Data science in India: the honest picture.

Where the hiring actually is.

If you picture data science in India as a room full of researchers in a consumer startup, you are picturing about five per cent of the market. The bulk of the work now sits inside organisations that have been collecting operational data for years and have finally built teams to use it. That shift matters, because it changes what a hiring manager is looking for.

Global capability centres are the single largest employer group. These are the India based engineering and analytics arms of overseas banks, retailers, insurers, airlines and pharmaceutical companies, concentrated in Bengaluru, Hyderabad, Pune, Chennai and Gurugram. They hire for defined, long running data functions rather than for experiments. Behind them sit the IT services and consulting firms, which staff data projects for client accounts, and the domestic banking, insurance and fintech sector, where the work is credit scoring, fraud detection, collections propensity and regulatory reporting.

Kerala sits inside that picture with its own tilt. Technopark in Thiruvananthapuram, Infopark in Kochi and Cyberpark in Kozhikode carry real volume in analytics, business intelligence and data engineering, much of it delivered for overseas clients. What Kerala has proportionally less of is deep research modelling, which still clusters around Bengaluru and Hyderabad. For most people reading this that is a detail rather than a barrier, because the first job is almost never a research job anyway.

Four jobs wearing one label.

Job boards use data science as an umbrella term, which is why so many applications land in the wrong pile. The four roles below want overlapping skills in genuinely different proportions, and knowing which one you are applying for is most of the battle in a screening call.

Two practical consequences follow. First, SQL is the common denominator across all five, and it is also the round most candidates fail, so it deserves more of your practice time than any single algorithm. Second, the data engineer route is the least crowded of the entry points and pays comparably, which makes it worth a serious look if pipelines interest you more than models do. The data and AI course track maps to these roles rather than to the umbrella term.

What the market actually pays.

Pay is the reason most people search this topic, so here is the shape of it, with the usual warning attached. These are market observations for India as a whole, not outcomes we are claiming for anybody, and the spread inside each band is wide because employer type moves the number more than job title does.

Three things move a candidate inside those bands. Employer type is the biggest lever: a global capability centre and a services firm can offer materially different packages for the same title and the same years. Location is the second, with Bengaluru and Hyderabad carrying a premium over Kochi and Thiruvananthapuram that partly disappears once rent is subtracted. Domain is the third, and it is the one people underrate, because a data scientist who understands credit risk or clinical trials is competing in a much smaller pool than a generalist. Our data analytics salary guide breaks the analyst side of this down stage by stage.

Getting in from a Kerala degree.

Almost every degree in Kerala is a workable starting point, and none of them is sufficient on its own. B.Tech graduates arrive with programming and are usually short on statistics. B.Sc statistics and mathematics graduates arrive with the reverse problem and tend to move fastest once they learn to code. BCA and MCA graduates are well placed for the data engineering side. B.Com and BBA graduates are not shut out at all, because commercial fluency is genuinely scarce in analytics teams, but they carry the longest technical run up.

The order you learn things in matters more than the total hours. SQL first, until joins, window functions, aggregation and date logic are automatic rather than looked up. Python next, specifically pandas and numpy, with an emphasis on cleaning messy input rather than on syntax drills. Statistics third, and properly: sampling, distributions, hypothesis testing, confidence intervals and the failure modes such as survivorship bias and Simpson's paradox that an interviewer will probe. Machine learning fourth, by which point you will understand why evaluation metrics matter more than model choice. One business intelligence tool and one cloud platform after that.

Internships are worth more than an extra certificate at this stage, including the structured internships that FYUGP students in Kerala can now take for credit. Six months of touching real, badly behaved company data changes how you talk in an interview in a way no course can replicate. If you want the full topic list before deciding, the data science and AI syllabus sets out what a complete curriculum should contain, and you can compare any provider against it.

The gap between a certificate and a portfolio.

This is the part most course pages skip. A certificate confirms attendance. It does not tell a hiring manager whether you can be trusted with a live dataset, and every other applicant has one too. What actually shortens the job search is a small number of projects that a reviewer can open, run and interrogate.

The standard classroom projects will not do this for you. A notebook that classifies the iris dataset or predicts survival on the Titanic tells a reviewer only that you followed instructions, because the data was clean, the target was given and the answer was already known. Reviewers have seen thousands of them and skim straight past.

Two or three projects at this standard beat ten notebooks every time. Build them while you study rather than after, and pick problems from a domain you already know, because your existing context is the unfair advantage nobody else in the queue has. Our data science and AI course is built around a reviewed capstone for exactly this reason.

What the interview actually tests.

Indian data hiring has settled into a fairly predictable set of rounds, and preparing for the wrong one is a common and expensive mistake. Expect four.

The SQL round comes first and eliminates the most people. It is live, it is against a schema you have not seen, and it targets joins with deliberate traps, window functions for ranking and running totals, deduplication, date bucketing and the difference between filtering in a where clause and in a having clause. The statistics round follows, and it is conceptual rather than computational: what a p value does and does not mean, how you would detect sampling bias, why an aggregate can reverse when you split it by group, and how you would design a valid test for a proposed change.

The case round is where analysts are separated from candidates who have only trained models. You are given a business situation, usually a metric that has moved, and asked how you would investigate it, what data you would want and what you would recommend. Structure matters more than cleverness. Finally the machine learning round, where the recurring topics are the bias and variance trade off, regularisation, class imbalance, target leakage, cross validation done correctly on time series, and why accuracy is the wrong metric for most real problems. Very few interviewers ask you to derive an algorithm. Most ask you to explain a decision.

The questions people actually ask.

The pages this guide points to.