In most telecom companies, field workers include both employees of the telco provider and those of its subcontractors. To document the quality of task delivery properly, they make significant amounts of photographs. It occurs, that some technicians try to cheat by uploading bad quality images, pictures taken during previous installations in other locations, pictures of other pictures, etc. – all this e.g. to report closing more tasks than have really been executed.
Even when the service provider is in possession of data necessary to check all these photos, it’s very time consuming to manually discover such frauds, due to the volume of information to process.
Comarch Field Service Management is a platform for field staff and equipment management, and as such, it gathers huge amounts of different kinds of data. When that data is properly prepared, it can be a valuable source of information for further processing.
One scenario in which the data can come useful is the said fraud prevention based on photo documentation.
Due to the amount of data to be processed, in order to find the potential frauds, it is much easier to run this scenario with the help of AI algorithms. The first approach is to utilize neural networks to find modifications on images (e.g. pasting an antenna onto a building image). Such cases can be reported for further manual inspection.
The second approach is to cross-compare pictures to assess if they all relate to the same task. The AI algorithm can enrich the image with additional data like picture geolocation, location of the task, field worker location, together with their timestamps. That is enough data to confirm the time and place where documentation has been prepared.
Yet another method is to use neural networks to combine photos delivered by a technician with satellite images from the same territory / of the same building / object. In this case a vertical image from the satellite can be projected onto a horizontal picture taken from the ground to assess if both present the same object / building, whether they were taken in the same time, etc.
Moving further, such dataset can be successfully used to validate correctness of the installation. Recognized object can be split into logical elements and compared with the knowledge base assessing quality of the installation and configuration.
Tell us about your business needs. We will find the perfect solution.