Sandhya Indurkar

Math, Applied

Exploring how math and data shape real-world decisions.

Breaking down complex systems into ideas we can actually use.

Ganita

Hi, I'm Ganita. Describe a decision or data problem in plain language and I'll point you to the best posts here.

Example: “We ran twenty A/B tests and three won. Should we ship them?”

Summarize the data

Means, spread, percentiles, and when one number misleads.

Prime Factorization Isn't Just Math: It's How You Break Down Real Problems

Prime Factorization Isn't Just Math: It's How You Break Down Real Problems

Clean batch splits, ETL jobs, and resource planning from number structure.

The Average Isn't the Answer: Mean and Median in Real Data

The Average Isn't the Answer: Mean and Median in Real Data

When one summary number misleads dashboards and product decisions.

Percent Change Isn't Intuitive: Growth Math in Real Decisions

Percent Change Isn't Intuitive: Growth Math in Real Decisions

Why recovery percentages do not undo drops, and how to set targets.

Variance and Spread: Same Average, Different Story

Variance and Spread: Same Average, Different Story

Standard deviation and range for delivery, ops, and reliability.

The Average User Isn't Average: Percentiles and Quartiles

The Average User Isn't Average: Percentiles and Quartiles

P90, SLAs, latency tails, and typical vs worst-case experience.

Weighted Averages: Roll Up the Number That Matches Volume

Weighted Averages: Roll Up the Number That Matches Volume

Simple vs volume-weighted rollups when segment sizes differ.

Up 10% Compared to What? Benchmarks and Baselines

Up 10% Compared to What? Benchmarks and Baselines

Same metric, different story: last month, last year, plan, or peer.

The New Users Look Great: Cohort Analysis in Real Decisions

The New Users Look Great: Cohort Analysis in Real Decisions

Headline retention vs signup cohorts: when the mix shifts and when product quality actually changed.

More Tickets, Same Workload? Rates vs Counts in Real Data

More Tickets, Same Workload? Rates vs Counts in Real Data

Total volume vs per-user rate: when headline counts rise because the base grew.

Up From Last Month: Is That Normal for March? Seasonality

Up From Last Month: Is That Normal for March? Seasonality

Month-over-month vs same month last year when the calendar drives the metric.

Experiments and uncertainty

Sample size, probability, intervals, and when to ship.

Twelve Data Points Isn't a Trend: Sample Size in Real Decisions

Twelve Data Points Isn't a Trend: Sample Size in Real Decisions

A/B tests, noisy averages, and when a readout is too thin to ship.

We Ran the Test: Could We Even See a Win? Statistical Power

We Ran the Test: Could We Even See a Win? Statistical Power

Detectability before you launch: sample size, baseline, and minimum lift together.

Probability in Real Decisions: Count Wins, Not Stories

Probability in Real Decisions: Count Wins, Not Stories

Estimate win rates from trials and know when the read is still shaky.

Will We Run Out? Probability of Stockout in Real Inventory Decisions

Will We Run Out? Probability of Stockout in Real Inventory Decisions

Set reorder points using explicit stockout risk instead of gut feel.

One Number Is Not Enough: Confidence Intervals

One Number Is Not Enough: Confidence Intervals

Ranges around conversion, CSAT, and defect rates before you ship.

Base Rates and Updating Beliefs: Rare Events, Loud Alerts

Base Rates and Updating Beliefs: Rare Events, Loud Alerts

Why a strong alert can still mean mostly false alarms when the base rate is low.

A/B Test Readouts: Significance Without Jargon

A/B Test Readouts: Significance Without Jargon

Lift, sample size, interval overlap, and when to ship without p-value talk.

When Everything Wins Once: Running Many A/B Tests

When Everything Wins Once: Running Many A/B Tests

False winners multiply when you run dozens of null tests in one sprint.

Simpson's Paradox: When Every Slice Wins but the Total Loses

Simpson's Paradox: When Every Slice Wins but the Total Loses

Segment tables before rollups so mix shifts do not flip the winner.

False Alarm vs Missed Win: Two Ways an Experiment Decision Goes Wrong

False Alarm vs Missed Win: Two Ways an Experiment Decision Goes Wrong

Ship a bad change vs kill a good one: pick which mistake you can afford.

Expected Value: Compare Bets Without Guessing

Expected Value: Compare Bets Without Guessing

Rank campaigns by upside, probability, and cost before spend.

Traps and causation

Bias, confounders, and conclusions that look right but are not.

Correlation Isn't Causation: Linked Data in Real Decisions

Correlation Isn't Causation: Linked Data in Real Decisions

Confounders, direction, and what you need before acting on r.

Your Best Customers Answered: Selection Bias

Your Best Customers Answered: Selection Bias

Surveys, betas, and who never made it into the sample.

Only the Winners Stay Visible: Survivorship Bias

Only the Winners Stay Visible: Survivorship Bias

Success stories hide failures that left the dataset before you measured.

You Hit the Target and Missed the Point: Goodhart's Law

You Hit the Target and Missed the Point: Goodhart's Law

When the metric becomes the goal, the dashboard greens while the real outcome reddens.

The Star Performer Slump: Regression to the Mean

Sales quotas, support metrics, and snap-back after extreme scores.

The Model Looked Perfect on Past Data: Overfitting in Real Decisions

The Model Looked Perfect on Past Data: Overfitting in Real Decisions

Train vs holdout: when forecasts memorize history instead of predicting new weeks.

Regression for Prediction: Data to Decisions

Regression for Prediction: Data to Decisions

Weekly orders from ad spend and email, with forecast, holdout, and scenarios.