Binary logistic regression
Concentrix: who will miss a premium?
Built a payment-default flag from auto-debit status and days past due, modeled it across 24,399 policy records, and read accuracy, sensitivity and specificity from the confusion matrix. Compared calls, SMS and email on a cost-versus-renewal-impact matrix.
Logistic regression · Confusion matrix · Cost matrix
Artificial neural network
Wallet churn with a one-node network
Min-max scaled usage and tenure, split the data 75/25, trained a neuralnet model to predict customer churn, plotted the network and checked it against a confusion matrix.
R · neuralnet · caTools · Scaling
Classification
Scaleneworks: predicting offer renege
Modeled whether 12,333 candidates who accepted an offer would actually join, using notice period, offered versus expected hike, joining bonus, relocation and sourcing channel.
HR analytics · Behavioral modeling
Retention modeling
Scholastic Travel: will the group come back?
Worked with 2,389 school trips and more than 50 variables, from parent meetings and payment plans to school size, to predict whether a group would re-book the next year.
Customer retention · Feature engineering
Demand forecasting
iD Fresh Food: forecasting perishable demand
Prepared beat-level daily sales for a short-shelf-life SKU, flagged outliers with the interquartile range, and framed a forecast where both stock-outs and waste are expensive.
Excel · IQR outliers · Time series
Linear programming
Shawls or Stoles: the most profitable mix
Costed seven woollen products, wrote the objective function to maximize gross profit, and solved for production quantities under 15 days of weaving and knitting labour.
Excel Solver · Objective function · Constraints
Scenario analysis and NPV
Trionym Systems: should the RB-5000 launch?
Built every combination of market size, share and production cost, projected five years of after-tax cash flow, and ruled out the scenarios that didn't clear the $50M investment by 10%.
Scenario modeling · Depreciation · Present value
Conjoint analysis
What buyers trade off
Used dummy-variable regression to turn preference rankings into part-worths: juice flavor, pack and size for a beverage brand, and city and plant size for Trionym's India factory.
Excel regression · Part-worth utilities
Analytic hierarchy process
Choosing a brand ambassador
Turned pairwise judgments on brand equity and fitness into weighted scores to choose among three Indian athletes for a sports-apparel launch, then applied the same method to a market-entry choice.
AHP · Pairwise comparison · Weighting
Market sizing
ByteIQ: license it or build a platform?
Estimated customers and first-year revenue for an analytics product under two business models, using the technology adoption curve for innovators and early adopters.
TAM · Adoption lifecycle · Revenue model
Apriori association rules
BigBasket: "Did you forget?"
Mined basket data for item pairs at 50% minimum support and 70% confidence, the logic behind a grocery app's Smart Basket and forgotten-item prompts.
Market basket analysis · Support · Confidence