Interactive Business Analytics Dashboard

I wanted this project to feel like a tool a store manager could actually open during a review meeting. The opening screen summarizes the uploaded workbook, identifies the detected variable types, and gives immediate confidence that the data was read correctly. That mattered to me because the dashboard was not meant to impress with code first. It was meant to reduce the distance between a messy workbook and a useful decision.

The workbook had a title row above the actual headers, which is common in business spreadsheets. I wrote the app to detect the likely header row instead of assuming that the first row was always clean. Users can also choose the worksheet and adjust skipped rows manually. This design choice made the dashboard more realistic: it expects spreadsheets to be human-made, not perfectly database-shaped.

Before I let users interpret charts, I made the dashboard show what might weaken the analysis. Wait_Time_Min had 54 missing values because wait time does not apply to every channel. Complaint_Theme was missing in most rows because most transactions did not have complaints. Loyalty_Tier had only three missing values. Treating these cases differently was important because missingness is not always the same kind of problem.

The numeric summary tab gives the basic statistics a manager expects: count, missing values, mean, median, standard deviation, quartiles, minimum, and maximum. I included both mean and median because customer metrics can be skewed, especially order value and NPS. I validated these summaries against independent calculations so the table could be trusted as an analytical baseline.

The categorical summary tab helped me move from columns to business questions. Region, channel, product category, customer segment, loyalty tier, complaint flag, return flag, and sentiment label are not just labels. They are ways managers organize action. By showing frequencies and percentages, the dashboard makes it easier to see where customers, complaints, or transactions are concentrated.

The grouped summary tab is where the dashboard becomes diagnostic. A user can compare a numeric metric such as order value, satisfaction, discount rate, wait time, or NPS across a categorical grouping such as region, channel, loyalty tier, or product category. This is how I found the practical findings: Phone had the weakest NPS, and Electronics and Beauty carried the highest complaint rates.

The plot tab lets users choose variables and generate charts directly. Numeric variables become histograms, categorical variables become bar charts, and optional grouping makes the chart more diagnostic. I included this because a dashboard should not only answer the questions I already know. It should help managers ask the next question while they are looking at the data.

I did not want the dashboard to become a black box. The raw-data tab keeps the filtered rows visible through a searchable table. This matters when a manager wants to sanity-check an outlier, inspect a complaint, or confirm that a filter is doing what they expect. The ability to move from chart to row-level evidence is a small feature, but it builds trust.

This case is personal to me because it shows how I like to combine analytics and product thinking. I did not only calculate summary statistics. I asked who would use the dashboard, what they would trust, where they might make a wrong assumption, and how the interface could guide them toward better questions. The result is an academic case study, but the mindset is practical: make the analysis usable, transparent, and honest about its limits.