Introduction
The use of analytical methods has gained immediate relevance in recent years. The practice of obtaining useful information from data has helped several companies improve their business performance.
Analytics enable companies to get a clear picture of past and future events of their performance. A glimpse into the future helps companies prepare for misfortune (Yes, there is) that is about to arrive.
Through analytics, companies can find answers to three main questions: “What happened”, “What's going on” Y “What will happen”. It would not be wrong to say that the rise in data has driven this scandalous penetration of the use of analytics.
Analytics is not just limited to gaining insights from the past, but also makes it possible to predict future results and make the most of commercial resources. Due, the most advanced forms of analysis, namely, the predictive and the prescriptive, have acquired greater relevance in supporting the decision-making needs of institutions.
In this post, I have explained the 3 main forms of analysis that categorize all the alternatives for analytical models applied in all countries.
According to a study, Institutions that focus on basic automation to expand their reporting capabilities can improve their ROI in a 188%. Despite this, Adding advanced analytical implementations that improve the organization's strategy can extend your ROI by up to a 1209 percent.
Then, in principle, What are the different types of analysis?
1. Descriptive analysis
Let's start with the most basic type of analysis, In other words descriptive analytics. The objective of descriptive models is to analyze historical trends and discover relevant patterns to obtain information on the behavior of the population.. Descriptive analytics involves finding answers to “what happened?”. It is the form of analysis most used by institutions for their daily operation and in general it is the least complex..
Descriptive models use basic mathematical and statistical techniques to derive KPIs that highlight historical trends. The main purpose of the model is not to estimate a value, but get information about the underlying behavior. Common tools used to run descriptive analysis include MS Excel, SPSS y STATA.
A typical example in the banking industry would be customer segmentation. Historical data is extracted to analyze customer spending patterns and wallet engagement to enable targeted marketing and sales focus. Such models are powerful tools for population profiling, but they have limited predictive power regarding the behavior of individual members of the same population.
Helpful Resources:
- Online resources for learning basic descriptive statistics can be found at Khan Academy: Link
- Here is a video on how to run descriptive statistics in SPSS: Link
- An essential MOOC on Coursera- Data Scientist Toolkit: Link
2. Predictive analytics
Predictive analytics use statistical models to establish the probability of a future situation or outcome occurring. It involves finding answers to ‘What could happen?’.
Predictive models focus on descriptive models as they go beyond using historical data as the primary basis for decision making, often using structured and unstructured data from various sources. They enable decision makers to make informed decisions by providing a comprehensive account of the likelihood of an event occurring in the future. They encompass various advanced statistical models and sophisticated mathematical concepts such as random forests, GBM, SVM, GLM, games theory, etc.
A predictive model relies on a descriptive model to predict future behavior. Despite this, unlike a descriptive model that only profiles the population, a predictive model focuses on predicting the behavior of a single customer.
The tools used to run predictive models vary depending on the nature of the complexity of the model, despite this, some of the most used tools are RapidMiner, R, Python, SAS, Matlab, Dataiku DSS, among many others. Online resources on using these tools can be found on Coursera.
A typical example in the banking industry would be advanced campaign analysis. Can help predict the likelihood that a customer will respond to a given marketing offer to drive product cross-selling and upselling. Another example would be predicting the likelihood of credit card fraud..
Helpful Resources
- MOOC in Coursera on R for beginners: Link
- A Complete Guide to Python for Beginners: Link
- Building predictive models on Coursera: Link
3. Prescriptive analysis
Prescriptive analytics is the most sophisticated type of analysis that uses stochastic optimization and simulation to explore a set of possible alternatives and recommend the best feasible action for a given situation. It involves finding answers to “What should be done?”.
Prescriptive models go beyond descriptive models that only address what is happening, and beyond predictive models that can only tell what will happen, as they go to advise what should really be done in the predicted future. Quantify the effect of future actions on key business metrics and suggest the most optimal action.
Prescriptive models synthesize big data and business rules using complex algorithms to compare the likely outcomes of a series of actions and select the most optimal action to drive business goals. Most advanced prescriptive models follow a simulation procedure in which the model continuously and automatically learns from current data to boost its intelligence..
These models are generally more complex in nature and, therefore, are being used by some big progressive companies, since they are difficult to administer. Despite this, when properly implemented, can have a strong impact on the effectiveness of a company's decision-making and, therefore, in your final results.
Having said this, technical advances such as supercomputers, cloud computing, Hadoop HDFS, Spark, processing in the database, MPP architecture, etc. have made the implementation of complex prescriptive models that use structured and unstructured data much easier. The tools used to run prescriptive models are mostly the same as predictive models, despite this, require advanced data infrastructure capabilities.
A common example of prescriptive models in the retail banking industry is the optimal allocation of sales staff at various bank branches to maximize new customer acquisition.. By combining geographic location information with the performance and potential of each branch, the model can prescribe the most optimal allocation of sales staff in all branches.
A more sophisticated prescriptive modeling approach is used in airline ticket pricing systems to make the most of the price of airline tickets based on travel factors, demand levels, purchase time, etc. to maximize profit margins, but at the same time does not deter sales.
According to an investigation, around the 10% of institutions currently use some form of prescriptive analysis, this figure has increased from 3% in 2014 and is expected to increase by 35% in 2020. Factors like massive investments in predictive analytics, Expanding IoT capabilities that complement prescriptive analytics are driving this growth and expanding the reach of prescriptive models.
Helpful Resources (at the same time as those of predictive analytics):
- Guide to building a recommendation engine in Python: Link
- MOOC on Coursera for Hands-on Machine Learning: Link
- Guide to learning random forests: link
Final notes
In this post, I analyzed 3 different versions of analysis used in industries today. These are the building blocks of the analytics industry around the world. It is fair to say that all models, Developments and discoveries made with data can be categorized into any of these three categories.
This post is intended to help people who are new to analytics or planning to switch to analytics to get a clear view of the domain. I hope the above resources help you start learning.
About the Author
Sajal jain is an analytics professional with more than 6 years of experience in banking and workforce analysis. Completed his Masters in Statistics from the London School of Economics (a scholarship) and is currently working with a research-based technology and consulting firm in Gurgaon.
Do you have experience in Business Intelligence / Machine Learning / Big Data / Data Science? Show your knowledge and help the DataPeaker community by posting your blog.
