numbersci

Supply Chain Retail Analytics

ani.tekawade@gmail.com (Ani Tekawade) — Sun, 18 Sep 2022 01:23:52 GMT

Ever-changing consumer tastes, the proliferation of competition, and the turbulent supply chain environment create significant obstacles for retail managers preventing them from operating their businesses successfully. Retail managers can overcome such challenges by getting business insights to guide their decision-making process better. Using the right analytics tools on the vast amounts of data available today, managers can achieve operational excellence and generate valuable insights about product demand and customers’ preferences.

For example, a manager can predict consumer demand and measure the importance of various attributes such as brand name, product category, shape, size, or color. They can provide the most value to the customer using predictive analytics, generate higher customer demand, and increase product popularity. Understanding how product features impact the market also influences service levels (inventory, stockouts, variety), which affects the bottom line. Thus, predictive analytics is also crucial for conducting better inventory planning.

A manager can also evaluate the effectiveness of past promotions (e.g., advertisements, discounts, endcap displays, price reductions). Some products sell well without a promotion, whereas others need that extra promotional boost to be noticed by customers. Thus, having an advertisement for products that technically don’t need it carves into retailers’ profitability.

Another critical way predictive analytics can be used in managerial decision-making is by evaluating substitution and cannibalization effects among products. This happens when products share similar features, so the customer can easily replace the functionality of the preferred outcome with an alternative. The absence of a highly substitutable product in the assortment will not do as much damage as the absence of a product that customers find challenging to substitute. In the former case, customers will quickly find a product that will do a similar job, but in the latter case, customers can turn away and buy this product from a competitor. Measuring this effect requires a deep understanding of the dynamic relationship among products’ features.

Transactions data is generated every time a purchase is made and contains detailed information about the features of the actual transaction, such as product features (brand name, shape, size, color), marketing mix activity (price, promotion, discount), store characteristics (location, size, type), method of sale (online/offline), timing (season, day of the week, time of the day). Different types of data can be used for the analysis. Additional information can be obtained through own or third-party sources can also include:

Online product reviews.
Customers’ level of education/occupation.
Social media posts.
Scraping competitors’ website information.

We help our clients utilize all available information to make the most profitable business decisions using advanced analytics tools.

SCM & Data Analytics

ani.tekawade@gmail.com (Ani Tekawade) — Sun, 18 Sep 2022 00:39:21 GMT

The abundance and richness of data and advances in data analysis have allowed managers to gain the most detailed insights into their businesses. One of the ways data can be used to improve business profitability is customer analysis. Customer analysis utilizes different types of data to identify various customer segments, understand and predict consumer behavior, and influence customers’ purchasing patterns.

Depending on the task, data for customer analysis can be obtained through various sources. Managers can utilize their historical sales transactions data collected over time. Transaction data covers detailed information on the time/date/location of the transaction, product attribute information, payment type, marketing mix information (pricing, promotions, discounts), and other purchase-related information. Another way to understand consumer behavior is to utilize household purchase data (also known as panel data), where households continuously disclose their purchasing habits. This data is very rich because it includes not only product and transaction information but also detailed demographic (i.e., education level, number of children, profession, marital status, etc.) and attitudinal information (i.e., feelings, beliefs, opinions) and spans across all product categories consumed by participating households. Such data is typically available through third-party data collection agencies. Additionally, data can be scraped on the web targeting specific information (e.g., online reviews, Twitter posts, mentions, etc.).

The simplest use of this data is descriptive analytics, which examines seasonal trends, purchasing patterns, correlations, outliers, growth rates, day-to-day changes, and so on. This type of analysis answers the question, “What happened?”. For example, one can detect a dip in historical sales or a spike in prior demand due to an external event that took place during that time frame. Data visualization tools can include charts, plots, graphs, and diagrams, and descriptive statistics can cover frequency distributions, central tendency measurements, variation, and percentiles. Numerous descriptive analytics tools exist on the market today. All statistical programs that provide descriptive statistics, such as R, Stata, Python, SAS, or Mathlab, also have excellent visual capabilities. Programs like Power BI, Tableau, and Alteryx, just to name a few, provide state-of-the-art visualizations, including highly interactive dashboards, graphs, and reports.

In contrast to descriptive analytics, predictive analytics takes a step further and beyond describing what has happened in the past and makes advanced predictions about what could potentially happen in the future. This type of analysis finds answers to the question,” What might happen?” Building an appropriate statistical model and using historical data to predict an outcome based on the model can be done in various powerful statistical analysis tools available using regression analysis, machine learning, and artificial intelligence algorithms. If data is not available, one can use a simulation model instead.

Descriptive analytics

What happened?

Dashboards, reports, bar charts, line graphs, scatter plots

Tableau, Alteryx, Power BI, Stata, Python, R, SAS

Predictive analytics

What might happen?

Regression analysis, simulation

Stata, Python, R, Matlab

Prescriptive analytics

What should happen?

Optimization

Gurobi, SAS, Matlab, Mathematica

Finally, prescriptive analytics offers a specific solution/recommendation based on the insights obtained in predictive analytics. This type of analysis answers the question, “What should happen?” A decision maker feeds predictive analytics values, business rules, and constraint information into an optimization model and, depending on the goal (e.g., maximize profits or minimize costs), determines the exact course of action to achieve this goal.

At NumberSci, we have developed a proprietary approach for collecting information quickly, optimizing and developing a solution that is the best fit for your company’s needs but also benchmarked against your competitors, utilizing the latest market and academic research to gain descriptive, prescriptive, and predictive insights into your business using advanced analytics tools.

Substitution Effects on Retail Products

ani.tekawade@gmail.com (Ani Tekawade) — Sun, 18 Sep 2022 00:39:19 GMT

When building predictive and prescriptive models, retail managers often struggle to capture the substitution rate among products they carry. This happens when products that share similar attribute features each other are easily substitutable, which allows the customer to easily replace the functionality of a preferred product with an alternative. For example, if two different snacks are highly substitutable (potato chips vs. veggie chips), in the absence of potato chips on a shelf (e.g., a temporary stockout), the customer who loves potato chips will buy veggie chips instead. In this case, the manager will be able to capture the demand through the sale of another product. In contrast, if the two snacks - potato chips vs. veggie chips - are not substitutable, the absence of potato chips on the shelf will cause the customer to turn away, and the potential sale will be lost. Unfortunately, managers rarely know the substitution rates of each SKU and often resort to heuristics and intuition.

Here is another example. Let’s assume a manager is looking to reduce their product assortment and seek to remove a single SKU from its assortment while choosing between SKU 1 and SKU 2, where each is selling ten units a week and generating the same profit margin per unit. At face value, the decision-maker is indifferent about which of the two SKUs to remove, as both SKUs would seemingly lead to the same unit loss of 10 units. However, in the table below, we illustrate how, unbeknown to the decision-maker, SKU 1 has a higher substitutability rate (0.7 vs. SKU 2's 0.2), which leads to the recapturing of a larger portion of SKU 1's lost sales. Here substitution rates capture the percent of customers that would switch, should an SKU not be available, to an alternative SKU that is available. Once the substitution rates are accounted for, it is clear that the decision-maker should eliminate SKU 1 and keep SKU 2.

Measuring this effect requires a deep understanding of the dynamic relationship among products’ attributes. And the outcome of this analysis has a significant impact on the bottom line of a retailer. Correctly measuring the substitution effects among products will influence the overall supply chain strategy, including inventory planning, assortment selection, and revenue-cost analysis, to name a few.

Data Science vs. Data Analytics

ani.tekawade@gmail.com (Ani Tekawade) — Sat, 17 Sep 2022 23:58:58 GMT

In recent years, it cannot be easy to keep up with the rapid proliferation of various data- and technology-related terms. In this article, we offer some explanations of the most commonly used words.

Data science and analytics are the terms sometimes used interchangeably by the scientific and practitioner communities. Although quite similar, there are some profound differences between the two - 1) the depth of scientific and computing training and 2) the amount of data.

Data Analytics

Data analytics (also known as business analytics) uses available datasets to draw business insights. In close collaboration with other organization members, a data analyst uses various visualization tools and statistical modeling to identify business trends and patterns or describe what happened in the past business period. A data analyst is responsible for identifying and explaining a business problem, seeking a meaningful solution, and guiding the decision process using excellent communication, visualization, and presentation skills. Data analysts have excellent knowledge of visualization and database tools such as Tableau, Alteryx, R, Python, Power BI, SAS BA, and many others. Data analysts typically have a master's degree or certification in business analytics. Budget forecasting and revenue prediction, risk evaluation, lead-to-customer conversion rate calculation, and sales projection are examples of data analytics.

Data Science

Data science is what we call data analytics on steroids. It is built on in-depth statistical, mathematical, and computing knowledge to deal with vast amounts of raw, disorganized data. Data scientists will use their immense coding and computing knowledge to "mine the data" - collect raw data, organize it, and build meaningful descriptive, predictive, and prescriptive models. One of the main capabilities of a data scientist is their ability to work with very large amounts of different data types. They can use cloud computing to work with millions of observations at once and analyze numerical, categorical, image, audio, and video data types. They leverage their extensive academic training in statistics, mathematics, probability theory, machine learning, and artificial intelligence to draw actionable business insights. Data scientists' problems are typically very complex, difficult to identify, and require sophisticated statistical analysis or machine learning modeling. Commercial tools used by data scientists include Python, TensorFlow, PyTorch, TPUs (tensor processing units), Amazon SageMaker, Azure Machine Learning, Databricks, and so on. Many data scientists are trained PhDs in math- and coding-heavy fields, including biostatistics, economics, and business. An in-depth understanding of customers' demographics and behavior, detection and prediction of fraudulent activity, movie recommendation system, sentiment analysis, and customer segmentation are examples of data science use.

NumberSci uses many data analytics tools to help its clients identify and solve the most challenging problems.

Leveraging Data to Optimize Your Supply Chain

ani.tekawade@gmail.com (Ani Tekawade) — Thu, 08 Sep 2022 05:51:48 GMT

The abundance and richness of data along with advances in data analysis have given managers an opportunity to gain the most intricate insights into their businesses. One of the ways data can be used to improve business profitability is customer analysis. Customer analysis utilizes different types of data to identify various customer segments, understand and predict consumer behavior, and influence customers’ purchasing patterns.

Depending on a task, data for customer analysis can be obtained through various sources. A manager can utilize their historic sales transactions data collected across time. Transaction data covers detailed information on the time/date/location of the transaction, product attribute information, payment type, marketing mix information (pricing, promotions, discounts), and other purchase-related information. Another way to understand consumer behavior is to utilize household purchases data (also known as panel data), where households continuously disclose their purchasing habits. This data is very rich because it includes not only product and transaction information but also detailed demographic (i.e. education level, number of children, profession, marital status, etc.) and attitudinal information (i.e. feelings, beliefs, opinions) and spans across all product categories consumed by participating households. Such data is typically available through a third-party data collection agencies. Additionally, data can be scraped on the web targeting specific information (e.g. online reviews, twitter posts, mentions, etc).

The simplest use of this data is descriptive analytics that examines seasonal trends, purchasing patterns, correlations, outliers, growth rates, day-to-day changes, and so on. This type of analysis answers the question, “What happened?”. For example, one can detect a dip in historic sales or a spike in prior demand due to an external event that took place during that time frame. Data visualization tools can include charts, plots, graphs, diagrams and descriptive statistics can cover frequency distributions, central tendency measurements, variation, and percentiles. Numerous descriptive analytics tools exist on the market today. All statistical programs that provide descriptive statistics such as R, Stata, Python, SAS, or Mathlab also have excellent visual capabilities. Programs like Power BI, Tableau, Alteryx, just to name a few, provide state-of-the-art visualizations including highly interactive dashboards, graphs and reports.

In contrast to descriptive analytics, predictive analytics takes a step further and beyond describing what has happened in the past makes advanced predictions about what could potentially happen in the future. This type of analysis finds answers to the question,”What might happen?” Building an appropriate statistical model and using historic data to predict an outcome based on the model can be done in various powerful statistical analysis tools available using regression analysis, machine learning, and artificial intelligence algorithms. If data is not available, one can use a simulation model instead.

Finally, prescriptive analytics offers a specific solution/recommendation based on the insights obtained in predictive analytics. This type of analysis answers the question, “What should happen?” A decision maker feeds predictive analytics values, business rules, and constraint information into an optimization model and depending on the goal (e.g., maximize profits or minimize costs) determines the exact course of action to achieve this goal.

At NumberSci ,we have developed a proprietary approach for collecting information quickly, optimizing and developing a solution that is a best fit for your company’s needs but also benchmarked against your competitors utilizing the latest market and academic research to gain descriptive, prescriptive, and predictive insights into your business using advanced analytics tools.

Supply Chain Disruptions

ani.tekawade@gmail.com (Ani Tekawade) — Tue, 06 Sep 2022 04:57:13 GMT

The past three years have proven to be exceptionally difficult for many businesses. At the onset of the COVID pandemic in 2020, as many as 630 US companies declared bankruptcy, the highest of the past decade (Irum 2021). Those who have barely survived continuously faced severe shortages of critical components and parts due to worldwide COVID-induced factory closures (Baker 2021). In today's time of significant uncertainty, the managers should continuously monitor the events by asking themselves, "In which environment do I operate?" This process of environmental scanning allows the company to track changes, identify threats and opportunities, and re-align business strategies to survive in an unfamiliar environment.

Anticipating potential disruptions, managers need to reconsider one of the core postulates of lean manufacturing that suggests that businesses should keep as little inventory on hand as possible. For decades, excess inventory has been considered wasteful, and the skill to keep production lines as 'waste-free' as possible has been an ultimate achievement and source of pride of any successful lean manufacturing manager. However, in today's world of uncertainty, when disruptions occur frequently and unexpectedly, this lack of buffer inventory can result in millions of dollars worth of lost future sales. The manager can reduce their risk of disruption and protect their production process by accumulating additional inventory that they would typically not need during normal, non-turbulent times.

Executives can reduce the risk of disruption by reducing the number of own suppliers. Supplier network typically consists of a vast web of Tier-1, Tier-2, and Tier-3 suppliers. Sometimes, the tiers extend beyond the third level making the supply chain so poorly visible to the original manufacturer that they no longer know who produces what and how. When one member of this complex supply chain gets disrupted, the whole supply chain falls apart as a result. Hence, the managers should focus on reducing the number of Tier-1 suppliers and working closely with them to seek efficient reduction of their own supplier network.

As a modern supply chains get more and more complex (e.g. Apple has as many as 608 first-tier suppliers just across China, Japan, US, and South Korea alone (O'Connor 2018), keeping the finger on the pulse of the surrounding environment by dealing with supplier relationship processes, product development, fulfillment, and customer relationships in a moment of crisis becomes a very stressful job.

Using data- and AI-driven supply chain solutions in environmental scanning and risk assessment, a firm can identify, address, and mitigate external environment risks in order to prevent potential disruptions. NumberSci brings to the table a unique skill set based on rich practical experience and extensive academic training to help supply chain managers make the most profitable decisions in risk modeling, identification, scoring, and prevention.