The Data Science Bubble: Why 80% of Data Scientists Do Excel Work

You enrolled in that ML bootcamp dreaming of building AI. Now you wonder why job descriptions ask for SQL and Tableau. You need the real picture.
P. Mishra · January 2026 · Data Science
4 min read · Reviewed by Editorial Desk · Correction path: Contact

Key Takeaways

  • This piece focuses on data science realities in India, not outlier narratives.
  • Compensation numbers should be interpreted with role scope, market cycle, and switching friction.
  • Use decision frameworks and evidence checks before acting on title or salary headlines.

On This Page

The Expectation

The Data Science Dream:
  • Build AI/ML models that change the world
  • Work on cutting-edge algorithms
  • Six-figure salary for writing Python
  • Be the smartest person in the room

What Bootcamps Sell: Learn TensorFlow, build neural networks, become a data scientist in 6 months. Jobs everywhere. Companies desperate for talent.

The Reality

Reality Check - What Data Scientists Actually Do:

📊 Data Scientist Time Allocation (Industry Survey 2024)

ActivityBootcamp PromiseActual RealityAt FAANG
Building ML Models60%10%25%
Data Cleaning/Wrangling10%45%30%
SQL Queries5%20%15%
Dashboards/Reporting5%15%10%
Meetings/Communication10%10%15%
Model Deployment/MLOps10%0%5%

The Title Inflation Epidemic:

In 2015, there were about 10,000 "Data Scientists" in India. By 2024, there are 200,000+. Did data science work increase 20x? No. Companies renamed existing roles for:

  • Better hiring (everyone wants to be a data scientist)
  • Higher salaries for the same work
  • Marketing to clients ("We use data science!")

Reality: 70% of "Data Scientists" are glorified Business Analysts with Python.

📈 What Companies Call "Data Science"

What You ThinkWhat It Actually Is% of Roles
Building Neural NetworksWriting SQL queries40%
Training LLMsMaking PowerPoint charts25%
Deep Learning ResearchCleaning Excel files20%
Actual ML DevelopmentActual ML Development15%

The Education-Industry Gap:

What bootcamps teach: TensorFlow, PyTorch, Deep Learning, Neural Networks, Computer Vision What jobs actually need: SQL, Pandas, Excel, Tableau, Basic Statistics, Communication

Case Study - The ML PhD Who Quit:

Amit, PhD in Machine Learning from IIT, joined a "Data Science" role at a unicorn startup. Reality:

  • Week 1: Setting up Tableau dashboards
  • Month 1: Writing SQL queries for marketing reports
  • Month 3: Realized 90% of work is ad-hoc business queries
  • Month 6: Quit for an actual ML Engineer role at 30% lower salary

The title said "Data Scientist." The job was "Business Intelligence Analyst."

The Skills That Actually Get Used:

📊 Skills Usage in "Data Science" Roles

SkillHow Often UsedHow Much Studied
SQLDaily1 week in bootcamp
Excel/SheetsDailyOften skipped
Python basicsWeeklyModerate
CommunicationDailyNever taught
Deep LearningRarely50% of bootcamp
NLP/LLMsAlmost Never20% of bootcamp

Related context: Salary Reality Check, CTC Decoder, more in Data Science.

Salary and Growth Reality

The Data Science Salary Reality:

💰 Salary by Actual Work Done (Not Title)

Actual Role0-2 Years3-5 Years6+ Years
ML Engineer (Real)Rs 15-25 LPARs 30-50 LPARs 50-80 LPA
Data Scientist (Research)Rs 12-20 LPARs 25-40 LPARs 45-70 LPA
DS (Analysis Focus)Rs 8-15 LPARs 15-25 LPARs 25-40 LPA
Analyst w/ DS TitleRs 6-10 LPARs 12-18 LPARs 18-28 LPA

The Top 5% vs Bottom 95%:

The salaries you see on LinkedIn and in news articles? Those are for the TOP 5% - people at FAANG, doing actual ML at scale.

For most people with "Data Scientist" in their title:

  • Starting: Rs 6-10 LPA
  • 3 years: Rs 12-18 LPA
  • 5 years: Rs 18-25 LPA

This is analyst money with fancier title.

Cross-check your take-home with the CTC Decoder and compare ranges in Salary Reality.

Where Most People Get Stuck

Where DS Aspirants Get Permanently Trapped:

Trap 1: The Kaggle Paradox You can win Kaggle competitions but cannot write production code. Companies need code that runs in prod, not notebooks that win medals. Kaggle is training you for a job that barely exists.

Trap 2: The Framework Obsession You know PyTorch AND TensorFlow AND JAX. You cannot write a SQL join. 95% of DS interviews have SQL rounds. You fail them.

Trap 3: The Research Fantasy You want to do "research" like you read about. Real DS research roles: maybe 500 positions in all of India. You are competing with PhDs from top global universities.

Trap 4: The Tool Collector Your resume lists: Python, R, SQL, Tableau, Power BI, Spark, Hadoop, TensorFlow, PyTorch, Keras, Scikit-learn... You are master of none. Depth beats breadth.

Breaking Free:

For actual ML work:

  • Target ML Engineer titles specifically
  • Focus on deployment (Docker, Kubernetes, MLflow)
  • Join companies with real ML in production (not "AI-powered" marketing)

For well-paying analyst work:

  • Accept the reality, optimize for it
  • Master SQL, communication, business metrics
  • Move toward Analytics Manager track

The middle path (vague "Data Scientist") leaves you in no-man's-land.

If this matches your current situation, run the Resignation Risk Analyzer before making your next move.

Who Should Avoid This Path

If you work at a FAANG company building actual ML systems, this does not apply to you. You are in the 5% doing real data science.

Decision Framework

Use this quick framework before changing role, company, or specialization.

  • If your take-home is not compounding with experience, benchmark externally before accepting internal narratives.
  • If role expectations keep rising without title/pay movement, escalate with documented outcomes.
  • If growth path is unclear beyond 6-9 months, run a switch-or-specialize decision cycle.

Common Mistakes Checklist

  • Treating outlier salaries as planning baselines.
  • Using title changes as a substitute for capability changes.
  • Delaying market benchmarking until after compensation stagnates.
  • Over-indexing on model demos without production deployment depth.

Real Scenario Snapshot

A professional stays in-role despite rising responsibility and flat pay. Growth recovers only after external benchmarking and a deliberate switch-or-specialize decision.

Originality Lens

Contrarian thesis: Career outcomes usually degrade from quiet trade-offs, not sudden failures.

Non-obvious signal: When responsibility rises but decision rights stay flat, stagnation risk rises even before pay slows.

Evidence By Section

Claim: Popular career narratives overweight edge cases and underweight base-rate outcomes.

Evidence: AmbitionBox Salary Insights, Glassdoor India Salaries

Claim: Observed market behavior diverges from social-media compensation storytelling.

Evidence: Glassdoor India Salaries, LinkedIn Jobs (India)

Claim: Salary and growth ranges vary by company type, leverage, and cycle timing.

Evidence: AmbitionBox Salary Insights, Glassdoor India Salaries, LinkedIn Jobs (India), Naukri Jobs (India)

Claim: Career plateaus are often linked to stale scope, weak mobility planning, and evidence gaps.

Evidence: LinkedIn Jobs (India), Naukri Jobs (India), Kaggle State of Data/AI

Final Verdict

The Uncomfortable Truth:

Data Science is real. But "Data Science jobs" are mostly not Data Science.

If you want actual ML work:

  • Target FAANG, well-funded AI startups, research labs
  • Accept Rs 0 during PhD or research fellowship
  • Build deployed projects, not notebooks
  • Apply for "ML Engineer" not "Data Scientist"

If you want good money and work-life balance:

  • Accept the analyst reality
  • Optimize SQL, BI tools, communication
  • Move toward Analytics Lead/Manager
  • Stop chasing the ML dream that company cannot use

If you are already in a "DS" role doing analyst work: You have two choices:

  1. Accept it, grow within it, become Analytics Manager
  2. Skill up for real ML roles, be prepared to job hunt for 6+ months

What does NOT work: Staying frustrated in a mismatch, collecting more certificates, hoping it changes.

The Test: On your current project at work, do you use:

  • Neural Networks? → You are doing actual DS
  • Gradient Boosting at scale? → Maybe actual DS
  • SQL and Tableau? → You are an analyst (that is okay!)

Stop letting job titles lie to you.

Last Updated: January 12, 2026
Found a factual error? Request a correction.

What Changed

  • January 12, 2026: Reviewed salary ranges, corrected stale assumptions, and tightened internal links for related reads.
  • January 12, 2026: Initial publication with baseline market framing and trade-off analysis.

Sources