The first step is to simplify your input otherwise you would need some complex logic to identify as you step through the table you have gone from Paris Dec_2021 to London Nov_2021.
If your input is truly structured as you have shown then you could very easily explode out this table using the Split by attributes tool.
This creates as many tables as you have locations, the output of one such table would be:
Depending upon your input structure you may need to sort the table into a correct date sequence.
Add a new field of type double to record your change then run the following python code in a field calculate tool
Code block is:
l = list() # remember lists start at index zero
with arcpy.da.SearchCursor("London","GrandTotal") as cur:
for row in cur:
l.append(row[0])
def offset(i,currentTotal):
global l
if i < len(l):
prev_month = l[i]
calc = ((currentTotal - prev_month) / prev_month)*100
return calc
In the example above I have hardwired it to read the table "London" but you could easily swap that out with an inline variable substitution from a model iterator as suggested by @Robert_LeClair