Business Analytics Explained: A Beginner's Guide

Every business makes decisions: what to stock, whom to hire, where to spend the marketing budget, which customers to call. Business analytics is the practice of using data to make those decisions better. It covers everything from a monthly sales report to a forecasting model, and it relies on analytical modelling and numerical analysis of many kinds (Wikipedia: business analytics). What unites it is purpose: the analysis exists to inform a decision, not to decorate a slide.

The four types of analytics, by the question they answer

Analytics is usually divided by the kind of question it answers. Microsoft's introductory training, for example, organises the types of analytics around the business questions each answers (Microsoft Learn: the analytics process):

Descriptive analytics draws insight from historical data through reporting and scorecards; predictive analytics uses statistical and machine learning models; prescriptive analytics recommends decisions using optimisation and simulation (Wikipedia: business analytics). Most of an analyst's work is descriptive and diagnostic, and that is where most business value begins. Descriptive vs diagnostic vs predictive analytics compares them in detail.

A worked example: from what happened to why

Take a real UK online gift retailer's 2011 transactions (UCI: Online Retail). The question a manager might ask: sales in November were far higher than in August, so what drove it? First, prepare monthly totals by market, excluding cancellations, postage and fees:

import io
import urllib.request
import zipfile
import numpy as np
import pandas as pd

url = "https://archive.ics.uci.edu/static/public/352/online+retail.zip"
with zipfile.ZipFile(io.BytesIO(urllib.request.urlopen(url).read())) as archive:
    raw = pd.read_excel(archive.open("Online Retail.xlsx"), dtype={"InvoiceNo": str, "StockCode": str})
services = {"POST", "DOT", "M", "m", "C2", "D", "S", "B", "BANK CHARGES", "AMAZONFEE", "CRUK"}
sales = raw[~raw["InvoiceNo"].str.startswith("C") & ~raw["StockCode"].isin(services)
            & ~raw["StockCode"].str.startswith("gift_") & (raw["Quantity"] > 0) & (raw["UnitPrice"] > 0)]
sales = sales.assign(month=sales["InvoiceDate"].dt.to_period("M").astype(str),
                     market=sales["Country"].where(sales["Country"] == "United Kingdom", "International"),
                     revenue=sales["Quantity"] * sales["UnitPrice"])
aug, nov = (sales[sales["month"] == m] for m in ("2011-08", "2011-11"))
for name, d in (("August", aug), ("November", nov)):
    print(f"{name}: revenue {d['revenue'].sum():,.0f} | orders {d['InvoiceNo'].nunique():,} | "
          f"average order {d['revenue'].sum() / d['InvoiceNo'].nunique():,.0f}")

August: revenue 725,514 | orders 1,339 | average order 542
November: revenue 1,457,746 | orders 2,751 | average order 530

That is descriptive analytics: revenue doubled. The diagnostic question is why. Revenue equals orders multiplied by average order value, so the growth can be split between the two, and the increase can also be split by market:

growth = nov["revenue"].sum() / aug["revenue"].sum()
from_orders = np.log(nov["InvoiceNo"].nunique() / aug["InvoiceNo"].nunique()) / np.log(growth)
print(f"revenue growth: {growth - 1:+.1%} | share explained by more orders: {from_orders:.0%}, "
      f"by order value: {1 - from_orders:.0%}")
by_market = pd.DataFrame({"August": aug.groupby("market")["revenue"].sum(), "November": nov.groupby("market")["revenue"].sum()})
by_market["share of the increase"] = (by_market["November"] - by_market["August"]) / (by_market["November"] - by_market["August"]).sum()
print(by_market.round({"August": 0, "November": 0, "share of the increase": 3}).to_string())

revenue growth: +100.9% | share explained by more orders: 103%, by order value: -3%
                  August   November  share of the increase
market                                                    
International   145555.0   171471.0                  0.035
United Kingdom  579959.0  1286275.0                  0.965

The answer is specific: all of the growth came from more orders, while the average order actually shrank slightly, and 96.5% of the extra revenue came from UK customers. For a gift retailer, the obvious hypothesis is the pre-Christmas season, which a comparison with the previous year would confirm. That is the difference between reporting "sales are up" and explaining why, and it changes the decisions that follow: stock and staffing for UK order volume in autumn, not efforts to raise basket size.

The next steps would be predictive, forecasting December and next autumn's volumes from the seasonal pattern, and prescriptive, deciding how much stock to hold given the forecast and its uncertainty. Correlation vs causation explains why a diagnostic finding like the seasonal one should be confirmed before it drives a costly decision.

The analytics process

1. Frame the question with the person who will act on the answer: what decision, by when, and what would change it?
2. Find and understand the data: sources, definitions, gaps.
3. Clean and prepare it, recording every exclusion.
4. Analyse: describe, break down, compare, and model where needed.
5. Communicate: the answer first, then the evidence, in the reader's language.
6. Act and measure: check whether the decision had the expected effect.

Most failed analyses fail at step 1, answering a question nobody asked, or at step 5, burying the answer.

Roles and tools

• Data or business analysts answer business questions with SQL, spreadsheets and BI tools such as Power BI.
• BI developers build the data models, reports and dashboards many people use.
• Data engineers build the pipelines and warehouses that make data available and reliable.
• Data scientists take on predictive and prescriptive work with statistical and machine learning models.

The tools form a stack from data sources through storage and transformation to reporting; the modern data analytics stack explains how the pieces fit.

Where business analytics is used

Almost every function has its own analytics questions:

• Sales: which products, regions and customers drive revenue, and which are slipping?
• Marketing: which campaigns bring customers who stay and spend, not just clicks?
• Operations: how much stock to hold, how many staff to schedule, where delays arise?
• Finance: where costs are rising faster than revenue, and what the next quarter is likely to bring?
• People: why employees leave, and which teams are stretched?

The questions differ, but the method is the same: a precise question, trustworthy data, an honest comparison and a clear recommendation.

Common beginner mistakes

• Starting with the data instead of the decision, and producing charts nobody needed.
• Reporting totals without breakdowns, which hides where change actually happens, as the market split above showed.
• Treating a pattern as a cause without checking alternatives such as seasonality.
• Ignoring data quality: exclusions, missing values and definitions change answers.

Skills a beginner needs first

• Spreadsheets and SQL, to get and shape data.
• One BI tool, to share results.
• Basic statistics: averages and medians, variation, sampling and uncertainty. Statistical concepts every analyst must know covers them.
• Business understanding: how the organisation makes money and what its managers worry about.
• Clear writing, because an insight nobody understands changes nothing.

Excel, SQL, Python, data visualisation, Power BI and business analytics concepts such as KPI and sales analysis make up the Data Analytics course.

Sources

• Business analytics (Wikipedia) — Wikipedia
• Explore the analytics process that turns data into insights (Microsoft Learn) — Microsoft Learn
• Online Retail (UCI Machine Learning Repository) — UCI Machine Learning Repository