How To design a solution for any Problem using Data Science methodologies within any Corporation in Pakistan or any South Asia Country

Muhammad Ammar Jamshed
Nov 15 · 6 min read

As we know Data Science has created many exaggerated expectations and curioisty over its potential for many tech enthusiasts and corporations trying to adopt modern technologies. Before we get on how to employ Data Science for any work in a corporation in Pakistan I would like to dispel many myths and exaggerations surrounding this field. First of all;


Many people think Data Science can turn revenues from zeros to 6 figures and grow companies with 10 clients to a corporation with Millions of customers.

for that the answer is YES MAYBE TO AN EXTENT AND BLOODY HELL NO as well.

Being a Data Scientist Implies being able to understand and analyze the data’s past, present and estimate what its future could look like and how the organization relying on it should prepare accordingly.

What data do have of what type of company and how do we design our plan when faced with a small firm?


As a data Scientist or Analyst or machine learning engineer. We need to first look at business settings and organizational political thinking before starting our analysis.


Lets say for example Companies with 3 departments such as HR, Sales and Finance would in most cases be small firms or companies relying on a single type of product to be sold for revenue such as tea cup making factories.


When they decide to store their data it’s mainly to keep a track of their sales activity and anything related to it as they would not be interested in advancing human thinking or administrative procedures given their conservative business settings of ensuring 1 product is sold smoothly and correctly. A data Scientist in this case could be hired to predict sales of tea cups next year.


The Data Scientist for this project would only require the tea cups sold every month and feed it to exisiting Machine learning algorithms like Linear Regression for time period sales and get the prediction for next year and show time series forecasts on the side for comparison and multi dimensional analysis for the sales team.


In this case the Data Scientist would need to hire an app developer to convert his/her Machine learning code into a working app with a predictive interface because a small company like the tea cup making company would not have a cloud service or data pipeline to run Machine learning service on.


What can the data be used for if it’s a Multinational?

When looking about large companies in the south Asian region such as Multinationals.

We have to first evaluate their organizational eco system as in how department share and interact from different data platforms such as how Sales department shares it Sales data with Human resources to evaluate which employee is making the most sales and how should they motivate other employees to do the same and if Human resources can design training programs accordingly for it or redesign the bonus infrastructure to motivate performance of employees in conducting sales.

Should we restructure the data?

We need to first check whether the data is being shared via a ERP system like Oracle or SAP or a shared repository like One Drive or just a common platform like Microsoft Sharepoint.

We then need to evaluate the type of data that is valuable to all departments such that employees numbers would be useful to HT but useless to sales but Sales per Employee is the the type of data useful to both Sales and HR as they both can evaluate Revenue flows and employee development through it.

In the final stage we can decide should we make this data available in tabular forms like excel sheet through shared control or a SQL database with access restrictions on certain aspects by both departments as Sales cannot view employee salaries associated with their sales but HR can.

What algorithms should we apply?

We need to see if the outcome of the data is can be achieved with supervised or Unsupervised Machine learning.

For numerical based results mostly supervised is a preferred algorithm while for building recommender systems like that on Netflix or spotify or Deep level Analytics like analyzing Blackhole images on Massive datasets we use Unsupervised algorithms and for making automated AI softwares such Alexa or Bixby we mostly rely on Reinforcement algorithms.

After completely viewing the nature of the data and its strategical purposes we should conducts exploratory data analysis to see if we can find anything further like:

  • Are there any errors or outlier within it. which we can do through data visualization using Business Intelligence tools like PowerBI or Tableau or even Python programming.

Should we apply Predictive analytics or simple Machine Learning Modelling?

If our desired predictions want to compare past to present data based on future digits like:

  • Salaries issued.
  • Number of employees.

Then we would use regression based algorithms but if we want the prediction for the case in a binary outcome like yes or no like if the employee is fraud or not then we would go for a classification based approach.

Once we are clear with the business objective at hand like if we want to see the future prediction of employee salaries then we would use tree based algorithms like Decision tree or RandomForest depending on the payroll structure of the organization like if the salaries are dependent upon related monetary factors like issue of last months bonus or how many bonuses claimed per year then Decision tree would be used to encode related variables to make the predicted salary of the future. If this is not the case then we will go for Random Forest Regressor to keep other variables independent from being included in computing the prediction value.

What should the outcome look like?

This would entirely depend upon the requirments of the stakeholders or client.

Based on their ability of thinking like if they want to see it in a Multidimensional form then we would convert our machine learning code into a Dashboard or if they want to see it in form of an interface tool then we can convert the Machine learning code into a web app or API where you can get the output shown once you input your desired data.





This is summarized life cycle of trying to run and implement a Data Science based project in any south Asian country where you face cultural conservatism and organizational complexity in technological variables.

I hope you understood my article and I look forward to what you think about it in the comments below.

I’m currently studying MSC Data Science and AI from University of London and I’ve been a freelance Data Science Specialist on Upwork and have worked with individual clients and organizations on Business Intelligence systems and Machine Learning models.

Please subscribe to my medium profile for more articles and do connect me on LinkedIn if you are interested in sharing professional knowledge or work on projects together.

Note: All images are not my copyright and dont belong to me.

Analytics Vidhya

Analytics Vidhya is a community of Analytics and Data…