To make sure that the accuracy of the collected data is as high as possible, each readout comes with a 'confidence score' (“high”, “medium”, or “low”), indicating under how much reliability the readout could be extracted. This can be used to alert the user that the readings could not be reliably extracted from the photo in case of a low confidence level. There could be many reasons for this, for example poor image quality (blurry or unsharp) or one or more of the digits being unreadable due to dirt, flashes, reflections, etc.
In the end, it is up to you to decide whether or not to use the confidence levels that come with each extracted meter reading, serial number and barcode. We would strongly advise to implement additional actions in case of a low confidence level, and potentially also in case of a medium confidence level.
One option would be to have your data validation team manually inspect (a random sample of) these rare, 'unconfident' readouts (on average around 1,8%). However, this does not allow for real time feedback to the user in case the relevant information can also not be extracted with the human eye and a photo retake would be needed.
That is why, the more commonly used action would be to implement a feedback loop to instruct the user in real time to perform an extra action. For example to retake the photo, or to double check the extracted information.
