence technologies help reduce workload by automating repetitive tasks? Orange Senegal has been investigating this on a specific case: checking the subscription files from mobile offers. In just a few months, the carrier has launched a tailored solution based on Robotic Process Automation (RPA) and Artificial Intelligence (AI) from open-source ecosystems.

“A newly released AI model that can be used by anyone that same day.”

The work carried out by the Orange Nomad Back Office team in Senegal seems never-ending and futile. Day after day, the team’s 17 members manually process 15,000 subscription requests submitted by resellers on the ground via the Nomad platform.

When AI helps humans

Their work involves verifying the authenticity of identity cards sent in the form of photographs and checking that the details on them match the information entered by the resellers. It’ is a repetitive task that takes, on average, one to two minutes per file. Under the leadership of Sandèné Ndao, teams in the Smart Data Department have therefore looked into the possibility of integrating artificial intelligence with a view to automating the process and relieving the Back Office of some of its workload. The project, which was launched last year, has quickly given way to a tailored and high-performance solution. It has been in operation for four months and plans are already in place for further technical developments and new use cases.

Combining AI with RPA

The system, which was designed and implemented by both the Smart Data Department and the IT Department, based on data provided by the customer service team, combines a variety of technological building blocks. The automated circuit runs as follows: a file is sent via the Nomad server, from which data is collected by an RPA system that then uses an API, which orchestrates the validation workflow. This API checks the validity of the file with the help of an AI image recognition system (TensorFlow) and a character recognition engine (Tesseract OCR). It then returns the file to the RPA with a validity indicator. If the indicator for the data in question (ID document and information entered by resellers) is above the reliability threshold of 80%, the subscription request is approved and resubmitted in Nomad so that the services can be activated. If the indicator is below 80%, the file is sent to the Back Office to be assessed by a human team.

Opportunities abound with open source!

The virtual agent created as a result of this composite assembly is even more remarkable as it was created from building blocks mostly from the open-source ecosystem. “Orange is looking closely at AI-related technologies, encouraging the use of open-source solutions rather than proprietary software,” explains Sandèné Ndao. “The world of open source provides us with new technologies that we can adapt to our business needs, ones we can test and troubleshoot quickly.” These environments have been used by the Group to build a fully automated architecture in no time at all, using mature software with a high level of technical performance. The TensorFlow platform can be used to develop and implement machine learning and transfer learning models with the added bonus of significant time savings. “In terms of accessibility,” Sandèné Ndao continues, “open source is a godsend in projects like ours. It presents some fantastic opportunities, such as newly released artificial intelligence models that can be used by anyone on the same day that they are published.”

A 24/7 engine

Most significantly, the original goal of reducing the Back Office team’s workload has largely been achieved. The AI/RPA model deployed runs 24 hours a day, 7 days a week, with an average processing time of 20 seconds – the actual data analysis takes less than 5 seconds -representing an output of 4,320 files per day. Encouraged by this first successful deployment, the team now wants to expand the initiative. “In particular, we are working to improve the performance of the Optical Character Recognition (OCR) engine, which currently takes on average 15 seconds to extract photo data. At the same time, we are looking into the possibility of parallelising the application and using more than just one robot, so that, in time, we can check 15,000 files. Finally, we are planning to broaden its use case to the Orange Money service, for which the sign-up process is similar in relation to the context of banking.”

In short, at Orange Senegal, the Back Office has become an AI showcase!

Read also on Hello Future

AI contributes to species conservation with TrapTagger

Getty Images - young scientist using a digital tablet while working with crops on a farm

Technology against food waste


AI: BigScience brings transparency and variety to large language models


What AI Can Detect About Humans Through Their Voices

An analyst looks at data on a screen

Online sentiment analysis, a promising predictive tool


Generative AI: a new approach to overcome data scarcity


Self-supervised learning paves the way for a common-sense AI


Drinking water network management grows in intelligence