How We Scaled a Single Delhi Dermatology Clinic to 7 Branches Using AI-Generated Meta Ads (And The Three Things That Almost Killed The Campaign)

In November 2023, Dr. Preeti Sabarwal sent me a WhatsApp message at 11:47 PM.

"Aman, the campaign is working. We had 47 walk-ins last week. Our front desk is overwhelmed. Can you slow it down?"

I remember the message because it was the first time in three years of running ADSWORM that a client had asked me to spend less, not more. Perfect 5 Clinic had one location at the time. By December 2025, they had seven.

This is the story of how we got there. Not the LinkedIn version. The real one — including the two months in Q2 2024 when the ROAS dropped to 1.8x and I almost lost the account.

If you run a dermatology or aesthetic clinic in India and you're reading this because you're wondering whether AI-driven Meta Ads actually work for healthcare brands, here's the answer: yes, but only if you build the system right. The wrong way burns money. The right way fills your appointment book past capacity.

Where Perfect 5 Was When We Started

Single location. Zoomed-in word-of-mouth model. Strong clinical reputation in the local catchment but invisible to anyone outside a 3 km radius.

They had:

• An Instagram page with 2,400 followers, posting twice a month
• Zero Meta Pixel installed on their website
• A WordPress site that scored 38/100 on PageSpeed mobile
• One previous "social media agency" who had run boosted post campaigns and produced exactly 11 patient enquiries in 5 months at ₹65,000 in spend

The conversation with Dr. Sabarwal lasted 90 minutes on the first call. She wasn't sceptical of digital marketing. She was sceptical of agencies. Fair, given the previous experience.

I told her three things up front:

• The first 60 days will look bad on paper. The Meta pixel needs conversion data to exit the learning phase. We'd be losing money on individual campaigns while the algorithm figured out what worked.
• We'd need before-and-after content. Real patient footage. Not stock images of women with perfect skin.
• The minimum monthly ad budget had to be ₹50,000. Below that, the algorithm wouldn't get enough conversion volume to optimise. She countered with ₹35,000. We agreed on ₹40,000 with a 90-day commitment.

She signed the same day.

The Stack We Built (And Why)

Before any rupee of ad spend, we set up the infrastructure. This is the part most agencies skip because clients can't see it. It's also the part that determines whether your campaigns work or burn.

Meta Pixel + Conversions API. Pixel-only tracking is dying. iOS 14 took out about 30% of trackable events. Conversions API sends data server-side, recovers most of that loss. Took us four hours to set up via WordPress plugin + custom server script.

Treatment-specific landing pages. We built three. One for laser hair removal. One for PRP hair treatment. One for HydraFacial. Each page had: headline, before/after carousel, doctor credentials, two-question lead form (treatment of interest + when looking to book), WhatsApp click-to-chat button, Google Maps embed. That's it. No top navigation. No "About us" link. The only path forward was the lead form.

The AI creative pipeline. Here's where we differ from most healthcare agencies. We don't shoot polished studio video. We use Midjourney + Higgsfield Cinema Studio + Kling AI for cinematic ad concepts that would cost ₹3 lakhs to film traditionally. For Perfect 5's first campaign, we generated 18 ad creative variants in 48 hours. A traditional agency would have shot 3 in two weeks.

The Midjourney piece matters specifically because aesthetic clinics struggle with Meta's healthcare ad policy. Real patient photos often get flagged. AI-generated illustrations of skin conditions, treatments, and result-state imagery passes review at about 4x the rate.

This is the AI part nobody talks about. AI doesn't just speed up creative production. It speeds up the testing cycle, which speeds up the ROAS curve.

Month 1: Looked Like a Disaster

ROAS in week 1: 0.8x. Spent ₹14,000, generated ₹11,200 in attributable revenue.

Dr. Sabarwal called me on day 9. She was understandably nervous.

I sent her a screenshot of every campaign I'd seen go through this exact pattern. The Meta algorithm needs about 50 conversion events to exit the learning phase. We were at 14. Patience was the only valid move.

By day 21 we were at 2.3x. By day 35 we hit 4.1x. By day 60 we crossed 5x consistently.

The thing that turned it was creative variant 12. A 28-second AI-generated Reel showing a woman walking through a misty morning street with a voiceover about confidence after laser hair removal. Cost ₹0 to produce (the AI generated it). Generated ₹2.4 lakhs in attributed treatment revenue in its first 3 weeks.

We killed the other 17 variants and scaled that one. Within 30 days we'd produced 8 derivative versions in different settings (rooftop, café, monsoon). The Meta algorithm rewarded the variation.

This is what most agencies miss. They produce 3 ad creatives, watch them all underperform for 2 weeks, panic, and shut down. The winning ad is rarely in the first 5 you make. It's in the first 50.

The Two Months That Almost Killed It

May and June 2024. The ROAS dropped from 6.2x to 1.8x in 6 weeks.

I won't lie — I thought we'd lost the account. Dr. Sabarwal stopped replying on WhatsApp for three days.

Here's what actually happened: Meta rolled out their new Advantage+ shopping campaigns and our budget was bleeding into a campaign type we hadn't optimised. Plus, monsoon hit Delhi NCR and aesthetic clinic search behaviour shifts dramatically — people don't book PRP when they're worried about their roof leaking.

Two fixes:

• We migrated the entire campaign architecture from Standard to Advantage+ Performance and added explicit creative exclusions for the underperforming ones.
• We built a monsoon-specific creative angle: "Treat your hair fall before it gets worse this monsoon." Hair fall PRP enquiries went up 4x in the next 30 days specifically because we leaned into the season instead of fighting it.

ROAS recovered to 5.7x by end of July. We never went below 4x again.

The Branch Replication Model

By month 6, Perfect 5 had a steady 80-100 leads per month from Meta. Dr. Sabarwal's front desk was struggling to convert them all (a real problem we didn't solve until month 9, when we built her a WhatsApp auto-reply funnel — but that's another article).

She asked me a question I wasn't expecting: "If we open a second location, can you replicate this?"

This was December 2024. We opened the second location in March 2025. The campaign launched on the same day the clinic opened, with location-specific creative and zip-code-targeted audiences within a 5 km radius. We hit 4x ROAS at the new branch within 7 weeks.

Same process for branch 3 (June 2025). Branch 4 (August 2025). By December 2025 we had launched seven branches in 12 months.

The replication model works because:

• We had 18 months of conversion data on the Meta pixel teaching the algorithm what a "good Perfect 5 patient" looks like
• We could clone the campaign architecture for new locations and just swap the geo-targeting
• The creative library was already built and tested

Each new branch costs about ₹85,000 to launch on the marketing side (₹40K agency fee for the launch month, ₹40K minimum ad spend, ₹5K for landing page modification). Within 60 days each branch is profitable on the marketing investment.

The Numbers That Matter

Not the prettiest LinkedIn-friendly numbers (8x or 12x is what people brag about — we hit 12x in single-month campaigns but the sustained number is what matters for a multi-branch business).

These are the kind of numbers that let a clinic owner sleep at night because they're predictable, not the kind that look good in a case study and then collapse next month.

What I Learned About AI in Indian Healthcare Marketing

Three things that surprised me, in case you're considering this for your own clinic.

One. AI creative isn't a cost saver. It's a speed multiplier. The campaigns where we won big are the ones where we produced 30+ creative variants in a month. Traditional agencies can't do this. AI-augmented teams can. The difference shows up in ROAS by month 3.

Two. Indian patients don't trust polished. The Reels that performed best for Perfect 5 were the ones that looked slightly raw — slightly imperfect lighting, real footage of the clinic interior cut into AI-generated sequences, doctor's voice (not voiceover artist). When we tried fully polished AI cinematic content, CTR dropped 38%.

Three. Compliance matters more than people warn you about. We've had three creatives pulled by Meta's healthcare ad policy reviewer in the last 18 months. Two were before-and-after photos that crossed Meta's "exaggerated results" line. One was an AI-generated image of a hyper-real skin tone change. We learned to use AI imagery for aspiration (a confident woman walking, sun on her face) rather than transformation (visible skin changes). Aspiration always passes. Transformation gets flagged.

Should You Try This For Your Clinic?

Here's the honest truth.

If your clinic has fewer than 50 monthly walk-ins right now, fix the front-of-funnel first. Get your Google Business Profile sorted. Get your reviews up. Make sure your existing patients can find you. AI-driven Meta Ads will accelerate a clinic that already has product-market fit. They won't fix one that doesn't.

If your clinic has good clinical reputation locally but you're invisible online, this approach can transform your business in 90 days. Not because of magic, but because the funnel works when you build all four pieces (pixel, landing pages, AI creative, treatment-specific campaigns).

If you're spending less than ₹40,000 per month on Meta Ads, the algorithm can't get enough data to optimise effectively and you'll see the kind of disappointing results that previous agency burned for Perfect 5. Either commit to the budget or don't run paid campaigns at all. Half-measures cost more than zero-measures.

Either way, hope this was useful. Most agency content is so polished it's useless. I tried to write the version I would have wanted to read three years ago when I was figuring this out.

— Aman