Crop
According to the NLP manual, the scans should be cropped in a way to hide the unexposed film base/borders. I often convert negatives without cropping and then set border buffer to 25% which can also get me better initial conversions with underexposed negatives. NLP looks for the “white” and “black” points in each of the r, g and b tone curves. Leaving blank areas will push NLP on a false track - unless you increase the border buffer. If the negative includes large areas of high density, it can help to set a crop to exclude those areas (e.g. overexposed skies)
White balance
I find that NLP is fairly tolerant to white balance and that wb can change the initial conversion nevertheless. AutoWB can help too.