Generative AI Meets Marketing: A Seven-Step Framework Called MARK-GEN, and India's Real Ledger

A while back, a friend in e-commerce vented to me.

His company had just rolled out ChatGPT for product copywriting. The team of six got cut to three — one to supervise the AI, one to run campaigns. His exact words: "We saved 30% on creative outsourcing. But the more I use it, the more anxious I get — I honestly can't tell if we're doing this right or wrong."

I didn't rush to weigh in. Because this anxiety isn't his alone.

India has 850 million internet users, 700 million smartphone users, 22 official languages, and 1,599 dialects. In 2024 alone, the digital advertising market reached 6.7 trillion rupees, climbing at 30% year over year. A pie this size — when generative AI arrives, everyone wants a bite. Flipkart used regional-language ads to boost click-through rates in Tier 2 and 3 cities by 18%. Hindustan Unilever fed 40 years of advertising archives to AI and slashed creative costs by a quarter. MakeMyTrip plugged in ChatGPT for itinerary planning and saw NPS (Net Promoter Score) jump 12 points. HDFC's AI-driven personalized emails pushed click-through rates up 25%.

The numbers are beautiful.

But here's the problem: the vast majority of Indian companies — especially the 40+ million small and medium businesses — are still in the "let's give it a shot" phase. Today they use Jasper to draft a couple of headlines. Tomorrow they generate an image with DALL·E. The day after, they hear MidJourney is better and switch again. No method to the madness.

It's like an army where every soldier has a great gun, but no one's ever taught them how to form ranks.

What Is MARK-GEN?

In 2025, a research study zeroed in on exactly this question: how should Indian companies systematically embed generative AI into marketing? The researchers interviewed 15 senior marketing leaders, ran two SME focus groups, and combed through public case studies from Flipkart, HUL, Paytm, and OYO. What they delivered was called MARK-GEN.

So what is MARK-GEN?

In plain terms: seven steps. From figuring out what you want to achieve, all the way to deploying the model, collecting feedback, and continuing to train it. The letters M-A-R-K stand for Marketing, and G-E-N for Generative. The name isn't particularly elegant, but the logic flows.

Let me break down these seven steps for you. For each one, I'll pair it with a real example that happened in India — so you won't walk away just remembering a string of English acronyms.

Step 1: First, Figure Out Where You're Going

This step gets skipped the most.

Many companies jump straight to "which model should we use" or "whose GPUs should we buy." Wrong. Ask yourself first: what problem am I actually trying to solve with generative AI?

Flipkart's answer was crystal clear — go deep. Users in Tier 2 and 3 cities couldn't make sense of English product pages. What they needed was localized content in Hindi, Telugu, and Bengali. Once that goal was set, everything that followed had direction.

HUL's answer was different. Their pain point: how to mass-produce ads with AI without losing the brand voice they'd spent 40 years building. Different goals, different paths.

So the first step is to write down one sentence: Use GenAI in X timeframe, in X scenario, to serve X audience, achieving X outcome. If you can't write this sentence, don't move forward.

Step 2: Pull the Data In

AI doesn't eat, but you've got to feed it.

In India, feeding it is especially tricky. You think there's data? There is. CRM systems hold tens of millions of purchase records. Social media generates hundreds of millions of comments a day. UPI (Unified Payments Interface) transaction volumes are astronomical. But here's the catch — this data is overwhelmingly English-centric.

A shopkeeper in Lucknow selling saris has customers who send WhatsApp messages entirely in Hindi mixed with Roman letters, without consistent punctuation. Feed this kind of "dirty data" straight into a model, and what comes out would make your hair stand on end.

ShareChat processed millions of Hindi and Tamil posts before its ad targeting became barely usable. Zomato's ability to serve up slogans like "Craving biryani in Hyderabad?" came from first cleaning regional food preference data.

Data is oil. But crude oil needs refining. This step can't be skipped.

Step 3: Clean the Material

Cleaning data, in the Indian context, means three things.

First, language normalization. Twenty-two official languages, 122 major languages — the same emotion gets expressed completely differently across states. You need translation, alignment, and sentiment analysis to smooth over pitfalls like "this phrase is a compliment in Gujarat but might be an insult in Punjab."

Second, cultural sensitivity. HUL stumbled here — AI-generated ads contained gender stereotypes that consumers caught and called out. They later added a dedicated cultural filter layer.

Third, bias correction. AI models naturally amplify biases present in their training data. In India, if caste, religious, or regional biases slip through, things can blow up in your face in minutes.

Step 4: Choose Your Model

Only at this step do you start picking weapons.

Writing text? Large language models — ChatGPT, Cohere, and the like. Generating images or video? DALL·E, Stable Diffusion, Runway. Building voice applications? India's homegrown regional-language TTS (text-to-speech) models.

But when Indian companies pick models, three hard requirements must be met.

Multilingual — at minimum, coverage of Hindi, Tamil, Bengali, and Marathi. An English-only model is basically half-crippled in India.

Cultural filtering — the ability to recognize what content is offensive in the Indian context.

Industry specialization — consumer goods, fintech, travel: each sector has its own language, compliance requirements, and consumer psychology. General-purpose models aren't enough.

Step 5: Feed It Local Grain

Once the model is chosen, you need to fine-tune it with local data.

HUL's approach is the most instructive. They organized 40 years of the company's advertising archives and fed them to the model. The ads that came out had the right brand tone — they wouldn't drift into a "tech-forward Silicon Valley aesthetic." This kind of archive training is something competitors can't copy: your history is your moat.

Reliance Jio, MakeMyTrip, and other large companies take the "global pre-training + local fine-tuning" route. It saves money and delivers results fast. Small and medium businesses don't have this option, so they're left grinding through with open-source tools.

Step 6: A Health Check Before Going Live

Model trained? Don't push it out just yet.

You need to test three things: accuracy, creativity, and ethical boundaries.

Accuracy is straightforward — if an ad claims "this phone supports 5G" and the phone doesn't, that's a hallucination. That's an incident.

Creativity is measured by engagement data. Sharing rates on Instagram Reels and genuine reactions in comment sections are more accurate than any subjective scoring.

Ethics is an especially sensitive checkpoint in India. Gender stereotyping, religious offense, deepfakes — any one of these can sink a brand overnight. In 2024, Paytm's AI ad produced a hallucination and got dragged on social media for an entire week.

There's also a very practical metric: did ROI actually go up? Flipkart boosting CTR (click-through rate) by 18% during Big Billion Days is hard evidence. AI projects without numbers backing them won't survive two quarters inside any company.

Step 7: Push It Live, Then Watch It

Deployment isn't the finish line — it's the starting line.

ICICI Bank pushed AI-driven personalized emails live and saw click-through rates rise 25%. But the real work comes after: you need to build a feedback loop. Social media sentiment — positive and negative — e-commerce reviews, customer service conversations — all of it needs to flow back into the model so it keeps learning.

Indian consumers give feedback fast and hard. A Zomato regional-language slogan goes viral, and within 24 hours the entire Hindi internet is sharing it. When something flops, it's the same — trending for all the wrong reasons within half a day. If your feedback loop runs slow, you're flying blind.

How Tightly Is This Framework Tied to India?

MARK-GEN didn't come from nowhere. It's anchored to three classic theories.

Resource-Based View (RBV) says a company's competitive advantage comes from things that are rare, valuable, and hard for others to replicate. In India, GenAI's value isn't just "auto-generating copy" — it's "regional language capability." HUL's ad model trained on 40 years of archives? Competitors who want to copy that need to go accumulate 40 years first.

Technology Acceptance Model (TAM) says whether a technology gets adopted depends on two things: is it useful, and is it easy to use. Indian marketers' judgment of "useful" is ruthlessly practical — did ROI go up? HDFC's +25% click-through rate: useful. But "easy to use" gets blocked by India's linguistic diversity. No matter how powerful the model is, if it's hard to wield, it's worthless. So this theory, applied to India, needs an added dimension: cultural usability.

Diffusion of Innovations (DOI) explains "who adopts first." India's diffusion path is clear: Reliance Jio and Flipkart lead as pioneers, HUL and MakeMyTrip follow, and 40 million SMEs are still standing at the gate watching. Not because they don't want in — but because GPU and cloud service costs are too much to bear.

An Unavoidable Knot: Personalization vs. Privacy

At this point, I have to address a paradox that gives every Indian marketer headaches.

Consumers want personalization. "This kurta pairs perfectly with last year's Diwali shopping cart" — written in Hindi, users love it.

But users also fear being spied on. Flipkart's regional-language ads are popular, but the moment an ad references specific UPI transaction records, the comment section explodes: "How do you know what I bought?" HDFC's personalized loan offers have high click-through rates, but accusations of "over-interpreting data" have never stopped.

India's Digital Personal Data Protection Act (DPDP Act) took effect in 2023, requiring explicit user consent for all data usage. This means: the more precise your personalization, the closer you are to the compliance red line.

There's only one solution: transparency. Tell users this was AI-generated. Tell users what data you used. Indian consumers' tolerance for "being deceived by AI" is razor-thin — the moment they discover an ad was entirely AI-made without disclosure, trust collapses within 24 hours.

A Few Plain Truths for India's Small and Medium Businesses

Big companies have money and people — they can run the full MARK-GEN playbook. But for 40 million SMEs, this framework needs to go on a diet.

First: don't rush to buy a big model. Hugging Face hosts a massive library of open-source GenAI tools, free or nearly free. Start there.

Second: focus on one language and one scenario. Don't try to cover all 22 languages at once. Nail the dialect your core customer base actually speaks first.

Third: ROI is the only yardstick. CTR, conversion rate, repurchase rate — if any one of them goes up, keep going. If it doesn't, stop and find out why.

Fourth: DPDP isn't a burden — it's a guardrail. Doing compliance well actually sends consumers the signal that "this shop is trustworthy."

My Own Take

The MARK-GEN framework, at its core, turns "generative AI in marketing" from mysticism into engineering. It's not perfect — 15 interviews and two focus groups aren't a large sample, and the cases are concentrated in digitally advanced cities like Delhi, Bangalore, and Mumbai. The vast rural market is largely untouched.

But its value lies in this: for the first time, someone tied India's linguistic diversity, cost constraints, and DPDP compliance together with a concrete GenAI implementation path.

The global GenAI marketing story is about "scale, personalization, efficiency." India's story is different. India's story is about "regional languages, low cost, compliance." Whoever can nail all three at once will capture the fastest-growing digital market on the planet.

My friend finally asked me: "So what should I do now?"

I said, go back and write out that one sentence from Step 1. Use GenAI in the next three months, targeting your core customer base, in the dialect they speak, to lift product page conversion rates by 10%.

Once you've written that sentence, you'll naturally know what to do next.