Hey There,
We have come across similar scenario for one of our client.  In our scenario , our website is getting lakhs of users daily on the along with Bots. Judgement of exact load on VM( Or app Service Plan ) was a real challenge. 
Load generally varies based on following factors : 
 
1. No of incoming requests 
2. Outgoing data ( Packet size ) 
3. Server Capacity to handle the requests - Faster the machine , more will the threshold.  
 
Using JMeter to configure load helped us to some extent BUT it is not 100% full proof. We have faced this problem in real time i.e.  using Jmeter  ,after testing it for specific Load(users , time duration ) , we were confident to move to production however our system couldn't sustain even upto 50% less traffic as tested with JMeter.  
I am still looking for something full proof from Azure which can help us to figure out the best costing mechanism. We should have a straight forward tools to configure/test these scenerios.
Just for clarity - We are using IAAS model (VM's) & Application is hosted on Windows machine , Developed on .Net Core ,MVC.