Data science internship hiring in India is done by four quite different kinds of employer, and they are not competing for the same candidate despite using the same job title.
Understanding which archetype you are — and therefore which candidates you should actually be chasing — is worth more than any list of company names.
Live from our candidate database
Counted at render time and refreshed every six hours. This is a supply-side view of one platform, not a national labour-market statistic.
Skills present in this pool
The four archetypes
Analytics GCCs hire the largest volumes and screen primarily on SQL and business reasoning. The work is closer to analytics than to research, and candidates who over-index on modelling often struggle with the stakeholder half of the job.
Product companies hire smaller numbers and screen on engineering ability alongside statistics — their data scientists ship code. Consulting firms screen on structured communication above all, because the output is a recommendation, not a model. And research-adjacent teams, the smallest group, screen on depth in a narrow area and tolerate gaps everywhere else.
The certificate problem
No early-career category has a worse signal-to-noise ratio on credentials. A large share of candidates hold at least one data science certificate, and the correlation between holding one and being able to do the work is close to nil.
The screen that works is dull and effective: give the candidate a small, messy dataset and ask them to answer a business question with it. Candidates who have only ever seen clean teaching data reveal themselves within minutes, and no amount of coursework compensates.
What separates the top decile
Three things, consistently. First, the ability to say why a model was chosen rather than which model was used — reasoning about the trade-off is what generalises. Second, evidence of having cleaned real data, with the specific war stories that only come from having done it. Third, discipline about evaluation: a candidate who volunteers how they knew their result was not an artefact is in a different tier.
None of these are visible on a résumé, which is why data science is the category where interview design pays the largest dividend.
Where to look
The data science pool draws from a wider degree base than most technical categories — statistics, mathematics and economics candidates frequently outperform computer science candidates on the reasoning half of the role, while lagging on the engineering half.
If your archetype is the GCC or the consulting firm, deliberately widening beyond engineering campuses is one of the highest-return changes you can make to a data science pipeline. If you are a product company, it is likely the wrong move — you need the engineering half.
Key takeaways
- Four employer archetypes hire "data science interns" and screen for genuinely different things — know which you are.
- Certificates carry almost no signal in this category; a messy-dataset exercise does.
- The top decile explain why, not what: model choice, real data cleaning, and evaluation discipline.
- Statistics, maths and economics candidates outperform on reasoning; widen beyond engineering campuses unless you need shipping ability.
Questions
Should we require a data science degree?+
Rarely. Requiring a specific degree cuts a large share of the strongest reasoning candidates out of the funnel, and the degree itself predicts very little at entry level. Screen on the exercise instead.
What is a realistic screening pass rate?+
Low, and you should plan for it. A well-designed practical exercise in this category typically eliminates most applicants — that is the exercise working, not a sourcing failure. Size the top of your funnel accordingly.
Do we need a separate pipeline for ML engineering?+
If the role involves production deployment, yes. Data science and machine learning engineering pools overlap but the screens differ: one is about inference, the other about serving. Hiring one for the other is a common and expensive mistake.