Build This Project · Portfolio Guide
Supply Chain Analytics Dashboard — Complete Project Guide
The brief
Imagine a mid-sized consumer goods distributor shipping from three warehouses, in Bhiwandi, Gurugram and Bengaluru, to retailers across India. The operations head has a weekly review and keeps asking the same things: Which warehouse is missing delivery promises? Are we shipping complete orders? Which products are sitting in stock too long? Your dashboard has to answer those in under a minute.
That framing matters for your portfolio. Reviewers are less impressed by many charts than by a dashboard that clearly solves one stakeholder's problem.
Step 1: Get the data
You need at least two tables:
- Orders / shipments: order ID, warehouse, customer region, product, order date, promised date, delivered date, quantity ordered, quantity delivered, shipping cost.
- Inventory snapshots: product, warehouse, date, stock on hand, unit cost.
Public options include supply chain and retail datasets on Kaggle (search for "supply chain" or "logistics"), the Sample Superstore dataset that ships with Tableau, or a dataset you generate yourself in Python with realistic delays. Whatever you choose, say so in the README; reviewers want to know whether patterns are real or simulated.
Step 2: Define the KPIs before you draw anything
| KPI | Formula | Question it answers |
|---|---|---|
| OTIF (on time, in full) | Orders delivered by the promised date and with full quantity ÷ all orders | Are customers getting what we promised? |
| Average lead time | Average of (delivered date − order date) | How long do customers wait? |
| Fill rate | Units delivered ÷ units ordered | How much demand do we actually meet? |
| Inventory turnover | Cost of goods sold ÷ average inventory value | How fast does stock move? |
| Days of inventory | 365 ÷ inventory turnover | How many days of stock are we holding? |
| Stockout rate | Days (or SKUs) with zero stock ÷ total days (or SKUs) | How often can we not sell? |
| Shipping cost per order | Total freight cost ÷ orders | Is fast delivery costing too much? |
OTIF is the headline metric for many distributors and retailers because it combines two failure types. An order that arrives on time but short is still a failure for the customer.
Step 3: Check the calculations on a small sample
Before building visuals, calculate the KPIs by hand or in a script on a few rows, so you know what the dashboard should show. This plain Python code needs no libraries:
from datetime import date
from collections import defaultdict
# order_id, warehouse, ordered, promised, delivered, qty_ordered, qty_delivered
orders = [
(101, "Bhiwandi", date(2026, 3, 2), date(2026, 3, 5), date(2026, 3, 5), 40, 40),
(102, "Bhiwandi", date(2026, 3, 2), date(2026, 3, 5), date(2026, 3, 7), 25, 25),
(103, "Bhiwandi", date(2026, 3, 3), date(2026, 3, 6), date(2026, 3, 6), 30, 27),
(104, "Gurugram", date(2026, 3, 2), date(2026, 3, 4), date(2026, 3, 4), 50, 50),
(105, "Gurugram", date(2026, 3, 3), date(2026, 3, 5), date(2026, 3, 5), 20, 20),
(106, "Gurugram", date(2026, 3, 4), date(2026, 3, 6), date(2026, 3, 9), 35, 35),
(107, "Bengaluru", date(2026, 3, 2), date(2026, 3, 6), date(2026, 3, 5), 15, 15),
(108, "Bengaluru", date(2026, 3, 3), date(2026, 3, 7), date(2026, 3, 7), 45, 45),
]
stats = defaultdict(lambda: {"n": 0, "otif": 0, "lead": 0, "ord": 0, "dlv": 0})
for _, wh, ordered, promised, delivered, q_ord, q_dlv in orders:
s = stats[wh]
s["n"] += 1
s["otif"] += delivered <= promised and q_dlv == q_ord # on time AND in full
s["lead"] += (delivered - ordered).days
s["ord"] += q_ord
s["dlv"] += q_dlv
print(f"{'warehouse':<11}{'orders':>7}{'OTIF %':>8}{'lead days':>10}{'fill %':>8}")
for wh, s in stats.items():
print(f"{wh:<11}{s['n']:>7}{100 * s['otif'] / s['n']:>8.1f}"
f"{s['lead'] / s['n']:>10.1f}{100 * s['dlv'] / s['ord']:>8.1f}")
Output:
warehouse orders OTIF % lead days fill %
Bhiwandi 3 33.3 3.7 96.8
Gurugram 3 66.7 3.0 100.0
Bengaluru 2 100.0 3.5 100.0
Notice what the sample already teaches. Bhiwandi's fill rate is 96.8%, which sounds fine, but its OTIF is only 33.3% because one order was late (102) and another was short (103). A dashboard that showed only fill rate would hide the problem. Bengaluru's order 107 arrived a day early and still counts as on time; decide and document whether early deliveries count, because some retailers penalise them too.
Inventory turnover works the same way. For one product line:
cogs = 1_800_000 # cost of goods sold for the year, in Rs
opening_stock = 320_000 # inventory value at start of year
closing_stock = 280_000 # inventory value at end of year
avg_inventory = (opening_stock + closing_stock) / 2
turnover = cogs / avg_inventory
days_of_inventory = 365 / turnover
print(f"Inventory turnover: {turnover:.1f} times a year")
print(f"Days of inventory: {days_of_inventory:.0f} days")
Output:
Inventory turnover: 6.0 times a year
Days of inventory: 61 days
About two months of stock may be normal for packaged goods and far too much for fresh food. Turnover only means something when compared across products, warehouses or months.
Step 4: Build it in your BI tool
Power BI, Tableau, Looker Studio or even Excel all work. In Power BI, for example, OTIF becomes a measure such as DIVIDE(COUNTROWS(FILTER(Orders, Orders[Delivered] <= Orders[Promised] && Orders[QtyDelivered] = Orders[QtyOrdered])), COUNTROWS(Orders)), with a date table so you can slice by week. Check that your tool's numbers match the script above on the same sample rows. If you are new to dashboard tools, dashboards in Google Sheets is a gentle start.
Step 5: Lay out the page for the weekly review
- Top row (cards): OTIF %, average lead time, fill rate, days of inventory, each with change versus last week.
- Middle left: OTIF by warehouse as a bar chart, sorted worst first.
- Middle right: weekly OTIF trend line, split into "late" and "short" failures.
- Bottom left: top 10 products by days of inventory (slow movers).
- Bottom right: stockouts by product and warehouse as a heatmap.
- Slicers: date range, warehouse, region, product category.
Keep colours meaningful: one colour for "on target", one for "needs attention", and grey for everything else. General principles are in what a BI dashboard is.
Step 6: Write the insight, not just the chart
End with three to five findings in plain language, for example: "Most OTIF failures at the western warehouse are late deliveries on Mondays, which suggests a weekend backlog," or "Three slow-moving SKUs hold a large share of inventory value." With simulated data, say the findings are from simulated data. Then suggest one action per finding.
Presenting it in your portfolio
- A one-paragraph business problem, the data source and the KPI definitions.
- Screenshots plus a link to the live dashboard (Power BI Publish to web, Tableau Public).
- Your validation step: how you checked the numbers.
- Limits: what the data cannot tell you, such as the reason a truck was late.
Pair this project with one from a different domain so your portfolio shows range: the sales analytics dashboard or the healthcare analytics dashboard. For how to group projects into a portfolio, see building a job-ready data analytics portfolio.