The business model behind free platforms-If You’re Not Paying, You’re the Product

THE DEAL NOBODY SAYS OUT LOUD

Every free platform, Instagram, Google, YouTube is funded by advertisers, not you. You pay with your data: your clicks, your searches, your 2 AM rabbit holes, the reel you re-watched three times.

The algorithm’s real job isn’t to serve you. It’s to keep you there long enough for ads to reach you.

That’s why outrage spreads faster than nuance online. Angry people click more, share more, comment more. The algorithm rewards this. Platforms profit. It’s not a bug; it’s the business model.

You are not the customer. You are the inventory.

Platforms know your political leaning, relationship status, and income bracket, often better than you’d expect. This data gets auctioned to advertisers in real time, before the page even loads.

THE BUSINESS MODEL, PLAINLY STATED

Meta, Google, TikTok, these are not really tech companies. They’re advertising companies that built very good technology to keep you in one place. The more time you spend, the more they learn. More precisely, they can sell your attention.

You create content and spend time → the platform collects your behaviour → behaviour gets turned into a profile → the profile gets sold to advertisers → advertisers pay to reach you at exactly the right emotional moment → revenue funds better features → better features keep you longer → the cycle repeats.

The scariest part isn’t that this exists. It’s that it works so well that most people never think about it.

WHERE BIAS ENTERS

Algorithms don’t just recommend videos. They decide who sees job listings, who gets a loan approved, who gets flagged by a facial recognition system. These systems are trained on historical data, a record of the world as it was, not as it should be.

If past decisions were discriminatory, the algorithm learns those patterns and repeats them at millions of decisions per second.

• Historical Bias : Training data reflects past unfairness directly.
• Representation Bias : Some groups barely exist in the data. Facial recognition fails up to 34.7% on darker-skinned women vs 0.8% on lighter-skinned men.
• Measurement Bias : The wrong thing is measured as a proxy for capability.
• Feedback Loop Bias : Biased outputs create new data, which trains the next model, making it more confidently wrong.

REAL CASES THAT PROVE IT

• Amazon (2018) trained a hiring AI on 10 years of CVs. It started penalising CVs that said “women’s” and downranked graduates from all-women’s colleges. Quietly scrapped after internal testing.
• Apple Card (2019) gave men up to 20x higher credit limits than their wives, even when wives had better credit scores. The algorithm “didn’t use gender.” It didn’t need to.
• COMPAS (US Courts) was twice as likely to wrongly flag Black defendants as high risk for bail decisions. People lost their freedom over an algorithm nobody audited
• Facebook Ad Targeting let advertisers exclude people from seeing housing and job listings through interest proxies, recreating illegal racial discrimination, just with better branding.
• Healthcare Algorithms recommended less care for Black patients than white patients with identical illness levels, because spending history was used as a proxy for health need.

THE BIGGEST LIE : “Algorithms Are Neutral”

Technically true. Practically meaningless. Math doesn’t care about fairness; it optimises for whatever it’s told to. And the people doing the telling are humans, with blind spots, deadlines, and quarterly targets.

• When a human rejects you, you can ask why and challenge it.
• When an algorithm does it, the answer is “the model said so”, a black box nobody’s required to open.
• Scale is what makes it dangerous, a biased human rejects hundreds. A biased algorithm rejects millions, overnight, silently.

The feedback loop makes it worse. A biased decision creates new data. New data trains the next model. The next model doubles down with more confidence.

THE INDIA ANGLE

This isn’t just a Western problem. As algorithmic systems expand across India for credit, welfare, hiring, policing, the same dynamics apply with less oversight and higher stakes.

• Alternative credit scoring using phone and location data can quietly encode caste, regional, and gender gaps.
• Facial recognition is deployed across Indian cities with no public accuracy data by demographic.
• AI content moderation fails regional languages, hate speech slips through, legitimate speech gets removed.
• Aadhaar-linked welfare systems have documented exclusion errors affecting the most vulnerable.
• There is currently no mandatory algorithmic audit requirement before deployment in India.

WHAT ACTUALLY HELPS

• Mandatory audits before high-risk AI are deployed, the EU’s AI Act is leading here.
• Right to explanation : you should get a real answer when an algorithm makes a decision about your life.
• Fairness-aware ML : techniques that test for bias during training, not after the damage is done.
• Diverse training data : still massively underfunded despite being the most obvious fix.
• Whistleblower protections : for engineers who flag biased systems internally.
• Public pressure : companies move when reputation is on the line. Awareness is the first domino.

THE ONE THING TO REMEMBER

The “free” internet was never free. The cost was always data. And when that data trains systems that decide who gets a job, a loan, bail, or a life opportunity, the stakes stop being abstract really fast.

A system that efficiently reproduces injustice isn’t a technical achievement. It’s just injustice at scale.