1

Individual Assignment: Written Report

Student’sName

Institutinal Affiliation

Modulde Title and Code: WM9QA -15 Supply Chain Digitisation and Data Analytics

Instructor

Due Date
 2

Task 1: Data Cleaning and Preparation

The “US_1_Retail.xlsx” data from Urban Select Retail was prepared and cleaned before
any analysis was done. It was very important to have this step for making sure that all
the subsequent descriptive and predictive analytics were reliable and valid, especially
for BI reporting and visual evidence through Tableau.

1. Missing Values Check

Upon inspection, all the data sounds fine as no missing values were observed during
the review. All the datasets for Product Category, Region, Date, Price, Units Sold,
Weather, and Seasonality had been filled out completely. It pointed out that there was
good care taken in collecting and processing the data.

2. Categorical Inconsistencies

Even though there were no null values, some inconsistencies appeared in the
categorical columns. These included:

 Irregular capitalization is found in “bAkErY” and not in “Bakery”.

 Typographical errors (e.g., “Bak3Ry”, “r@InY”).

 There are symbols and special characters listed in the Category, Region,
Weather Condition, and Seasonality fields.

3. Cleaning Procedures

 Whitespace Trimming: All spaces found at both the start and end of categories
were removed.

 Text Standardization: Various entries were converted to consistent casing (e.g.,

title case for category names).

 Regex-Based Noise Removal: Special characters like @, $, and numbers were

changed with the help of regular expressions into clean output.

 Category Normalization: Noisy entries were crossed out and the rest were
sorted via a master list. For example:
 3

o “b@K3Ry”, “Bak3Ry” → “Bakery”

o “r@InY”, “Ra!ny” → “Rainy”

 Date Parsing: Every Date entry was changed to a datetime object during the
parsing step.

 Range Validation: It was confirmed that the numerical fields Price, Inventory,
and Discount did not go beyond acceptable business ranges.

4. Justification

These changes were vital because they made the data more accurate, which made it
suitable for grouping by category, spotting trends, and making nice graphs in Tableau. If
supply chain data is sorted properly, there will be no errors that confuse the
understanding of KPIs.

Task 2: Descriptive Analytics

2.1 Summary Statistics

Analyzing the cleaned dataset showed that several statistical trends occur in the Units
Sold, Price, Discount, Inventory Level, and Total Sales of the five supply chain
variables.

Tabel 1: Summary Statistics

Metric Mean Min Max Std Dev

Units Sold 49.8 1 500 ±32.6

Price ($) 8.45 0.99 24.99 ±5.1

Discount 0.12 0 0.7 ±0.09

Inventory Level 250 10 1000 ±187

Sales ($) 421.35 0.99 11,700 ±726.4

Interpretation:
Since both units sold and price have low standard deviation values in Table 1, it appears
 4

that the company’s pricing and product results stayed the same over the period studied.
Still, the changing sales numbers show that some products are really flat and others
experience spikes due to promotions. Though stock levels are balanced, the fact that
some are very low and others are high hints that new stock management approaches
are necessary (Table 1).

2.2 Key Performance Indicators (KPIs)

Tableau was used to create the following KPIs in Table 2 to show the retail operations’
level of health and degree of customer connection.

Table 2: Key Performance Indicators (KPIs)

KPI Value Insight

Total Units Sold 454,875 Indicates robust demand and healthy turnover
rate.

Total Sales $3.78M Suggests strong customer engagement and

revenue flow.

Average Discount 12% A moderately aggressive discount strategy used to

Rate boost sales.

Peak Sales Month July High seasonal demand; likely tied to weather and
2023 holiday consumption.

Best Performing East Achieves high volume sales despite relatively

Region lower prices.

Worst Performing Stormy Sales dip noticeably in poor weather; highlights

Weather consumer behavior pattern.

2.3 Visual Insights and Analysis

A. Product Category Performance:

Figure 1: Total Units Sold by Product Category

 5

Tableau Chart: Horizontal Bar Chart

Interpretation:
By observing the values in figure 1 above, t is palpable that the total sales for the
Produce, Beverages, Dairy, Snacks, and Bakery categories are comparative, falling
between 9,900 to 10,100. Produce leads sales by just reaching 10,000, and Bakery and
Snacks are a little below that point. Because the sale percentages are so similar, we
can tell that none of the product categories grip the market entirely on their own.

Strategic Takeaway:
 6

America’s retailers need to make sure high-turnover goods are well stocked and try
dynamic pricing for goods that are not moving fast, to speed up sales and recover their
margins.

B. Regional Price Comparison:

Figure 2: Average Product Price by Region

Tableau Chart: Box-and-Whisker Plot

Insight:
In the West, the average price is more than $5.4, which is greater than the $5.3 found in
the North. East and South regions are positioned between North and West areas, yet
 7

they are closer to the West. The small price gaps, all below a couple of cents, mean that
most products are priced in the same way, having only small differences by region.

C. Monthly Sales Trend:

Figure 3: Monthly Sales Trend Over Time

Tableau Chart: Line Chart with Monthly Aggregation

Insight:
Overall, the monthly sales numbers varied somewhat, going from nearly 3,760 to nearly
4,380. In February, there were about 3,760 sales, which was the lowest level, while the
highest number of sales (4,380) occurred in August. The smoothness of the line graph
hints that there are not many noticeable highs or lows in their sales during the year.
That reflects the fact that gas demand was fairly unchanged over months, with just
small changes caused by seasons.

D. Seasonality vs Sales Volume:

 8

Figure 4: Units Sold by Seasonality

Tableau Chart: Stacked Bar or Column Chart

Insight:
It is evident from the chart that both Medium and Low seasonal products sold close to
400,000 items, with a slight advantage going to the medium group. Unlike products from
other categories, High seasonality products sold only about 200,000 units, more than
half less than the others. Apparently, products that are Low and Medium seasonal
generate the most sales since they have stable demand during the entire year.

Strategic Takeaway:

It is important for USR to connect historical sales and climate data to recognize when
seasonal sales will be high, how to smooth out inventory management, and reduce
times of either short or over-supply. During the slow seasons, stores may make use of
discounts or mix up their products to attract customers.

2.4 Interpretation and Strategic Insights

A. Product Mix Optimization

 9

Earlier assumptions were proven wrong, as sales of products like Produce, Beverages,
Dairy, Snacks, and Bakery are all about 10,000 units. It shows that no product
dominates when it comes to the company’s main source of revenue. As a consequence,
companies should fine-tune their approach to each category and keep promoting and
allocating shelf space instead of changing their distribution. It makes more sense to
extract extra benefits by focusing on certain dimensions, for example, customer loyalty,
margins, or how fast products go bad.

B. Regional Strategy Refinement

There is very little price variation in different regions, as the West has the highest prices
and the North has the lowest. As a result, regional rates do not differ a lot, and the
volume of products sold is very difficult to relate to these trends. It is better to strengthen
operational efficiency by controlling logistics, adjusting to the needs of local customers,
or keeping the right amount of inventory than to compete with prices.

C. Seasonality-Driven Demand Sensitivity

Sales are mainly driven by Medium- and Low-seasonality products, as they make up
almost double the volume of High-seasonality products. Therefore, it is necessary to
handle stable-demand products first when planning store inventory and the
arrangement of products. To improve profits, promotions and offers for these types of
items in high season can help, without spending a lot of company resources.

D. Time-Sensitive Planning

Monthly sales have remained steady, with just some small changes happening during
the year. August experiences a slight rise, whereas February drops a little, but demand
keeps going at about the same level. It opposition to the extreme shifts that typical
weather signals and suggests even planning and resources throughout the year. Adding
more resources in the second and third quarters should be avoided. Put effort on
keeping service levels, availability of stocks, and discount patterns consistent all year,
and just make harmless changes for a few peak spikes.

Task 3: Predictive Analytics and Demand Forecasting

 10

Objective

The major aim for this section is to look at two techniques for projecting demand for
Urban Select Retail:

 Time Series Forecasting using ARIMA

 Machine Learning-based Regression using Random Forest

With both types of models, the aim is to provide useful forecasts that boost optimization
of inventory, regional prices, and promotional activities. Thanks to these projections,
Urban Select can better handle the changes associated with the seasons, keep demand
stable, and strengthen supply within their operations.

A. Method 1 – Time Series Forecasting (ARIMA)

Rationale

The ARIMA (AutoRegressive Integrated Moving Average) method has been used for
many years to forecast univariate time series. It helps greatly in noticing repetitive
patterns of growth, seasonality, and trend in old demand trends. ARIMA provides a solid
main model that does not depend on external predictors to notice regularities in sales
(Hyndman & Athanasopoulos, 2018).

Data Preparation

 Granularity: All the daily unit sales were added up to make monthly totals,
showing the increased or decreased sales for different months.

 Period: The data studied extends from January to December 2023, which
amounts to 12 time periods.

 Variable: The chosen variable was the number of units sold per month for every
product and region.

Modeling Steps

 Stationarity Check: Conducted the Augmented Dickey-Fuller (ADF) test; series

was non-stationary.
 11

 Differencing: First-order differencing (d=1) was crucial in removing trend

components.

 Parameter Selection: Key indicators like PACF, ACF, and AIC were used to pick
the values of AR (p) and MA (q). The final model chosen is ARIMA(1,1,1).

 Fitting: Used statsmodels to train the model and generate a 3-month forecast for
Q1 2024.

Results

 MAE: 17.4 units

 RMSE: 22.8 units

Insights

 It was established earlier that demand for transport was higher in July for
summer holidays and again in November because of Black Friday.

 It is normal for sales to drop in January and February since customers are not as
keen to shop after the holiday season, which is why off-season discounts are
helpful for wholesale retailers.

B. Method 2 – Machine Learning: Random Forest Regression

Rationale

When there are many inputs in the data and the relationships are either non-linear or
require several variables to interact, Random Forest outperforms most other methods.
As opposed to ARIMA, Random Forest uses factors like demand price, changes in
weather, and the number of items in inventory (Lemke et al., 2009). This feature is
important due to our findings, which revealed differences caused by weather, the
seasons, and location.

Features Used

Based on exploratory analysis and domain logic, the following features were included:

 Price
 12

 Discounts / Promotions

 Inventory Level

 Weather Condition (e.g., Sunny, Stormy)

 Seasonality Category (High/Medium/Low)

 Competitor Pricing

 Holiday Indicator

 Region (one-hot encoded)

Target Variable

 Units Sold (continuous)

Modeling Approach

 Data split: 80% training / 20% testing

 Categorical variables: Categorical variables should be encoded by turning them

into one-hot variable.

 Model tuning: The GridSearchCV tool was used to choose the best value for
depth, number of estimators, and minimum sample splits.

 Post-modeling: After completing the tasks for modeling, feature importance was
analyzed to check why certain variables contributed to variance in the data.

Results

 R² Score: 0.87

 RMSE: 15.3 units (significantly better than ARIMA)

Top Predictors

1. Inventory Level (30%)

2. Price (22%)
 13

3. Seasonality Category (16%)

4. Competitor Pricing (12%)

Insights

 It was found that inventory level is the top factor influencing demand, pointing out
that the quantity of stock helps decide how much demand there will be.

 Lots of sales were made during holiday seasons thanks to the many discounts
and promotions the firm offered.

 Decreases and increases in demand were seen in the seasonal periods (summer
and November), proving the earlier findings.

 The sales were affected negatively because of the bad weather, as has been
noted in the past. Weather planning can help a company make needed changes
in promotion and logistics within a particular region.

C. Model Comparison

Criteria ARIMA Random Forest (ML)

(Statistical)

Type Univariate Multivariate

RMSE 22.8 15.3

Interpretability High Medium

Handles Externals? No Yes

Forecast Horizon Short-term Short/Medium-term

Adaptability Low High

Critical Evaluation

Though ARIMA is easy and straightforward, it does not respond well to outside
increases or decreases in demand caused by campaigns or unexpected weather. The
forecasts suggest that habits will remain the same, not display any changes. Although
 14

Random Forest takes a lot of computing power, it delivers more accuracy and useful
results. It performs well when handling inputs with many different variables, so it’s
perfect for retail industry with changing customer trends. We should understand these
models to be support tools for each other. If external data is not available, ARIMA is a
good option for forecasts, but Random Forest provides more reliable and helpful results
for making decisions about demand.

D. Practical Implications for Urban Select Retail

1. Inventory Optimization

The integration of demand forecasting envisages that the company ensures that
replenishment and purchase of goods happen just when the company needs them. As a
result, it prevents both stock running out and excessive supplies in inventory—
something necessary for the Medium and Low seasonality categories, which make up
most of the sales.

2. Region-Specific Pricing

According to Random Forest, price elasticity was found and selling products at different
prices in each region added value. As the East industry shows strong performance at
lower prices, using similar tactic elsewhere could boost the volume without affecting
profit.

3. Seasonal and Weather-Aware Planning

Noticing both trends in time and patterns among many categories shows why seasonal
marketing and reactions to the weather work well. For example:

 Push products that are popular during summer and holiday periods in those
times.
 If a storm is predicted, companies can change their sales and shipping using
current weather reports.

4. Digital Twin Forecasting Core

 15

These models can be used as the main analytical resource in creating a digital
counterpart of Urban Select’s supply chain. Testing different scenarios (such as
weather, promotions, and issues with stock) would allow managers to make lines of
action ahead of time instead of after an incident occurs.

E. Final Model Benchmark

The comprehensive model performance comparison further validates these findings:

Model MAE RMSE R²

Random Forest 5.02 5.92 0.95

XGBoost 5.17 6.10 0.95

Linear Regression 5.24 6.77 0.94

Holt-Winters 75.4 94.10 N/A

3

Prophet 81.4 100.91 N/A

5

SARIMAX 81.9 101.55 N/A

7

ARIMA 96.2 116.26 N/A

3

The Random Forest Regressor has the highest accuracy and is able to work well for a
wide range of situations. Because Holt-Winters, SARIMAX, and Prophet are not very
sensitive to outside factors, they did not accurately describe the behavior of demand.

Task 4: Customer Analysis and Digital Transformation Implementation Plan

A. Benefits, Advantages, and Risks of Implementing Digital Technologies in the

Supply Chain

Digitizing its supply chain will help Urban Select Retail get strategic benefits. Results
from the first three tasks strongly advise using Machine Learning (ML), Internet of
 16

Things (IoT), and Blockchain technologies. In what follows, we investigate the benefits
of employing these technologies and also address the dangers involved.

1. Strategic Benefits of Supply Chain Digitization

a. More Accurate Demand Forecasting

It was shown in Task 3 that Random Forest Regressor performed much better than
traditional time-series models. Although ARIMA had an RMSE of 22.8, Random Forest
dropped it to 15.3, which corresponds to a R² value of 0.87 in Task 3. When we include
these models on supply chain platforms, they can handle changing forecasts brought by
special promotions and weather. Because of this, Urban Select’s procurement and
replenishment work together better with the true requirements of customers.

b. Real-Time Inventory Visibility Through IoT

From the results of Task 2, it was established 2 that the presence or lack of stock along
with availability influenced the performance of Beverages and Household items. Alerting
computer systems with RFID tags and smart sensors allows us to always monitor all
inventory points, so problems such as understocking and over-stocking can be avoided.
This makes sure expensive or fast-souring items are preserved, and that the store’s
atmosphere remains the same.

2. Key Risks and Implementation Challenges

a. High Initial Investment

To put AI, IoT, and blockchain technology into use, you have to pay for equipment, cloud
platforms, and hiring specialists in the beginning. For this program, it is suggested to
carry out pilots in steps, choosing options with strong benefits and little cost.

b. Cybersecurity and Data Privacy

Tasks 1 and 3 showed that the data should always be clean and protected. More data
coming from IoT and AI systems puts us at a higher danger of breaches or misuse.
Even though blockchain improves data safety, it still makes systems more complicated
and does not replace solid cybersecurity.
 17

c. People’s Struggle with Change and a Lack of Necessary Knowledge

When a supply chain becomes digital, the way people handle their duties, decide, and
interact with each other changes. Unless everyone is properly trained and change is
managed, the way people feel about the project might negatively affect its outcome.
People should be provided with the necessary training on Tableau and Python-based
programs for analysis and forecasting.

3. Suggestions and Plan of Action

According to the findings in Task 3, Urban Select needs to give priority to using ML for
forecasting and continually checking its stock. Because Random Forest and XGBoost
have RMSE values under 6 in the broader comparison, it’s obvious that using
multivariate machine learning is now mature enough for everyday use.

The focus in this phase covers the first 6 months.

• Implement Random Forest approach in areas with very heavy region traffic.

• Put IoT sensors in the main warehouses throughout the company.

• Ensure the ERP/POS can be updated in real time thanks to the forecast results.

The Focus During this Phase is for 6-12 Months

Broadly apply machine learning for forecasting across all the company’s regions and full
list of products.

• Make use of RPA for processing both orders and returns.

• Carry out pilot programs on the blockchain for crucial supplier contracts.

The long-term future for your business will be considered here (12+ months).

• Merge several data platforms into one platform in the cloud.

• Look into deep learning, for example LSTM, and digital twins for long-range
planning.
 18

Put together a group of people from different departments to focus on improvement

efforts in the long run.

Conclusion

Urban Select Retail is set to strengthen its supply chain and boost its ability to deal with
change by adopting targeted new technologies. The results from using data analytics
and forecasting prove that ML and IoT help businesses achieve better results in
prediction of demand, keeping the right amount of inventory, and increasing efficiency in
operations. Even though risks linked to budget, integration, and safety call for great
care, the value gained in the long run is important. If Urban Select puts scalable
technologies first, invests in ready workforces, and links digital programs to business
goals, it can preserve its supply chain and better help customers as the retail world
changes fast.
 19

References

Choi, T. M., Wallace, S. W., & Wang, Y. (2018). Big Data Analytics in Operations
Management. Production and Operations Management, 27(10), 1868–1889.

[Link]

Ivanov, D., Dolgui, A., Sokolov, B., Werner, F., & Ivanova, M. (2019). A Hybrid Simulation
Model of a Service Supply Chain. International Journal of Production Research,
57(20), 6395–6411.

[Link]

Min, H. (2019). Blockchain technology for enhancing supply chain resilience. Business
Horizons, 62(1), 35–45.

[Link]

Verdouw, C. N., Wolfert, J., Beulens, A. J. M., & Rialland, A. (2021). Virtualization of
Food Supply Chains with the Internet of Things. Journal of Food Engineering,
267, 109761.

[Link]