Data Scientist Careers in Retail Industry Explained

# Demand forecasting — a classic retail data science problem
# (predicting sales per SKU per store per week)

import pandas as pd
import numpy as np
from sklearn.metrics import mean_absolute_percentage_error
import xgboost as xgb

# Load historical sales data
df = pd.read_parquet("s3://retail-lake/sales-history/")

# Feature engineering — this is 70% of the work
df["date"] = pd.to_datetime(df["date"])
df["week"] = df["date"].dt.isocalendar().week
df["month"] = df["date"].dt.month
df["quarter"] = df["date"].dt.quarter
df["day_of_week"] = df["date"].dt.dayofweek

# Lag features — what sold in previous weeks (highly predictive)
for lag in [1, 2, 4, 8, 52]:
    df[f"sales_lag_{lag}w"] = df.groupby(["sku", "store"])["sales"].shift(lag)

# Rolling averages — capture trend
for window in [4, 8, 13]:
    df[f"sales_rolling_{window}w"] = (
        df.groupby(["sku", "store"])["sales"]
        .transform(lambda s: s.rolling(window).mean())
    )

# External features — promotions, holidays, weather
df = df.merge(promotions_df, on=["sku", "week"], how="left")
df = df.merge(holidays_df, on="date", how="left")

# Drop rows with too many missing lag values (start of series)
df = df.dropna(subset=["sales_lag_4w"])

# Train/test split — last 8 weeks hold out
cutoff = df["date"].max() - pd.Timedelta(weeks=8)
train = df[df["date"] < cutoff]
test = df[df["date"] >= cutoff]

# Train XGBoost
features = [
    "week", "month", "quarter", "day_of_week",
    "sales_lag_1w", "sales_lag_2w", "sales_lag_4w", "sales_lag_8w",
    "sales_rolling_4w", "sales_rolling_8w",
    "promo_flag", "is_holiday", "price",
]
model = xgb.XGBRegressor(
    n_estimators=500, max_depth=6,
    learning_rate=0.05, subsample=0.8,
    early_stopping_rounds=30,
)
model.fit(
    train[features], train["sales"],
    eval_set=[(test[features], test["sales"])],
    verbose=False,
)

# Evaluate
preds = model.predict(test[features])
mape = mean_absolute_percentage_error(test["sales"], preds)
print(f"MAPE: {mape:.2%}")  # Typical retail: 15–25%

# Real skills: feature engineering, lag features, XGBoost, MAPE evaluation.
# This is what retail data science interviews actually test.

# Price elasticity — how much does demand change when price changes?
# (the foundation of dynamic pricing)

import numpy as np
import pandas as pd
import statsmodels.api as sm

# Historical sales + price data
df = pd.read_parquet("s3://retail-lake/price-experiments/")

# Log-log regression: log(units) ~ log(price)
# The slope coefficient IS the price elasticity
df["log_units"] = np.log(df["units_sold"] + 1)
df["log_price"] = np.log(df["price"])

# Add controls (promotions, seasonality, competitor price)
X = df[["log_price", "promo_flag", "is_holiday", "competitor_price"]]
X = sm.add_constant(X)

# Fit per product category (elasticity varies by category)
elasticities = {}
for category in df["category"].unique():
    subset = df[df["category"] == category]
    X_cat = subset[["log_price", "promo_flag", "is_holiday", "competitor_price"]]
    X_cat = sm.add_constant(X_cat)

    model = sm.OLS(subset["log_units"], X_cat).fit()
    elasticities[category] = model.params["log_price"]

# Interpret: elasticity of -1.5 means 10% price increase → 15% drop in units
for cat, e in elasticities.items():
    print(f"{cat}: elasticity = {e:.2f}")

# Real skills: log-log regression, elasticity interpretation, business reasoning.
# This is the core of a retail pricing data science role.

# Extensions: causal forests, uplift models, price optimisation with RL.
# But most of the time, you start with simple regression + business sense.