What I've done is to take about 1-2 weeks of recent data and put it in a spreadsheet
Delete any rows without a mid-cycle test
Sort it by dose and pre-shot glucose ranges (ex 200-300, 300-400 etc)
Make a column which calculates the difference between pre-shot and approximate nadir value for any rows with this data
For a given starting range (ex 250-350) and dose (ex 2 units, look at the maximum drop a given dose had.
Figure the range starting 100 mg/dL higher may need more insulin and add 0.25 to the dose for that
Figure the range starting 100 mg/dL lower may need less insulin and subtract 0.25 to the dose for that
Experiment - Try this out and be prepared if the lower range heads to 50 mg/dL
Adjust up or down if the estimate doesn't work.
Its a trial and error approach.