Descriptive Statistics for Interviews: Centre, Spread and Shape Made Simple
The most dangerous sentence in analytics is: “The average is fine.” A quick-commerce city can show a healthy average delivery time while a small but painful tail of customers waits far too long - and that tail is where churn, refunds and angry social posts live.
- Descriptive statistics summarize what happened in a dataset; they do not prove why it happened.
- Always read a dataset through three lenses: centre - typical value, spread - variability, and shape - pattern of the distribution.
- Mean is useful for balanced data but gets pulled by outliers; median is safer for skewed business data like income, delivery time or order value.
- Standard deviation measures average fluctuation around the mean; IQR focuses on the middle 50% and resists outliers.
- Skewness tells you whether the tail is on the left or right; right-skewed data is common in revenue, waiting time and transaction size.
- A strong answer never quotes one number alone. Pair centre with spread and shape: “Median is 23.5 minutes, mean is 27.2, so a right-tail outlier is inflating the average.”
Big Picture: The Three-Question Dashboard
Descriptive statistics is not a formula list. It is a compact dashboard for turning raw rows into business meaning: what is typical, how much it varies, and whether the distribution has a dangerous tail.
Core Explanation: Centre, Spread and Shape
Centre answers: “What is a typical value?” The main measures are mean, median and mode.
- Mean = sum of all values divided by number of values. Best when data is reasonably symmetric.
- Median = middle value after sorting. Best when data has outliers or skew.
- Mode = most frequent value. Useful for categorical or repeated-value data, such as most common payment mode or shirt size.
Spread answers: “How inconsistent is the data?” Two teams may have the same average sales but very different risk if one team is highly volatile.
- Range = maximum minus minimum. Quick but sensitive to extremes.
- Variance = average squared distance from the mean. Mathematically important but not intuitive in original units.
- Standard deviation = square root of variance. Interpretable in the same unit as the data.
- Interquartile range = Q3 minus Q1. Captures the middle 50% and resists outliers.
Shape answers: “What does the distribution look like?” This is where many candidates miss the story. A long right tail means a few very high values; a long left tail means a few very low values.
Worked Example: One Outlier Changes the Story
Suppose a city operations manager reviews 10 delivery times in minutes:
18, 20, 21, 22, 23, 24, 25, 26, 28, 65
The business read is sharp: normal operations are stable, but the long right tail needs root-cause analysis. That could mean rider shortage, restaurant delay, payment issue, bad address quality or weather disruption - descriptive statistics tells you where to investigate, not the final cause.
Key Measures, Formulas and What “Good” Looks Like
There is no universal good number in descriptive statistics; it depends on the KPI. But interviewers expect you to know the formula and the practical signal each measure gives.
Definitions: Say These Cleanly
NIST: “Descriptive statistics are numbers that are used to summarize and describe data.”
Zepto: The Average Delivery Time Is Not the Promise
Zepto’s quick-commerce model shows why operations teams must read delivery performance through centre, spread and shape, not just an average.
Situation: Quick commerce in India competes on speed, reliability and local availability. For a company like Zepto, a city-level average delivery time can hide the real customer experience. If most orders are fast but a small tail is painfully delayed, the average may still look acceptable while customer trust weakens.
The move: The operating model is built around dense dark stores close to demand, localized assortment, technology-led routing, fast picking and replenishment discipline. The primary driver is the dark-store proximity model; supporting drivers include curated SKUs, inventory visibility, picker SOPs, rider allocation and demand forecasting.
How descriptive statistics fits: A manager would not stop at mean delivery time. They would track median delivery time for the typical order, IQR for consistency, P90 or P95 for the slow-tail experience, and skewness to see whether a few extreme delays are pulling the average upward.
Outcome or lesson: Descriptive statistics turns an operating promise into a measurable control system. The strategic “so what” is simple: in service businesses, the average may satisfy a dashboard, but the tail decides customer memory.
How AI Changes Descriptive Statistics
AI does not replace descriptive statistics; it makes the first pass faster and the second pass more dangerous if you stop thinking. In 2026, the best analysts use AI to profile data quickly, then verify the logic themselves.
- Automated data profiling: Tools can instantly summarize missing values, outliers, duplicates, skewed columns and suspicious category levels. This speeds up the “what am I looking at?” stage.
- Natural-language business intelligence: GenBI systems let managers ask, “Which city has the highest P90 delivery time?” and receive charts plus summaries. The risk is hallucinated interpretation if the data model is wrong.
- Smarter anomaly detection: ML models can flag unusual spikes in delivery time, payment failure, sales returns or churn. Descriptive statistics still anchors the explanation: mean, median, spread and shape tell you how unusual the event is.
Use ChatGPT or Claude with a small CSV: ask it to compute mean, median, IQR, standard deviation and skewness, then ask, “Which single metric would mislead a manager here and why?” Always verify formulas in Excel, Python or Google Sheets.
Interview Relevance
“You are given sales or delivery-time data. How would you summarize it for a business stakeholder, and when would you prefer median over mean?”
A strong answer sounds like this: “I would not present the mean alone. I would compare mean versus median, add IQR or standard deviation, then check the histogram for skew. If there is a long right tail, I would use median as the typical value and separately investigate the outliers.”
Common Mistake
The mistake: Quoting the average and stopping there. It costs candidates because it shows formula memory, not business judgment. One-line fix: Always say, “I will pair the average with median, spread and shape before making a decision.”
What to Revise Next
Now move from summarizing data to questioning whether the summary is misleading. Revise these next in order: