Business and operations
Machine learning for business analytics
In business analytics, machine learning predicts which customers will respond to a campaign, forecasts demand for stock and staffing, and sharpens the reading of A/B tests. Each model feeds a specific decision, so its value is measured by how that decision improves.
How machine learning is used in business analytics
Targeting campaigns at customers who respond
A response model scores each customer on how likely they are to act on an offer, using past purchases, tenure and earlier campaign history. Uplift models go a step further and estimate how much the offer itself changes behaviour, so budget goes to customers who would not have bought anyway. Logistic regression, gradient-boosted trees and meta-learners built on them are typical.
Real exampleCausalML (Uber)An open-source Python package from Uber for uplift modelling and causal inference that lists campaign targeting optimisation as a typical use case.
Forecasting demand for stock and staffing
A forecasting model predicts sales, orders or contacts per day or week from past values, seasonality, holidays and promotions. The forecast feeds purchase orders and rotas, so its errors are judged in units or hours, not just percentages. Seasonal regression models and gradient-boosted trees on lagged features are common starting points.
Real exampleProphet (Meta)An open-source forecasting procedure released by Facebook's Core Data Science team that fits an additive model with trend, yearly, weekly and daily seasonality and holiday effects.
Reading A/B tests more precisely
When a team tests a new page, price display or email, regression on each customer's pre-experiment behaviour removes noise that has nothing to do with the change. The same traffic then gives a narrower confidence interval, so smaller real effects become visible. This is usually linear regression, sometimes with model-based predictions as the covariate.
Real exampleMicrosoft Experimentation Platform: variance reductionMicrosoft's experimentation team explains how CUPED adjusts metrics with a regression on pre-experiment data and reports an effective traffic multiplier for each variance-reduced metric.
Can you learn machine learning for business analytics?
Yes. Analysts who already work in spreadsheets, SQL or BI tools have most of the business context; the gap is usually Python, statistics and supervised learning. Learn Python and pandas, then the statistics of testing, then classification and forecasting on customer and transaction tables.
- Python for analysisLearn enough Python to load an export, write functions and repeat an analysis without copying spreadsheet formulas.Learn it in Python
- Tables with pandasJoin customers, orders and campaigns, aggregate them per customer and per week, and build one modelling table without leaking future data.Learn it in Data Tools
- Statistics for experimentsConfidence intervals, hypothesis tests and regression are what you need to read an A/B test and to say whether a model's gain is real.Learn it in Statistics
- Classification and forecastingTrain logistic regression and gradient-boosted trees for response and churn, and time-series models for demand, each compared with the rule the business uses today.Learn it in Machine Learning
Moving into machine learning for business analytics
If you already work in the field
Business analysts already know the metrics, the data warehouse and the decisions a model would change. Add Python, a held-out evaluation habit and one or two model families. A realistic first move is to rebuild an existing targeting or forecast rule as a model and compare both on last quarter's data.
If you come from data science or ML
If you come from ML or data science, learn how the business measures success: margin, retention, cost per contact and how experiments are run. Spend time with the teams who act on the scores, because a model that does not change a decision has no value to them.
Roles that do this workBusiness analystData analystData scientist, marketing analyticsCustomer analytics managerAnalytics engineer
Tools you will use
| Tool | What you use it for |
|---|---|
| pandas | Clean, join and aggregate customer, order and campaign tables into one modelling table. |
| DuckDB | Run fast SQL over large CSV or Parquet exports on a laptop before modelling. |
| scikit-learn | Fit and cross-validate response, churn and segmentation models with a shared pipeline. |
| statsmodels | Fit interpretable regressions and test whether an observed lift is larger than chance. |
Public data to practise on
- UCI Bank Marketing: Phone campaign records from a Portuguese bank; predict whether a client subscribes to a term deposit.
- UCI Online Retail II: Invoice lines from a UK online retailer; analyse customer value, repeat purchases and product demand.
Your first project
Download the UCI Bank Marketing data, predict which clients subscribe with logistic regression, and compare it with contacting everyone in call order using ROC AUC and the subscriptions found in the top 10% of calls.
Common questions
What is the difference between business analytics and machine learning?
Business analytics is the wider work of measuring and explaining what happens in a business, often with dashboards and reports. Machine learning is one set of methods inside it, used when you want to predict an outcome for each customer, product or day.
Do business analysts need to code for machine learning?
Some BI and cloud tools offer point-and-click models, but Python gives you control over features, validation and comparison with a baseline. Most analyst roles that include modelling now expect Python or R alongside SQL.
Which machine learning models are used most in business analytics?
Logistic regression and gradient-boosted trees for yes-or-no outcomes such as response and churn, regression and time-series models for demand, and clustering for segmentation. Simple, explainable models are often preferred because managers need to trust the scores.
How do you show a model is worth using?
Compare it with the current rule on a later period it never saw, in the units the business cares about, such as extra sales found per hundred contacts. Where possible, confirm the gain with a randomised test.
Courses for these skills
- Practical Machine Learning in PythonFit outcome predictors without contaminating the experiment estimate.51 published lessons · more lessons in production
- Statistics and Probability for Machine LearningUnderstand randomisation, variance and confidence intervals.41 published lessons · all planned lessons open
- Data Tools: NumPy, pandas, Matplotlib and scikit-learnPrepare dated business tables and preserve feature availability.23 published lessons · all planned lessons open
New to programming or maths? Start with Python, or follow the machine learning path in order. Published lessons are free to read; compare Free and Pro.