I'm only on my second project using LSP CAD, but I thought I would share a tip that I have found to work well for me so far.
I start off by measuring the response and impedance of the drivers in the cabinet as usual. I also then measure the total system response with no crossover, and the total system response with a very rough crossover in place. This gives me three sets of measurements - the usual set of driver measurements, plus two different system responses.
I then start using LSP CAD. I create two dummy projects - one with no crossover, and one with the rough crossover. The aim is to verify that I can accurately reproduce the measured responses in each case. When I can reproduce in LSPCAD each of the two measured responses, then I know that all my input data is good and my measurements are good, and I can then start designing the actual crossover with confidence that everything is working.
You can go further, and take measurements with the mic in a different position, say twice as far away, then create a dummy project with the virtual mic in LSP CAD moved to the new position. Again, if all your input data is correct, then you'll get really good agreement between what LSPCAD predicts and the actual measured effect of moving the mic. The more sets of data you can get to agree, the more chance that you have eliminated all the errors. If anything doesn't agree, then chances are that something is wrong and that will mess up your final crossover design - ie you won't get the predicted response when you build the final crossover and measure it. The rough crossover can just be a guess, using whatever chokes and caps you have available (within reason).
This may seem like an unnecessary step, and in theory it is unnecessary. But I found that until I did this I wasn't getting anywhere. There are so many things you have to get right in LSP CAD. An example is the acoustic centre of the drivers. If you have multiple sets of measurements, you can tweak the offset a mm at a time and when its right, all the little wrinkles in the response will start to line up with the measurements. Without multiple measurements, its hard to figure out if you have a phase error or a position error.