Learning Diary
Notes for self-reflection from IIM Ahmedabad Post Graduate Diploma in Advanced Business Analytics
[April 28, 2024]
What you cannot measure, you cannot improve.
Extending this thought from Peter Drucker:
What you measure, is only what improves.
So, if you measure the wrong metric, you improve in the wrong direction.
Hence, it becomes crucial to measure the right metric as a yardstick of growth. This here is an attempt to simply record one thing.
What I know now that I did not know last week?
This will then become living diary of reflections on external inputs and introspections within to lay out in simple terms my learnings at the level of life, career, institute, program, module, course and session. There will be cross-learnings as well and certain infusions from the peer group as well that cuts across the granularities mentioned above.
1. Life
[April 28, 2024]
Expertise is not signalled by how much you know. People don’t turn into wikipedias. It is about the number of higher tiers of conceptual chunking you are able to do of the subject matter and how new information fits into your knowledge model and existing information is articulated and applied. Simply put, the more connected complex chunks you build that reduce the total number of simpler chunks, the higher your expertise becomes.
2. Career
[April 28, 2024]
Quarter-to-Quarter helps address seasonality which occurs due to quarterly target setting and natural patterns. It is also called hockey stick pattern in sales.
3. IIM Ahmedabad
3.1 Philosophy of Pedagogy
[April 28, 2024]
While other schools merely adopted the case study method from Harvard, IIM-A has a six decade history of learning case study method from its early founders at Harvard. These roots are felt in the teaching style used by professors at IIM-A. The course sessions are built around carefully picked cases which reveal the learnings to students through a in-depth story of an organisation. Most of the cases used are from HBS itself. Even when you read the case and make your decision sheet before class, the 75 minutes of class by the end will reveal something that the case had all along just hidden in plain sight. The secondary learnings of a industry and real-world scenarios are an add-on in the case method. At fortunate times, you might also learn from valuable points from peers in the case discussion. Best, if someone challenges your point of view and pushes you to think deeper or consider alternatives.
See this article from the IIM-A Archives to know more.
Another philosophy is keeping things simple and away from jargons that forces you to think from fundamental principles and not hide behind cliche/complex sounding terms.
3.2 Evaluation
[August 27, 2024]
Similar to other schools, IIM-A follows a credit-grading system. A unit or credit means studying one hour per week for 25 weeks i.e. one semester. This is the measure of the amount of course load.
Evaluation of these units is done on a 4-point passing scale from A to D where A means Grade Point of 4.0 and D is 1.0 out of 4.0. F is the Fail grade. These grades and grade points are assigned from grade averages calculated from the components of the course. Each component is evaluated on this four point scale further divided with 3 levels each (+, neutral and -). To keep calculations simple, these component grades are assigned values from 2 to 13 out of 12 corresponding to D- to A+. Yes, A+ means scoring more than maximum on your assignment! Finally, these components are simply multiplied by their weightage for the course and overall grade average is divided by 3 giving the grade for the one credit or one unit. If it was a 0.5 unit course, component grade average, overall grade average and grade point will be multiplied by 0.5.
In a class, the grading is relative but the bucket sizes for each grade A,B,C,D and F are undisclosed publicly.
4. Post Graduate Diploma in Advanced Business Analytics
This section describes the :
- Overall points of observation related to the diploma; and
- Cross-module learnings.
This section does not include these:
- Cross-course learnings which are highlighted within each module.
- Course level learnings which are within their respective sections.
4.1 Diploma Overview
[August 11, 2024]
This course lies between a business program and rigorous data analytics program.
[April 28, 2024]
Just like Toyota treats its employees with respect, IIM-A treats all its students with respect regardless of your program.
[May 12, 2024]
I finally understood, what ‘Advanced’ means for Business Analytics. It does not necessarily mean using the most cutting edge theories of analytics or the most complex algorithms that most would not have heard of. That is not the goal, neither is it reflected in the course curriculum. What ‘advanced’ means is the absortion of knowledge to a depth, establishing clear fundamentals and correct application of knowledge from across the modules. This is facilitated by high quality professors and teaching material.
[June 23, 2024]
This network illustration is an attempt to understand the links of courses in series of modules across the academic areas.
4.2 Cross-Module Learnings
[July 11, 2024]
There are multiple cross-module learnings. Pre-term statistics set the foundations for the insight formations in module 1 for statistics concepts such as confidence intervals and hypothesis testing. Module 1 is about data.
[August 20, 2024]
Further, module 2 which is on models, takes these concepts and apply them in the context of categorical variables through estimators.
5. Module 0: Pre-Term
5.1 Introduction to Python
[April 28, 2024]
The scientific calculations in python done using sci-kit learn library are not exact. Since, these are done using sampling. For example, the integration results will be accurate to some decimal places but not the ‘exact’ values.
5.2 Basic Statistics and Probability
[April 28, 2024]
Inference is the bridge from sample to population.
[July 9, 2024]
Every probability and statistics class is not the same. Elementary school introduced these concepts at the age of 13 with simple measures of central tendencies such as mean, median and mode; and frequency based probabilities, high school extended these concepts to histograms and variance and later college took us to distributions and inference. This course of IIM-A did not simply repeat these concepts. The pre-module content focussed on building core concepts that directly translate into application. The entire pre-module course did not once mention random variable or distribution. The module 1 course covered inference in detail, again with a focus on building core intuitions around the concepts which can be extended beyond the theoretical knowledge.
[August 20, 2024]
Sampling can be done in various techniques such as simple random, clustered or stratified. If population has similar groups but each group is diverse internally, picking few groups makes a good sample. E.g. people living in townships across India. Picking few townships as samples. If population has different segments which are dissimilar but internally homogenous, go for stratified sampling. E.g people from different age groups in a population. Pick few samples proportionately from each group.
5.3 Introduction to R
[July 9, 2024]
It is important to think in terms of vectors instead of loops for writing optimised codes.
5.4 Basic Linear Algebra
[April 28, 2024]
One of the most underrated subjects, it forms the basis for most of the ‘magic’ beyond computer graphics and human-machine interactions.
Spectral Decomposition is a beautiful concept which underlines an algorithm to convert a coordinate system heavy linear transformation into a set of rotation -> basis scaling -> reverse rotation. Take this to your own life, you can have a revelation of ideas simply by changing the basis of axioms on which you make your mental models of reality.
6. Module 1: Data
Cross-course learnings in this module include the link between Business Analytics and Data Visualisation. Business Analytics stack is built on top of IT stack. The descriptive components are rendered in data visualisation channels. Later, with a significant leap of technical maturity of the organisation, predictive analytics is built on top of the descriptive analytics.
6.1 Business Analytics
[April 28, 2024]
Delineate your assumptions.
Sometimes, there are illusions of choice in life. It may be a Hobson’s choice of ‘take it or leave it’ which is no true choice. Or it may be a catch-22, a paradoxical choice of two options that can’t be traversed without the other and hence leave no real choice. A dilemma is when you have choices but no good option.
Computational resources are helping us take the unstructured data (BLOBs) and turn them into ‘structured’ types for analysis.
Uncertainty is caused by lack of information.
Statistics is the quantification of this uncertainty.
Probability is the toolset which helps in decision-making on top of these statistics.
Planning Analysis helps in predictive maintenance. Often relied by suppliers of products to retain control from the customers of these products or use rental business model over one-time sale. Most common use cases in large equipment suppliers in power-plants, oil refineries and other heavy industries; medical equipment like MRI/CT-scan machines where manufacturers rent out these machines by pay-per use and timely predictive maintenance is in their best interest.
6.2 Data Visualisation
[April 28, 2024]
Data visualisation is done for:
- Analysis — Interactive dashboards
- Communication — Static reporting with annotations
Often a table is the best visual and no graph is needed. If you need to compare individual values, table is the best option. Only for trends should you even think of any graphs.
Even if a designer, knows everything there is to know, that knowledge might not transfer into the right application.
A bar chart must start from 0 since, the bars are a visual representation of quantity. A trend chart/line chart may not start from 0 since, the visual representation is over a trend with delta/differences and not absolute values from x-axis.
We scan from top to bottom left to right (based on language). Hence, place the important items on top and left. Dictates ordering of ascending or descending in ranked horizontal bar chart.
Right skew shifts to left on a distribution plot (frequency polygon or histogram)
Box-plots with multiple categories should sort their series carefully. (Only for reporting not for analysis).
Use maps only when there is a spatial trend.
How to add a time dimension to correlation graph?
Option 1: Use dual axis
Option 2: Use color gradient
Option 3: Connect the dots using a line
Tufte’s recommends to use more intensity of the same color if the size is smaller for visual consistency. Hence, scatter plot should use more intense color for dots than a bar chart on the same dashboard just to maintain the balanced perception of same color.
Legends can be replaced with the series labels in the color of the series next to the lines themselves.
How to decide which plot to use for time-series data?
Heatmaps can be used in multivariate analysis by simply adding column for each new variable.
6.3 Operations
[April 28, 2024]
Flows of material, information and cash.
Little’s law of average throughput.
Supply-Demand at levels of operation, tactics and strategy.
Tradeoffs.
Efficiency vs Effectiveness
Organisation layout — Process vs Product
Toyota Production System
Cycle time is a better measure than overall capacity in terms of volume
Vicious procurement cycle in arms-length/arms-twisting relationship
Virtuous procurement cycle in arm around relationship
Kraljic Matric for procurement strategies
If you could only ask two questions to understand a company?
Q1- Volume — High/Low
Q2- Variety — Low/High
6.4 Big Data Management
[April 28, 2024]
It takes guts to start your course with the learning that not every problem can be solved using Big Data Analytics (BDA). This revelation was shocking.
[June 23, 2024]
This course was a ten 75-minutes session long course that ended a month ago. It was a well-designed course split between strategy-technical-deep dive in 30–30–40. A top-down approach from CXO suite to management to seamless transition to technical level designs and then into hands-on exercises.
6.5 Probability and Statistics
[May 12, 2024]
We just covered conditional probability for 3 hours. And the term ‘Bayes Theorem’ was not even used once. This shows the focus on the actual concepts and not the nomenclature. Most textbooks and college courses stop at how the accuracy of a medical test is high, yet the chance of actually having the disease, knowing the test came out positive is quite low. And with that the concept of conditional probability is covered. The students are left with little confidence in the medical tests and half baked clarity of conditional probability. Here, this was just the beginning. This was followed by understanding the role of prevalence and it’s impact on the conditional probabilty to further build a correct intuition model. Later, the actual medical solutions to cover this were discussed such as conformity tests and repeat tests. Prisoner’s paradox and real life cases like Sally Clark. The role of conditional probability in modelling especially in LLMs and recommendation systems was covered in detail with fundamental business differentiation of Amazon and Big Basket. All this just build on top of a simple concept like ‘Conditional Probability’. This is what makes it advanced.
[June 23, 2024]
In today’s discussion of confidence interval estimation under statistical inference using Normal distribution with unknown mean and variance, a simple question mid-way was thought provoking enough to reveal the fundamental interpretation and philosophical differences between Bayesian and frequentist approach. The Prof. was not dismissive, rather gave an entire hour to this because it highlighted the importance of curiousity, thought-provking questions and avoiding the ignorance and false pretence of understanding something without deep introspection and questions.
7. Module 2: Models
7.1 Applied Causality and Experiments for Business
[July 21, 2024]
To be studied
7.2 Bayesian Analysis
[July 21, 2024]
To be studied
7.3 Big Data Analytics: Analytics of Text and Social Media Data
[July 21, 2024]
To be studied
7.4 Business Simulation
[July 21, 2024]
Simulations are an alternative to analytical models with stochastic processes. Similar to Finite Element Analysis in mechanical engineering, business simulations can help identify bottlenecks and generate performance metrics from particular scenarios.
- Monte carlo simulations are simple models without any spill-over effect i.e. one entity does not affect another in the system. (Newsboy model). These can be modelled in Excel. These are also the bedrocks of Financial models. Hence, the extensive use of excel in finance industry.
- Discrete Event Dynamics includes spill-over effects where one entity affects another such as in a model of restaurant one customer delay would impact later customers. Softwares such as Arena allow for basic Discrete Event Dynamic models. More recent softwares such as Anylogic also support agent-based simulations and system dynamics.
[August 15, 2024]
Investors, Don’t replace your range by average! Because financial models have non linear functions with respect to time (such as compound interest), the time range of a certain investment must not be replaced by the mean time to evaluate ROI or make investment decisions. For example, if certain investment will be held for 2–5 years. The ROI calculations done using 3.5 years as a ‘shortcut’ will be inflated by ~15% more than a simulation done using uniform distribution in 2–5 years range. The wider the range of ‘uncertainty’, the bigger the difference between simulation and analytical methods.
[August 20, 2024]
Business simulations can be powerful decision making tools where analytical methods are just not enough. They can help identify bottlenecks, answer “what-if” questions and give power in the hands of business owners and managers to experiment and optimise. These are highly interpretable and low cost alternatives to pilot projects or AB testing when done right. For example, see this project created from scratch as part of course end submission.
7.5 Categorical Data Analysis
[July 21, 2024]
To be studied
7.6 Machine Learning with Big Data
[July 21, 2024]
To be studied
7.7 Model Thinking
[July 21, 2024]
To be studied
7.8 Network Analysis
[September 11, 2024] [Narrated and then grammatically refined using AI]
Network theory is not a new concept, as it is introduced in most undergraduate engineering programs, particularly in computer science. However, this course provided new insights into the practical applications of network theory.
In practice, most network analysis focuses on the giant component. This often means around 30% of the nodes are excluded from the analysis, as they are either isolated or form smaller minor components.
Additionally, we explored how central nodes in a network are identified. The simplest method is to calculate degree centrality, which identifies nodes based on their number of direct connections. This creates multiple layers of node importance.
Another method is closeness centrality, which measures how central a node is by calculating the average distance to all other nodes. Nodes with the shortest average distance are considered to have high closeness centrality. It is common for a node’s neighbors to have similar closeness values, and this measure tends to taper off smoothly across the network.
A third method is betweenness centrality, which considers how often a node acts as a bridge along the shortest path between two other nodes. Unlike closeness, betweenness centrality is not continuous, and typically highlights critical junctions within the network.
Finally, we explored eigenvector centrality, which measures how influential a node is based on the influence of its neighbors. This measure is iterative, and as the number of iterations increases, the centrality values converge, often resulting in one node having the highest eigenvector centrality.
Lastly, we examined how real-world graphs evolve over time. These networks often follow a power-law distribution, where nodes with higher degrees initially attract more connections in the future — commonly known as the “rich get richer” phenomenon. This principle is clearly observed in networks generated through models like the scale-free network.
7.9 Non-linear Optimisation
[July 21, 2024]
To be studied
7.10 Optimisation Problems in Business
[August 12, 2024]
Never work with unit profits! Use unit contributions/revenue. This is because in optimization problems, profits will not have a linear relationship due to fixed costs, but revenue/contributions will.
Shadow price can help assign value to every resource you have. It is a crucial negotiating lever too. And a very real way to capitalise information. For example, for a company’s perspective, two engineers each earning 24 lakhs per annum might have very different real costs to the company. An engineer working on a 1 crore project, whose absence can lead to loss of 40 lakhs is worth more than an engineer whose absence from a 10 crore project has no effect. The second engineer, is just leading to actual cost of 24 LPA while the first is potentially covering a cost of 40 LPA.
Duality is a less understood but important topic. It has a strong application in real life. An intuition is, for example — a maximisation problem of your salary for you is a minimisation problem for the company. You are constrained by the pay band, while the company might be in the position to decide the range of pay bands.
7.11 Panel Data Analysis
[July 21, 2024]
To be studied
7.12 Regression Analysis
[July 21, 2024]
TBU
7.13 Time Series Analysis
[July 21, 2024]
TBU
