Customer Analytics with Python Training Course

Participants can structure a reproducible Python project for customer analytics work.

• Jupyter notebooks organised with inputs, transformations, outputs and documented assumptions
• Python environments managed through virtual environments and requirements files
• pandas, NumPy and Plotly imported using standard analytical conventions
• Project folders separate raw data, processed data, notebooks and reports
• Functions reduce repeated cleaning and metric calculation code
• Random seeds recorded for repeatable sampling and model results

Participants can import customer datasets and diagnose their analytical fitness.

• CSV, Excel and Parquet files loaded with pandas readers
• Data types checked for identifiers, dates, categories, currency and quantities
• Missingness profiles calculated by field, customer segment and reporting period
• Duplicate customer records identified using keys, names and contact attributes
• Frequency tables reveal invalid categories, rare values and coding inconsistencies
• Descriptive statistics expose implausible values and extreme transaction amounts

Participants can clean customer records and create a reliable customer-level dataset.

• Primary keys assessed for uniqueness, stability and cross-system coverage
• Standardised dates, currencies and categorical values prepared for analysis
• Missing values treated through business rules, imputation or exclusion
• Customer records joined using one-to-one, one-to-many and many-to-many relationships
• Entity resolution rules reconcile duplicate names, emails and account identifiers
• Data quality logs document corrections, exclusions and unresolved exceptions

Participants can identify meaningful patterns in customer activity and value.

• Distribution plots compare purchase frequency, order value and engagement measures
• Customer activity analysed by channel, product, geography and acquisition source
• Time-series charts identify seasonality, growth patterns and anomalous periods
• Cross-tabulations reveal behavioural differences between customer groups
• Percentiles distinguish typical customer value from high-value tail behaviour
• Outlier treatment decisions preserve legitimate premium customer activity

Participants can calculate consistent customer metrics from event and transaction data.

• Active customer definitions specify qualifying events and observation windows
• Acquisition, activation and repeat-purchase rates calculated from customer histories
• Average order value separated from revenue per active customer
• Retention rate formulas distinguish gross retention from net revenue retention
• Churn definitions account for contractual, behavioural and voluntary attrition
• Metric dictionaries record formulas, exclusions, refresh frequency and owners

Participants can build cohort tables and interpret retention patterns over time.

• Acquisition cohorts assigned using first purchase, signup or activation dates
• Retention matrices calculated by cohort age and calendar period
• Revenue cohorts separate customer survival from spending behaviour
• Heatmaps reveal retention differences across acquisition periods and channels
• Survival curves show probability of remaining active over time
• Cohort findings converted into onboarding, service and reactivation questions

Participants can create interpretable customer segments using recency, frequency and monetary value.

• Recency calculated from each customer's most recent qualifying transaction
• Frequency windows selected to reflect purchase cycles and product cadence
• Monetary value defined using revenue, margin or contribution measures
• RFM scores assigned with quantiles, fixed thresholds and business rules
• Segment labels distinguish loyal, promising, at-risk and dormant customers
• Segment profiles compare value, behaviour, channel preference and retention

Participants can create customer-level features suitable for segmentation and prediction.

• Transaction histories aggregated into customer-level behavioural feature tables
• Purchase intervals calculated to represent regularity and recency patterns
• Channel mix features quantify digital, branch, call-centre and partner usage
• Product breadth measures capture cross-sell depth and portfolio concentration
• Engagement features combine sessions, opens, clicks and service interactions
• Feature windows prevent future information entering historical prediction datasets

Participants can develop and assess clustering-based customer segments.

• Feature scaling prepares monetary, frequency and ratio variables for clustering
• K-means clustering groups customers by multidimensional behavioural similarity
• Elbow plots and silhouette scores inform cluster number selection
• Cluster profiles compare value, needs, engagement and attrition indicators
• Principal component analysis supports visual inspection of cluster separation
• Business usability tests determine whether clusters support differentiated actions

Participants can estimate customer lifetime value using transparent commercial assumptions.

• Historical customer value calculated from revenue, margin and service costs
• Discounted cash flow estimates incorporate contribution margin and discount rates
• Retention assumptions linked to observed cohort and survival patterns
• Simple CLV models distinguish contractual from non-contractual customer relationships
• Acquisition cost compared against expected lifetime contribution by segment
• Sensitivity analysis tests value estimates under alternative retention assumptions

Participants can define a churn prediction problem with valid labels and observation periods.

• Prediction targets distinguish cancellation, inactivity and revenue decline outcomes
• Observation windows separate available predictors from future churn labels
• Positive and negative classes checked for imbalance and label reliability
• Business intervention windows determine useful prediction lead time
• Train, validation and test periods preserve chronological data order
• Baseline rules establish comparison points before machine learning models

Participants can train baseline and supervised churn classification models in Python.

• Logistic regression provides interpretable baseline churn probability estimates
• Decision trees identify nonlinear risk patterns and actionable splits
• Random forest models capture interactions among behavioural customer features
• scikit-learn pipelines combine preprocessing, modelling and evaluation steps
• Class weights address imbalanced churn outcomes without discarding observations
• Hyperparameter searches compare candidate models using validation data

Participants can evaluate churn models against operational costs and intervention capacity.

• Confusion matrices quantify true positives, false positives and missed churners
• Precision and recall assessed against campaign capacity and contact costs
• ROC-AUC compared with precision-recall curves for imbalanced classification
• Probability calibration checks whether predicted risk matches observed churn
• Decision thresholds selected using expected value and intervention constraints
• Lift charts rank customers by concentration of churn risk

Participants can assess campaign results using suitable comparison groups and metrics.

• Campaign funnels measure delivery, response, conversion and downstream value
• Control groups estimate incremental outcomes beyond natural customer behaviour
• Randomised test design addresses selection bias in campaign evaluation
• A/B test results interpreted using confidence intervals and practical significance
• Uplift metrics compare treatment performance against control group outcomes
• Attribution assumptions documented for multichannel customer journeys

Participants can select simple recommendation approaches for cross-sell and content personalisation.

• Recommendation objectives distinguish relevance, conversion, margin and customer experience
• Popularity models provide transparent baselines for product recommendation
• Content-based filtering matches customer preferences with product attributes
• Collaborative filtering uses similar customer behaviour for recommendations
• Cold-start problems affect new customers, products and sparse datasets
• Offline evaluation uses precision at k and recall at k

Participants can communicate customer findings through clear visual and written narratives.

• Chart selection matches comparison, distribution, trend and relationship questions
• Plotly charts add filters, hover details and stakeholder-ready annotations
• Segment dashboards combine size, value, risk and recommended treatment
• Executive narratives state decision, evidence, uncertainty and proposed action
• Model explanations distinguish correlation, prediction and causal claims
• Recommendations prioritised by expected value, feasibility and implementation ownership

Participants can apply governance controls to customer analytics datasets and models.

• Purpose limitation defines permitted uses of customer personal data
• Consent status and lawful basis recorded for analytical data use
• Data minimisation removes unnecessary identifiers from analytical datasets
• Protected characteristics assessed for unfair segment or model effects
• Model documentation records data sources, assumptions, performance and limitations
• Monitoring plans detect drift, degraded accuracy and changed customer behaviour

Participants can plan how customer analysis will become a repeatable operational process.

• Analytical workflows scheduled around data refreshes and decision cycles
• Data pipelines specify source systems, transformations, checks and outputs
• Segment refresh rules maintain stable customer treatment definitions
• Model monitoring dashboards track performance, drift and intervention outcomes
• Stakeholder handoffs assign ownership for actions, exceptions and feedback
• Implementation roadmaps sequence quick wins, dependencies and measurement milestones

Participants can build, review and present a complete customer analytics project with actionable recommendations.

• Business question defined with target customers, decisions and success measures
• Customer dataset cleaned, joined and documented using reproducible Python code
• Segments, cohorts and value metrics developed from behavioural features
• Churn or campaign analysis evaluated with appropriate business metrics
• Governance risks, assumptions and limitations reviewed before stakeholder presentation
• Final recommendations prioritised with owners, timing and measurement criteria