Vision : Dawn of Digital Cops — part 2

I hope you enjoyed reading my previous article titled Dawn of Digital Cops that used RPA to monitor field activities. As for any interesting movie, there is a sequel — and that gets better. This story uses very different technology — Machine learning and Computer Vision. A cop that can intervene and foster corrective actions.

Challenge:

I am a proud provider of essential services during the pandemic era! By the end of December 2020, my team worked relentlessly to provide internet service to millions of customers; and the number keeps growing through 2021. In order to deploy the network modem and a router at the customer premise, the company uses several partners or subcontracting companies for this purpose. Due to the explosion in volumes, the quality of interventions is not always consistent and it is sometimes necessary to intervene several times to satisfy a customer. It is impossible to control all interventions and only 5% to 10% of the installations are controlled.

The explosion in the demand for interventions puts pressure on technicians who, due to lack of time to manage the most difficult situations, may fail to follow certain procedures. As a result, the quality of service is gradually deteriorating due to non-compliant installations.

Solution

A traditional solution would be to have the contractor take photos and an audit team review them. But that is a resource-intensive and ineffective process. Resolution meant another trip to the customer premise to fix the problem.

From post-visit quality control to real-time analysis

I ideated a quality control solution to detect the presence of anomalies in technical intervention photos. Using Deep Learning and advanced Computer Vision AI methods, to recognize devices and their parts and associate them with common issues and resolutions.

Equipped with a proprietary app on a smartphone, the technician is mandated to take a photo of the modem, router, and any other devices that are used to connect a customer to the internet. The emphasis is to take a picture of the exact location where the connection between the network devices is made.

The photos are analyzed in real-time in the cloud by AI-driven image recognition neural networks and report anomalies such as unconnected cable, bad positioning of antenna, equipment placement, etc. The real-time analysis allows errors to be corrected immediately instead of having to plan a rework later. In case of doubt, the algorithm raises an alert and a technician makes the final decision, thus making it possible to solicit support expertise on complex cases only.

Outcome and future

Improved quality of service and customer satisfaction!

The neural networks trained by experts in the field successfully recognizes more than 75% of defects, a figure that is constantly increasing. Thanks to the continuous learning platform, production performance improves and algorithms adapt to the appearance of new defects. The application could cover all the different areas of intervention of the technicians.

  • Safety measures