Case Study
150M views
on ₹5L
I head the content team at FloBiz, the company behind myBillBook, a GST billing app for Indian small businesses. In September 2026 our content did 150M views on ₹5L of spend. That works out to about ₹3.3 per 1,000 views.
Views · September 2026
0M
Trajectory
10M to 150M a month
Monthly reach went from about 10M to about 50M views over the first half of 2026. In September it hit 150M.
Monthly views
The system
Four engines
Creator reels
We collaborate with creators who already speak to small-business owners. 300+ creator reels went out between Oct 2025 and May 2026.
In-house channels
Accounts the team owns and runs, where reposts and original pieces go out.
Reposts
Strong reels get a second life as branded, credited reposts, built through the Repost Portal.
AI-made content
Content made with AI tools, run as its own engine next to the other three.
How it moves
From one reel to a branded search
Two tools I built sit in the middle of this loop. The Repost Portal makes the videos. The view tracker tells me what they did.
Tool 1
Repost Portal
Paste a reel, pick a layout, and get back a branded, credited repost. About 30 seconds of human time and roughly 5 minutes of machine time. 7 layouts, 19 add-ons (caption bars, story chips, zoom, music and SFX, logos), and 550+ videos rendered since May 2026.
Fetch
yt-dlp pulls the source reel.
Cut
ffmpeg trims, crops and masks.
Compose
An HTML template system (HyperFrames) builds the overlay: layout, caption bar, chips, logo.
Render
A render worker renders the final video and uploads it.
Principles
How we repost
The creator's work stays intact
Their audio, on-screen text and closing shot are never touched.
Credit lives in the caption
Every repost names the original creator in the caption.
About 10 reposts per account per 30 days
Instagram cuts reach for accounts that mostly post unoriginal content. The cap keeps every account original-first.
The value-add is real
Each repost carries genuine creative input: a layout, captions, context, sound.
Tool 2
View tracker
Reads three platforms daily
Instagram, YouTube and Facebook view counts, every day, including Hindi “लाख views” text.
Cohorts by posting month
Each month's views split by the month the reel went out, so I can see old reels still compounding next to new ones.
Heals its own login
When the Instagram session expires, it logs back in by itself.
Did it move anything?
Branded search, about 22% above trend
1.65%
monthly growth before
2.91%
monthly growth after
After the creator programme started in September 2025, branded Google searches for the brand grew about 22% above the counterfactual trend. The monthly growth rate went from 1.65% to 2.91%. This is an observational read, not a controlled experiment.
Numbers
What it delivers
150M
Views
September 2026
₹5L
Spend
September 2026
₹3.3
Cost
per 1,000 views
300+
Creator reels
Oct 2025 – May 2026
550+
Videos rendered
Repost Portal, since May 2026
~30s
Human time
per repost
Decisions
Why I built it this way
Why reposts at all?
A good reel already did the hard part. Giving it a branded, credited second run costs about 30 seconds of someone's time and five minutes of a machine's. That math lets a small team publish far more than it could shoot.
Why cap each account at about 10 reposts a month?
Instagram cuts reach for accounts that mostly post unoriginal content. A hard cap keeps every account original-first, so the reposts add reach on top instead of eating into it.
Why lock the creator's audio, text and closing shot?
That is their work and their voice. My layer frames it: layouts, caption bars, story chips, zoom, sound. Credit goes in the caption every time.
Why render on a Mac with a launchd worker?
Rendering is ffmpeg plus a browser composing HTML overlays. A machine I already run handles it, and a self-healing requeue picks up any job that dies mid-render.
Why build my own view tracker?
Views sit across three platforms, sometimes as Hindi text. I needed one daily number per reel and a cohort view by posting month, so I could tell whether growth came from new reels or old ones still running.
Why measure branded search?
Views say people watched. Branded Google searches say people went looking for the brand afterwards. That is the signal I tie the content programme to.
Stack
What powers it
Next.js
Portal app (Vercel)
Supabase
Jobs, users, counts
launchd worker
Mac render queue
HyperFrames
HTML overlays
ffmpeg
Trim, crop, mask
yt-dlp
Fetch + view counts
FastAPI
View tracker
Razorpay
Credits + team tier
Want reach like this?
I build content systems and the tools that run them. Let's talk about yours.
© 2026 Sarthak Goel