“Because running out of stock is bad… but overstocking is how you end up selling 2022 calendars in 2025.” 😬

🏢 The Business Setup¶

Meet CoffeeCo, a small but ambitious coffee retailer ☕. They sell premium beans, mugs, and caffeine-themed tote bags.

Lately, the CEO is worried:

“We either sell out too soon or end up with enough unsold mugs to build a fortress.” 🏰

Your mission, should you accept it: 📊 Use time-series forecasting to plan next quarter’s inventory.

🧰 Step 1. Gather the Data¶

The company tracked monthly sales of coffee beans for the past 3 years. Here’s what it looks like:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

np.random.seed(42)
months = pd.date_range(start='2021-01-01', periods=36, freq='M')
sales = 200 + 15*np.arange(36) + 50*np.sin(np.arange(36)/3) + np.random.normal(0, 30, 36)
df = pd.DataFrame({'ds': months, 'y': sales})

plt.plot(df['ds'], df['y'])
plt.title('☕ Monthly Coffee Bean Sales')
plt.xlabel('Month'); plt.ylabel('Units Sold')
plt.show()

☕ The pattern? A growing trend, seasonal peaks, and occasional chaos (a.k.a. TikTok made them viral again).

🧙‍♂️ Step 2. Train a Forecasting Model (Prophet to the Rescue)¶

from prophet import Prophet

model = Prophet(yearly_seasonality=True)
model.fit(df)

future = model.make_future_dataframe(periods=6, freq='M')
forecast = model.predict(future)

Just like that, Prophet gives you 6 months of data-driven hope.

📈 Step 3. Visualize the Forecast¶

from prophet.plot import plot_plotly, plot_components_plotly

plot_plotly(model, forecast)
plot_components_plotly(model, forecast)

🎨 You’ll see:

• A steady trend 📈

• Predictable seasonality (every winter = caffeine spike) ❄️

• Random holidays or marketing chaos (hi, “Pumpkin Spice Month”) 🎃

🧮 Step 4. Convert Forecast → Inventory Plan¶

The CEO asks:

“How much coffee should we stock for the next 6 months?”

You smile confidently (and open Excel).

future_sales = forecast[['ds', 'yhat', 'yhat_lower', 'yhat_upper']].tail(6)
future_sales['inventory_plan'] = future_sales['yhat'] * 1.2  # 20% buffer
future_sales

🧾 Interpretation:

• yhat: Expected sales

• yhat_lower / yhat_upper: Uncertainty range

• inventory_plan: Add buffer so you don’t run out during sudden caffeine frenzies ☕💥

📊 Step 5. Evaluate Forecast Performance¶

Before you show your boss the results, test how good the model actually was:

from sklearn.metrics import mean_absolute_percentage_error

train = df.iloc[:-6]
test = df.iloc[-6:]

model = Prophet()
model.fit(train)
future = model.make_future_dataframe(periods=6, freq='M')
forecast = model.predict(future)

mape = mean_absolute_percentage_error(test['y'], forecast['yhat'][-6:])
print(f"MAPE: {mape*100:.2f}%")

If your MAPE < 10%, celebrate with a triple espresso. If > 20%, tell your manager it’s a “data alignment issue.” 😅

💼 Step 6. Presenting to Management¶

Here’s how you sound smart in the meeting:

“Based on historical trends and seasonality, our forecast predicts a 15% growth in Q2. With a 20% safety stock buffer, we recommend ordering ~3,200 units for optimal coverage.”

Translation for business folks: ✅ No empty shelves ✅ No storage overflow ✅ You get credit for “AI-driven planning”

🧠 Step 7. Business Takeaways¶

🎯 Final Result¶

Prophet says:

“Next quarter will be strong — order confidently, but not recklessly.”

The CEO says:

“Excellent. Also, make me a dashboard.”

You say:

“Coming right up… after one more coffee.” ☕💻

🧾 TL;DR¶

“Inventory planning is the fine art of predicting tomorrow’s chaos using yesterday’s confusion — preferably before finance notices.” 💼😅