Project: an exploratory analysis report

Questions 1 and 2: growth and product lines

import pandas as pd

def dataset(name):
    return pd.read_csv(f"https://api.gradient.test/v1/datasets/{name}.csv")

orders = dataset("orders").drop_duplicates()
orders = orders[orders["sku"] != "TEST-000"].copy()
full = orders.merge(dataset("catalog")[["sku", "cost", "line_id"]], on="sku").merge(dataset("product-lines").rename(columns={"name": "line"})[["line_id", "line"]], on="line_id")
full["revenue"] = full["qty"] * full["unit_price"]
full["profit"] = full["revenue"] - full["qty"] * full["cost"]
full["year"] = full["date"].str[:4].astype(int)

by_year = full.groupby("year")["revenue"].sum()
print("overall growth:", round((by_year[2025] / by_year[2024] - 1) * 100, 1), "%")
stores = full.pivot_table(index="store", columns="year", values="revenue", aggfunc="sum")
stores["growth_pct"] = ((stores[2025] / stores[2024] - 1) * 100).round(1)
print(stores.sort_values("growth_pct", ascending=False))

lines = full.groupby("line").agg(revenue=("revenue", "sum"), profit=("profit", "sum"))
lines["margin_pct"] = (lines["profit"] / lines["revenue"] * 100).round(1)
print(lines.sort_values("profit", ascending=False).round(1))

Revenue grew 8.4%, but not everywhere. The web shop and Portland grew; Denver slipped a little; Seattle fell by 16%. Before you report that as a crisis, remember that Seattle is a small store, with about 200 orders a year. A small total swings more from year to year by chance, just as small groups did in the grouping lesson. Report it as something to watch, and check it against more data, such as visits or the first months of next year.

Ridgeline (tents) earns by far the most profit, because tents are expensive. Ember (camp kitchen) has the best margin. Which one "matters most" depends on whether leadership wants more profit dollars or more profit per dollar sold, so report both.

Question 3: a test, and a null result

Members get discounts and points. Do they spend more per order? Collapse lines to orders first, as the grouping lesson showed, then compare the two groups of order totals with Welch's t-test:

import pandas as pd
from scipy import stats

orders = pd.read_csv("https://api.gradient.test/v1/datasets/orders.csv").drop_duplicates()
orders = orders[orders["sku"] != "TEST-000"].copy()
orders["revenue"] = orders["qty"] * orders["unit_price"]

per_order = orders.groupby("order_id").agg(member=("member", "first"), total=("revenue", "sum"))
members = per_order[per_order["member"] == 1]["total"]
others = per_order[per_order["member"] == 0]["total"]
result = stats.ttest_ind(members, others, equal_var=False)
low, high = result.confidence_interval(0.95)
print(len(members), "member orders,", len(others), "others")
print("mean order:", round(members.mean(), 2), "vs", round(others.mean(), 2))
print("gap:", round(members.mean() - others.mean(), 2), " p =", round(result.pvalue, 3))
print("95% interval for the gap:", round(low, 2), "to", round(high, 2))

Members' orders average about $8 less, but p is 0.27, and the interval runs from about −$23 to +$6. So the data shows no clear difference either way. That's a real finding. Report it honestly as "no evidence that members spend more per order". It is not "members spend less", and it is not "proven equal". A null result is still worth reporting: it stops the team from assuming the loyalty program raises order size.

Also note what this comparison can't show. Members chose to join, so they may differ from other customers in many ways. Even a clear gap wouldn't prove the program caused it.

import pandas as pd
lines = pd.DataFrame({"order_id": [1, 1, 1, 2], "member": [1, 1, 1, 0], "revenue": [10, 10, 10, 50]})
print(lines.groupby("member")["revenue"].mean().to_dict(), lines.groupby("order_id")["revenue"].sum().to_dict())

Questions 4 and 5: seasons and stores

Seasons need month-of-year, so every January lands in the same row, whichever year it's from:

full["month_of_year"] = full["date"].str[5:7].astype(int)
season = full.pivot_table(index="month_of_year", columns="category", values="qty", aggfunc="sum")
print(season.idxmax())

Apparel peaks in winter, around January, while tents, packs and footwear peak from June to August. That's useful for stock planning, and a good reason to look at categories separately: added together, the opposite seasons partly cancel out and hide both patterns.

Per-visit revenue repeats the joins lesson: total revenue per store, joined to the stores table, divided by two years of visits. Portland earns most per visit and the web shop least. That's the expected pattern for browsing online, and a useful comparison to the growth numbers. The web shop is growing fastest, but each visit is worth the least.

Writing it up

A report is for someone who won't read your code. Put the answers first:

• Lead with findings, one per question, each with its number: "Revenue grew 8.4%."
• Give sizes and uncertainty, not just directions: intervals for tests, and small samples called out.
• One chart per finding, titled with the finding itself ("Apparel peaks in winter"), not the data ("Units by month").
• Caveats in plain words: what was removed in cleaning, what the data can't show (cause and effect), and what you'd check next.
• Keep the code reproducible, so next year's review is a rerun, not a rewrite.

Leave out the dead ends. You may have made thirty charts to find five findings, but the reader needs the five. Keep the rest in your notebook, in case someone asks.

import matplotlib.pyplot as plt
import pandas as pd

def dataset(name):
    return pd.read_csv(f"https://api.gradient.test/v1/datasets/{name}.csv")

orders = dataset("orders").drop_duplicates()
orders = orders[orders["sku"] != "TEST-000"].copy()
full = orders.merge(dataset("catalog")[["sku", "cost", "line_id"]], on="sku").merge(dataset("product-lines").rename(columns={"name": "line"})[["line_id", "line"]], on="line_id")
full["revenue"] = full["qty"] * full["unit_price"]
full["profit"] = full["revenue"] - full["qty"] * full["cost"]
full["month"] = full["date"].str[:7]

monthly = full.groupby("month")["revenue"].sum()
lines = full.groupby("line").agg(revenue=("revenue", "sum"), profit=("profit", "sum"))
margin = (lines["profit"] / lines["revenue"] * 100).sort_values()

fig, (left, right) = plt.subplots(1, 2, figsize=(12, 3.8))
left.plot(range(len(monthly)), monthly.values, marker=".")
left.set_xticks(range(0, 24, 3), monthly.index[::3], rotation=45)
left.set_ylim(bottom=0)
left.set_title("Revenue grew 8.4% in 2025, peaking in October")
left.set_ylabel("revenue ($)")
right.barh(margin.index, margin.values)
right.set_xlim(left=0)
right.set_title("Ember has the best margin; Scree the worst")
right.set_xlabel("profit margin (%)")
fig.tight_layout()
print(monthly.idxmax(), round(margin.max(), 1), round(margin.min(), 1))