DDLJ: Data’s Deep Learning Journey
Today we will start a series of “DDLJ”. First Topic of series is Monte Carlo Simulation integrated with Machine Learning.
Monte Carlo, very simply, can be said as a method to understand the relationship between multiple variables using random sampling.
Monte Carlo helps you to understand the different possible outcomes from your decisions and hence can help you to assess the impact of “Risk”

Business Benefits:
Today, Monte Carlo simulations are used to
- Staff a Call Centre by understanding/experimenting on different scenarios on the basis of Average Time and Volumes of Calls …
- Plan investments by experimenting on past returns (and their fluctuations) from multiple investments …
- Study traffic patterns in multiple scenarios and experiments on Frequency of Vehicles, Volume/Number of Vehicles, Signal times …
- Plan Inventory or Safety Stock by experimenting on Average Demand and its variation, required Customer Service Level …
Lets first understand how Monte Carlo Simulation helps to solve a particular Business Case/Problem and then lets understand how it can complement AI/ML.
Business Case:
This is an hypothetical use-case, based on my experience. Suppose your company is about to release a data product, lets assume the product is a Platform that helps Defects Reduction Problem or improves productivity. I am keeping it general so that the idea can be extended to any domain.
Objective is to estimate the first year net-profit from the product. There are three major components or decision variables that will effect profit.
- Volume/Number of Products sold
- Selling Price
- Costs (Variable+Fixed)
Lets discuss more about the business case, when we start formulating a solution.
Before we start building the solution, below given is the general approach to formulate and Solve a Business Problem using Monte Carlo Simulation.
Monte Carlo Simulation Approach:
- Identify/Formulate the Transfer Equation and Range for Optimisation
- Identify/Define the input parameters and Variability
- Create Random Data
- Run/Simulate and Analyse
Using the above approach, we will try solving our business case
Based on a market research, input from Client Partners and interactions with Customers, there are equal chances for the market to be niche, normal or mass.
So in case of niche market, lets assume you will be able to deploy 10 licenses at 500k dollars/license. For normal and mass lets assume the figures are 20 and 40 licenses at 300k and 100k dollars respectively. (For normal and hot , assumption is the targets clients are smaller organisations with lesser ROI and hence the licensing cost is less).
So the net profit is a function of 2 random variables — number of licenses sold and Licensing Price.
What you can see here is, we have already completed our first 2 steps of the MCS approach. Now lets move to the next step i.e. creating random numbers using the average(Normal Case) Price and Number as the most likely estimate and the other 2 cases as the ranges for the random number. (We can assume a distribution here, lets assume normal distribution, to create data).
For each scenario, net profit will be calculated and average net profit calculated will be the final estimate is your business objective defined above.
Same could have been calculated using averages directly and be used to make a business decision related to operations/staffing/geography…
Monte Carlo helps you to understand the risk of uncertainty in the business decision and can help you to plan better for the positive and negative impacts of your decision.
You can try simulations for 10000 runs and compare both the methods of averages and MCS. It will help you to appreciate the MCS method.
Now, what about machine learning?
Think of the very first step we talked about having a transfer equation. What if we dont have any such equation at your disposal? Think of it.
Also read the paper in the link below for more understanding and I will come with more details on the integration of Monte Carlo and Machine Learning in my next blog.
“Only those who will risk going too far can possibly find out how far one can go” — T. S. Eliot
