What is data science?

Data science combines statistics, programming and domain knowledge to turn stored records into decisions. A kirana chain sitting on three years of billing data does not need more data; it needs someone who can ask the right question of it and hand the owner an answer they can act on this week.

The workflow

1. Ask a clear business question — Which customers are likely to churn next month?
2. Collect data from databases, APIs, spreadsheets.
3. Clean — handle missing values, duplicates, wrong types.
4. Explore (EDA) — distributions, relationships, outliers.
5. Model — statistics or machine learning.
6. Communicate — dashboards, reports, recommendations.

The list looks linear, but real projects loop. EDA shows that a third of the customer IDs are blank, so you go back to collection. The model works but the manager asks a slightly different question, so you reframe. Expect two or three passes before anything is final.

Where the time actually goes

Four kinds of question

Most of the value in an analytics team's first year comes from descriptive and diagnostic work done properly, not from machine learning.

A small taste of the loop

import pandas as pd

bills = pd.read_csv("bills.csv", parse_dates=["bill_date"])
monthly = (bills.groupby(bills["bill_date"].dt.to_period("M"))
                .agg(revenue=("net_amount", "sum"),
                     customers=("customer_id", "nunique")))
monthly["spend_per_customer"] = (monthly["revenue"] / monthly["customers"]).round(0)
print(monthly.tail(6))

Six lines, and you can already tell whether revenue grew because more people bought or because each person spent more. That one distinction decides whether marketing should chase new customers or push larger baskets.

Data science and its neighbours

• Data engineer — builds and maintains the pipelines and warehouse the rest of the team queries.
• Data analyst — answers descriptive and diagnostic questions, builds dashboards and reports.
• Data scientist — adds statistics and predictive modelling, and designs experiments.
• ML engineer — puts models into production and keeps them running.

In most Indian SMEs one person does all four. Knowing which hat you are wearing at a given moment keeps the work honest: a modelling problem and a plumbing problem need very different fixes.

Common mistakes

• Reaching for a model before checking whether a well-made table answers the question.
• Optimising accuracy when the business cares about cost, stock or churn value.
• Skipping the domain expert; the shop manager knows why Tuesdays look odd.
• Delivering a notebook instead of a recommendation with a number attached to it.

Try our free CSV / Excel Data Analyzer to run instant EDA on any file.