@Quickads_ai & @88ventures | Engineer by training, a 'hacker' at heart, and a recovering CxO
Joined April 2020
- Tweets3.4K
- Following1.2K
- Followers417
- Likes1.6K
Sane advice especially for post AGI works where building is easy, moats disappearing
This is why I recommend younger founders to build profitable, bootstrapped business, make money, and then think of building a moonshot, venture-scale business.
I did exactly this myself. Built my first venture, sold it, made good money and became financially free. Now I'm building a moonshot at @GreyLabsAI taking a big goal of disrupting the call center industry using AI.
Worst case if it doesn't work, I won't find myself on the road. Best case if it works, it's going to be a great outcome for all.
Growth_Hacker retweeted
"Certain class of founders"?
Such a weird flex - reeks of misplaced derision about "incremental tech" bets.
It was the success of these v1 incremental bets - Flipkart, Myntra, Zomato, Freshworks - that spawned founders whose angel investments catalysed pretty much every deep tech category in India - EV, robotics, space tech, wearables. Categories that VCs themselves never found worthy of their time and money until they broke out. Neither these v1 incremental bets nor the following deep tech bets were sure things when they started out - yet who can argue that they have not been transformational?
By your own framing, Oyo also belongs squarely in the “incremental” bucket. Lightspeed happily banked the founder’s debt-financed secondary - your largest payout from India? So what if it was unearned and contrived for your convenience?
Your frontier maths hardly supports the superiority either. A 1% shot at 100x, with zero otherwise, is 1x expected gross MOIC before fees.
“Minuscule” odds whose returns can therefore fall far short of beating the supposedly “guaranteed 2–3x” you so casually dismiss!
Finally, “guaranteed 2–3x” on incremental bets is a remarkable claim from someone in the risk business.
Sachin Bansal's quip comes to mind...
Look, India has two kinds of VCs right now. One set continues to invest in things that are 'here and now", the ice cream brands, the new age high end goods, and frankly a variety of local maxima incremental tech including a lot of stuff in AI.
The reality is those are going to make some money, no doubt. Per capita disposable income in increasing, some of the local maxima tech cos will be m&a fodder, etc, etc etc. I don't have anything against these VCs, and 100% believe founders should chase whatever dream they are energized by and can find capital partners for (VCs are but just one source). A lot of this incremental stuff is guaranteed 2-3x MOIC. But the future of India will be built on stuff that's transformational.
The market there is in bringing the frontier to the present and chances of success are miniscule. Everyone investing there, everyone starting a company there is very likely to get <1x of the capital back. But that's where the >100x returns are. I wake up thinking most likely things are NOT going to work, most likely we will all fail, but WHAT IF WE ARE RIGHT? WHAT IF WE WIN?
I think these sorts of probabilities attract a certain class of founders and it needs a certain class of investors that need to co-exist. These founders have global ambition and are global maximas. The investors need to match that both in their own understanding of what global maximas are, and how to amplify this ambition through capital, network, experience. That's the business partnership that India needs.
Growth_Hacker retweeted
Beyond Amitabh Bachchan's bungalow, there are only 2–3 industrialists' bungalows on that road.
There is one grand bungalow on that road. It is called "Niranter."
This bungalow is three times larger than Amitabh Bachchan's. It has a lawn of nearly 3 acres and is extremely opulent.
Anyone would be stunned to hear that such a lavish bungalow exists in an elite area like Juhu, Mumbai!
But this bungalow doesn't belong to a superstar or an industrialist.
Do you know who owns it?
**Teesta Javed Setalvad** – just a "social activist"!
Now read on patiently to the end – your eyes, ears and mind will all begin to open.
Between 2004 and 2012, she received crores of dollars in funding from abroad.
For what? To uplift the poor?
But there is something more...
Why are these people so anti-India?
Their forefathers sowed this seed.
👇Start here. Read the history Congress hid from us.
The Hunter Commission was the inquiry committee that gave General Dyer a clean chit – the same Dyer who ordered the firing at the Jallianwala Bagh massacre.
A member of this commission was Harilal Chimanlal Setalvad, great-grandfather of Teesta Setalvad.
In other words, the ancestor of the present owners of the bungalow described above.
It was Harilal's son, Motilal Chimanlal Setalvad – Teesta's grandfather – who acquitted General Dyer.
After independence, Nehru appointed this same Motilal Setalvad as India's Attorney General.
This is living proof of Nehru's loyalty to the British.
The story doesn't end here...
**Read on.**
While General Dyer was being tried, Diwan Bahadur Kunj Behari Thapar, showing his loyalty to the British, collected Rs. 1.5 lakh for General Dyer and honored him by presenting him with a kirpan and a turban.
This same Kunj Behari Thapar is the great-grandfather of **Karan Thapar**.
The Thapar family grew rich during British rule, especially by supplying troops and materials during the First World War.
Today, when the Thapar family displays its loyalty, it should be no surprise where the kirpan and turban came from.
All this happened in the management of the Golden Temple, where two contractors, **Sujan Singh** and **Sobha Singh**, were the main contractors.
When the British shifted India's capital from Calcutta to Delhi, these two took over all the construction contracts.
Sobha Singh's son **Khushwant Singh** was a famous writer and a supporter of Indira Gandhi. He wrote articles in favor of the Emergency.
Khushwant Singh's son, **Rahul Singh**, keeps promoting anti-India views on NDTV by glorifying figures like Teesta Setalvad and Arundhati Roy.
Speaking of the Thapar family...
Karan Thapar's father, **Pran Nath Thapar**, was the commander in the 1962 China war – a war India lost.
Earlier, General K.S. Thimayya had recommended Lt. Gen. S.P.P. Thorat as his successor.
But Nehru rejected that recommendation and appointed Pran Nath Thapar.
– Another example of the British mindset.
Not only that – Pran Nath Thapar's brother is **Mayadas Thapar**, and his daughter is **Romila Thapar**, after whom Indian school history textbooks are named.
Nehru made this appointment too.
Strangely, the names of those who lost the 1962 war are not found in these history books.
Just because these books were written by their nieces!
This is the same wealthy family – which grew rich through bribes during the British era – that still holds India's history and identity hostage today.
These are the same people who regard themselves as the monopolists of progress.
These are the same people who spread lies and stop India's true identity from emerging.
After independence, Nehru and Congress protected British interests by giving these very people important posts.
Teesta Setalvad went to jail – this was not just the fall of one fraudster... but the fall of an entire parasitic system, whose roots were planted generations ago.
So true.
A company that was refused what 80 times by sand hill and went on to defy exponential growth.
Hats off to founders and early team
Banger
How a static ad hits $120k+ in spend:
→ Call out your specific audience [people on Mounjaro]
→ Open a curiosity loop [this paragraph is an eye-opener]
→ Validate the result they want [weight loss]
→ Introduce the overlooked problem [loose skin after weight loss]
→ Challenge the obvious solution [another skin cream]
→ Reveal a new mechanism [Type I & III collagen]
→ Make the solution stupidly simple [2 tablespoons in coffee]
That’s how you make a random book image print $2k/day lol.
Cost per lead says ₹93 beat ₹123. Sales might have an objection.
Our published QuickAds campaign report gives those two rounded CPLs. It doesn't publish their qualification rates. So the dashboard supplies a starting point, not the verdict on which deserves the next rupee.
Here's the number that would change the decision.
The ₹123 creative needs roughly 123 ÷ 93 = 1.323 times the other's qualification rate to tie on cost per qualified lead.
About 32% higher. Not 32 percentage points.
Take an explicitly hypothetical case. If 20% of the ₹93 leads qualify, each qualified lead costs ₹465.
At the same 20% qualification rate, the ₹123 leads cost ₹615 per qualified lead. Clearly worse on this measure.
But if that second creative qualifies at 30%, its qualified-lead cost becomes ₹410. Now the supposedly expensive lead is cheaper.
The break-even is about 26.5%. Above it, the ₹123 creative wins this particular comparison. Below it, it loses. Those qualification rates are examples, not results from our report.
This is why “but the leads are better” needs a number, and “the CPL is higher” isn't enough to dismiss it.
Reusable rule:
Required qualification rate for B = A's qualification rate × (B's CPL ÷ A's CPL).
Use your exact unrounded inputs for a live decision. These rounded published figures make the threshold approximate.
Then be annoyingly strict about the denominator. Same qualification definition. Same observation window. Comparable follow-up. A lead sales called within an hour should not be compared casually with one left untouched for a week.
And stop at the boundary of the maths. Qualified leads still aren't customers. If close rate, order value or margin differs, the revenue decision needs those numbers too.
There is even a useful limit: if the required qualification rate goes above 100%, the higher-CPL option can't catch up on this metric, however enthusiastic the campaign review becomes.
Our report cannot tell us which creative cleared the quality hurdle, because it doesn't contain that column. That's the point: ask for the missing evidence before moving the budget.
The case numbers:
quickads.ai/creative-intelli…
Answer engine optimization gets awkward when the invoice starts with a special file Google says you don't need.
Google's own guidance for AI Overviews and AI Mode says there are no additional technical requirements beyond normal Search eligibility. A supporting page must be indexed and eligible to appear with a snippet. There is no special AI text file or schema requirement.
That is a specific statement about Google's Search features. It is not a claim about every assistant on the internet.
But it should change the sales conversation.
If someone is charging you for “AI readiness,” ask what will actually be different on the site when they finish.
A file can be a deliverable. Its existence doesn't establish a useful outcome.
Suppose an illustrative merchant has a sizing chart only inside an image, an old product page linked in the navigation and conflicting dimensions between the visible description and structured data.
You could install something with an impressive new acronym. The buyer would still be trying to work out whether the thing fits.
The useful project would name the broken page, make the important details available as text, correct the conflict and verify the intended page can be found. Those are concrete changes someone can inspect.
Google's guidance also says eligibility doesn't guarantee crawling, indexing or appearance. So even a technically clean project needs a narrower promise than “we will get you cited.”
My preferred acceptance test for a proposal: show the current problem, the exact change, how you'll verify it and what the check cannot establish.
For example, confirming that a page is indexed proves that step. It doesn't prove the page will be selected as a supporting link for the next AI answer. Seeing one citation proves an observation at that time, not a permanent placement.
Keep those results in separate columns. Otherwise, a vendor can deliver a setup task while everyone mentally books a sales result.
There is useful work to do around evidence, discoverability and accurate product information. It deserves a budget when the problem is real.
A fashionable filename shouldn't be the thing that makes the proposal sound technical.
Google's actual documentation:
developers.google.com/search…
Absolutely
Did a price comparison and not sure why anyone is paying 400% more!!
quickads.ai/quickads-vs-higg…
AI shopping and search appear to be getting along better than the people arguing about them.
Similarweb's September ecommerce report says that when consumers used AI in their shopping research, they also used search 89% of the time.
Read the denominator before announcing a winner. That's co-use within AI shopping research. It does not say 89% of all shoppers used AI.
The same release says direct AI referrals rose more than 200% over the preceding year, but remained small compared with search traffic to ecommerce sites.
So two things can be true: more people are using AI, and search can still matter to the purchase. A growth rate cannot tell you the size of a channel without its starting point.
Here's the practical mess hiding behind the channel-war headlines.
Imagine a buyer asks an assistant to compare two backpacks for a rainy commute. They get a shortlist, search a brand name, inspect the size chart and buy through the site.
That's an illustration, not a tracked customer in Similarweb's report. But it shows why a last-click dashboard may answer a different question from “what helped this person decide?”
The assistant comparison and the later search can do different jobs. The sale doesn't have to declare loyalty to one of them.
For the merchant, consistency gets more important. If the product page, shopping feed and comparison page disagree about the material or warranty, the buyer carries that disagreement between tools. Another SEO-versus-AI argument won't fix it.
I'd take the five questions that keep appearing before purchase and follow each one across the actual evidence the buyer can see. Where does the answer change? Where does a confident claim lose its source? Where does the right variant become unclear?
Fix those handoffs before constructing a precise budget model from a tiny referral sample.
You can track assistant mentions, visits and completed orders separately. Record what the analytics cannot connect. A purchase survey may add context, but a remembered answer is still a remembered answer, not perfect attribution.
And if journeys using both tools show better conversion, don't leap straight to “AI caused it.” People doing more research may already differ in intent. Similarweb's metrics are estimates, not a randomized trial.
The buyer is trying to buy something. Our organisational chart and our favourite channel are probably not their main concerns.
Source:
ir.similarweb.com/news-event…
Agentic commerce has a more interesting story than “every store needs an AI shopping button.”
A screenless tin-can phone. Slow-grazing horse hay nets. Carefully machined metal pill cases.
Those are the kinds of products Shopify puts in its May research on specialist businesses. Much more useful than another generic slide about the future of shopping.
The number: 71% of Shopify's 2025 AI-attributed orders came from categories outside its top 100 by GMV. AI attribution means an AI-powered channel appeared in the buyer's discovery path.
It does not mean 71% of all ecommerce used AI. Or that 71% of Shopify orders did. Or that AI caused those purchases.
It also isn't research about long-tail keywords. The tail here is product categories. Small distinction, until someone builds the entire strategy on the wrong noun.
What interests me is the product question.
“Buy a pill case” leaves a lot of products in contention. Shopify's account of Ikigai Cases describes a different starting problem: two brothers whose father struggled to open his pill box. That gives the object a more specific reason to exist.
My inference: a specialist brand should be very careful before smoothing all that specificity into “premium accessories for modern living.” You may have just deleted the information a particular buyer needed.
For an illustrative product-page review, imagine someone needs a case they can open easily. A glossy photograph of a closed box cannot settle that. Show the opening mechanism. State what was actually tested. Avoid inventing an accessibility promise the product team can't support.
The same goes for the ad. A hand opening the actual case could answer a useful question. Another slow pan over a shiny object may be prettier while saying less.
This is where the research changes a brief, rather than merely decorating a prediction: choose the specific buying constraint and show what the product does about it.
Don't turn the 71% into a budget allocation. We don't have your store's AI-referral volume, margin or incremental conversion from that figure.
We do have a good reason to stop apologising for a product being specific.
The niche may be the answer. Make sure the page actually says what the question was.
Source, including Shopify's definitions:
shopify.com/news/entrepreneu…
Creative testing gets pretty shallow when the whole brief is “let’s try static vs UGC.”
We inspected GIVA’s India ads in Meta’s public library on October 20. Four examples across GIVA and Heer by GIVA made the problem obvious: the useful differences were the reasons someone might buy. The file type was the least interesting part.
Start with the maroon gift boxes. GIVA’s October 12 ad puts a large 15% discount beside an open jewellery box, a bracelet and earrings. The offer requires two pieces and a first order; the copy leaves room for buying for yourself or somebody else. That matters. The packaging makes the gift feel ready, while the quantity condition gives someone a reason to add another piece. Whether that actually lifts basket size needs account data. The creative only tells us what the offer is trying to do.
Now look at the October 19 ad. The captured frame shows a silver ring with two green stones, isolated on white. No person. No cinematic life milestone. The copy addresses an ambitious woman and invites her to celebrate her own journey; the link headline carries a 20% discount code. That is a self-reward proposition using an ordinary product photo. You do not need a creator holding a phone to test self-purchase.
The October 12 gold ad is quieter again. The captured version shows one gold ring against white. The copy starts with everyday styling, moves into gifting, then adds material claims, hallmark language, returns and buyback terms. I would call that a purchase-reassurance hypothesis. It anticipates the questions someone might have before buying jewellery online. A pretty product image and product proof are different jobs, even when they appear in the same ad. Those are the advertiser’s claims, not independently verified product findings.
Then there’s the October 18 Heer by GIVA partnership ad, co-labelled with she_is_samrageepalit. Calling this “UGC” would hide most of what makes it different. The 54-second video shows a couple in a styled kitchen, a gift-box reveal, jewellery worn on the body and a Karwa Chauth end card. This is a produced relationship story, attached to an occasion. The partner label is real. The assumption that every creator ad must look like a casual testimonial is not.
What would I actually test?
Pick one product and one offer. Write three briefs: reward yourself, give something meaningful, reduce purchase uncertainty. Keep the landing page consistent. Then choose the format that makes each reason believable. A creator can demonstrate fit. A still can make the design legible. A story can make the occasion feel specific.
Otherwise, the “UGC winner” might just be the only ad that gave someone a relevant reason to care.
QuickAds’ jewellery swipe file is useful as an intent map: it describes 56 patterns grouped into seven buyer intents. I’d use that taxonomy to generate the briefs, then judge them with actual account results; not treat a swipe-file example as a proven winner.
All four ads were marked active when checked on October 20. That is a receipt, not a ROAS report. The static ads also have multiple versions; the screenshots show the specific frames inspected.
Source ads:
facebook.com/ads/library/?ac…
facebook.com/ads/library/?ac…
facebook.com/ads/library/?ac…
facebook.com/ads/library/?ac…
Intent taxonomy:
quickads.ai/jewelry-ad-hooks…
ALT Actual GIVA Meta ad creative with maroon gift packaging and a first-order, two-item offer.
ALT The specific green-stone ring creative displayed in GIVA's multiple-version self-reward ad.
Viral marketing: can we talk about the bit of Duolingo’s dead-owl campaign that actually asked people to use the product?
“Be more unhinged” is an expensive lesson to take from it.
Look at the original February 11 post. Five words in the caption. More than 100 in the attached obituary.
Please, no “short posts win” takeaway. Most of the writing is sitting in the image.
That image gives people several ways into the joke. There’s a death to investigate. There’s the suggestion that your neglected lesson might be responsible. There’s a Duolingo Max joke, followed by a request to respect Dua Lipa’s privacy.
It reads like an official announcement from a company that has completely lost the plot. But the lesson is already inside the story.
Then look at February 12.
Duolingo posts two app-style screens. One is a lesson result with a dead owl and a button to bring him back. The other has daily quests, including “Learn how Duo died” and “Extend your streak as tribute.”
Those are the campaign materials they posted. I’m not claiming every user actually received those screens.
The useful detail is how they changed the meaning of the task. A daily lesson could now count as investigating a death or paying tribute. The familiar product behaviour had a role in the joke.
By February 17, the invitation gets more concrete. Duolingo links to a rescue page. The attached GIF shows a 50-billion-XP target, a progress bar and a country leaderboard. Its instruction is to do a lesson.
The audience can contribute beyond another funny reply. Progress in the product can move a shared outcome in the story.
That’s the part I’d put in a campaign brief.
A lot of brand campaigns give you a joke to watch, then bolt a completely different sales request onto the end. Here, the thing Duolingo wanted people to do was presented as the way to help resolve what they were watching.
There’s a commercial receipt, with a limit worth keeping attached.
In its Q1 2025 shareholder letter, Duolingo reports 1.7 billion organic campaign impressions and a meaningful lift in new and reactivated users. It doesn’t give us a campaign-specific lift number or enough detail to independently establish causality.
The same quarter’s 49% year-over-year DAU growth is not a number we can simply credit to a dead owl. Other product improvements were happening too.
And 50 billion XP is a campaign target, not proof of 50 billion additional XP caused by it.
Before copying this, I’d ask a less glamorous question than “what’s our outrageous stunt?”
What can somebody actually do inside our product that matters to the story?
Write that answer before writing the announcement. Then decide how you’ll separate ordinary usage from the extra activity the campaign produces.
If the answer is still just “share the post,” there’s a gap in the plan.
Duolingo gave people a reason to reopen an app they knew. Killing the mascot was the invitation. Doing the lesson was their part.
The actual follow-ups:
nitter.cf/duolingo/status/188978…
nitter.cf/duolingo/status/189161…
Company results:
duolingo.gcs-web.com/static-…
ALT Duolingo's actual February 12 campaign post showing a lesson result and death-themed daily quests. These are published campaign materials, not a claim every user received the screens.
We have discovered a new lead in unearthing the mystery of Duo's death. Please contribute where you see fit: bringback.duolingo.com/
Your UGC ads need a reason to exist beyond "make it authentic." Look at what Mielle does with "Be ready in 5!"
Three Mielle partnership ads in Meta’s Ad Library make the distinction clear. Same brand. Different reasons to care before anyone starts explaining the bottle.
The useful creative choice: give the routine a job in the story.
1. Chelsea as of Late: the five-minute deadline.
The opening is a phone message: “Be ready in 5!” Chelsea looks up, startled. We move from a seated, fully dressed creator to a close-up of edge gel, brushing along the hairline, then oil applied with a dropper between braids.
At roughly ten seconds, she answers “Coming down!” Then we get walking shots and different outfits outside.
The product demo occupies the gap between an invitation and leaving the house. That makes the application shots part of the action rather than a commercial break. The finished-hair footage supplies a visible styling payoff. It cannot demonstrate long-term strengthening.
The card says Amazon availability and uses Shop now. Nineteen seconds. One tiny plot.
2. Carolyn “Maria” Gray: protect the style through the week.
She opens by touching and showing her silk press. The captions introduce the question of maintaining it. Product bottles appear early, followed by lathering, washing, towel-dried hair, an oil dropper and application.
Then the setting changes: getting out of a car, browsing books, posing in an outfit, a museum and evening scenes.
That sequence makes a three-product routine legible: what goes in, where it goes, and the kind of week it is meant to fit. The caption identifies shampoo, conditioner and scalp-and-hair oil, with a Walmart shopping instruction.
There is a useful distinction here. The montage illustrates her style and schedule. It does not establish how many extra days the products made the style last. Borrow the sequencing. Substantiate your own promises.
3. Jenna Evans: an awkward conversation earns the bathroom scene.
This one takes about a minute. Large yellow type, a handheld microphone and an embarrassed reaction open a story about someone commenting on her hair after a workout.
The bathroom becomes the next scene for a reason. We see product dispensed into a palm and worked through hair, followed by outfit choices, styled braids, a date setting and a later scalp-oiling close-up.
The creative moves from embarrassment to doing something about it to going out. Product use is observable; any strengthening effect remains a claim. The ad card closes with Shop now.
Here is the brief I would take from these, rather than ordering three more interchangeable testimonials:
Pick one interruption. A last-minute invitation, maintaining a finished style, or getting ready after exercise. Give the creator one visible application action that belongs in that situation. Finish with the original plan happening.
For a clean test, keep the creator, product, offer and landing page fixed. Change the opening situation. Read the downstream purchase results alongside watch and click metrics. A funny story that attracts the wrong clicks is still an expensive story.
These ads were active when checked on 20 October. Meta’s public cards do not reveal their ROAS.
Source: Mielle partnership ads in Meta Ad Library. Chelsea: facebook.com/ads/library/?ac…
Carolyn: facebook.com/ads/library/?ac…
Jenna: facebook.com/ads/library/?ac…
ALT Actual frames from Mielle's Chelsea partnership ad: a message about being ready in five minutes, product application and departure.
ALT Actual frames from Mielle's Carolyn partnership ad showing the hair routine and lifestyle scenes, not proof of product efficacy.
Fashion Nova is running the same photo in two different Meta ads.
Same woman. Same brown top. Same black jeans. Same handbag.
Different reason to buy.
Meanwhile, somewhere, a creative meeting has decided the problem is “we need another shoot.”
These six examples come from Fashion Nova's actual Meta Ad Library, with screenshots, IDs, copy and video frames.
Here's what was actually there.
1. The photograph stays. The sales pitch changes.
One ad opens with “Your next Fashion Nova haul is calling” and gives you GET20.
Another opens with “Up to 80% off sitewide” and warns that the best pieces could disappear.
Both show the same outdoor outfit photograph. Both name the same black skinny jeans beneath it. Both say Shop Now. Both were marked active, with September 3 start dates.
One frames the purchase as another haul. The other frames it as a deal worth catching.
That's a useful creative-testing distinction: changing the buying reason is not the same thing as changing the footage.
No, I can't tell you which won. The library doesn't show that. Anyone turning these screenshots into a ROAS victory lap is writing fan fiction.
2. The denim video answers a different question.
A 24-second mirror video shows pale-blue jeans from the front, then closer at the waist and legs, then from the side, then moving toward and away from the mirror.
Black top. Bedroom. Phone in hand. No elaborate production.
The surrounding copy promises stretch and a flattering fit. The video at least lets you inspect shape and movement. It doesn't just repeat “flattering” in a nicer font.
If fit is the hesitation, another cinematic opening isn't automatically the answer. Showing the bloody jeans might be.
3. The sandal ad splits the work.
In a 10-second clip, a hand rotates a sculptural wedge sandal. Then we see it on feet, from different angles.
The on-screen line is “it's a wedge type of summer.”
Outside the video? Closet indecision, FNFAST and free one-day shipping.
The footage shows what makes the shoe interesting. The copy gives a reason to order now. Those are separate jobs - not a request to cram seven selling points onto the screen.
4. The Halloween setup gets a second job.
One five-second costume video has a woman in red and gold moving toward the camera under “when the bob is bobbing.” Shop Now leads to the Halloween collection.
Another uses the same costume-and-room setup, but changes the treatment: white branded frame, “POV: Halloween fit secured early,” different inner caption.
The CTA is now Install now. The destination is the Fashion Nova app listing.
Same recognizable setup. Different edit, framing and requested action.
“Make more content” is a wonderfully convenient brief. Nobody has to decide what the existing content is supposed to do.
A more useful brief names the variable:
The buying reason. The product question. The offer. The destination.
Then decides whether new footage is actually required.
Our broader Fashion Nova catalogue teardown explores the product-asset infrastructure behind that way of thinking:
quickads.ai/fashion-nova-cas…
These six ads are observations, not proof of performance or their entire strategy. Checked October 20, US active-ad view.
But they are considerably more useful than calling every new hook a “game-changing creative.”
Actual photo pair: facebook.com/ads/library/?ac…
facebook.com/ads/library/?ac…
ALT Original Fashion Nova Meta Ad Library card offering GET20 with an outdoor outfit photograph and named black jeans.
ALT Original Fashion Nova Meta Ad Library card using the same outfit photo with an up-to-80-percent-off message.
Bursting hype vs. reality on the AI ads (real cost, issues and time to ship a usable on-brand version)
Your UGC ads don’t need another “I’m obsessed” monologue. Look what RYZE put on a dining table.
A pile of products. Handwritten price cards. Someone behind the camera pointing at the stuff. A creator answering with her hands. Neon-green highlights. A giant $0.
It looks more like a slightly chaotic shop counter than a skincare confessional.
That’s the interesting bit.
Look at these four actual RYZE creatives in Meta’s Ad Library. The 33-second tabletop video is worth studying because the creator has a job beyond looking relatable: direct your attention through the offer.
You see the basket. You see which object the pointing finger means. You see the green emphasis move the conversation along.
The face isn’t carrying the entire sale. The table is doing some actual work.
Compare that with RYZE’s lilac gift ad. A hand holds the coffee pouch on the left. The right side itemizes the jar, magnet, creamer, utensils and shipping. The biggest message is the claimed $100 gift value.
Same broad buyer question: what am I getting?
Different answer. One is a visual inventory. The other turns the inventory into a little interaction.
Calling the video “UGC” and the image “static” tells you how they were packaged. It barely tells you what to brief next.
Then there’s the Halloween ad.
Huge apology opening. Bats. Orange background. The copy tells recent buyers they missed out, then tells the people who waited that a flash sale has arrived.
That’s a very specific little emotional switch: hesitation becomes good timing. You weren’t indecisive. You were apparently early to the better deal.
Clever framing. But look below the cleverness.
The artwork says 40% off. The native ad’s destination-description says 25% off.
I haven’t checked checkout, so I’m not calling the offer fake. There may be different conditions. I am saying the displayed messages make the buyer reconcile two discounts. “More hooks” is a weird priority while that’s sitting there.
The fourth creative pushes a belly-related outcome, then puts $27 beside a crossed-out $45. Two wooden utensils are tied with red ribbons.
Now the order of persuasion has changed: a physical concern first, price second, gifts as support.
But the image supplies no evidence for the physical outcome. Copying the promise because the ad is active would be a spectacularly lazy way to write a brief.
These ads were active when I looked. That is not a ROAS report.
What I’d actually borrow is the tabletop mechanism. Give the creator the exact basket, the actual offer conditions and one buyer question to answer. Compare that walkthrough with an itemized static using the same terms.
Then you can learn whether the human interaction helps explain the purchase.
Another person holding a pouch and being obsessed mostly teaches you that you can hire another person to hold a pouch.
Source examples: facebook.com/ads/library/?ac…
facebook.com/ads/library/?ac…
facebook.com/ads/library/?ac…
facebook.com/ads/library/?ac…
Observed October 20, 2026. Visual analysis; no inference about measured sales.
ALT RYZE's original gift-bundle ad, reviewed as an example of itemized offer communication.
ALT A frame from RYZE's actual tabletop creator ad, showing products, handwritten prices and a pointing hand.
ALT Original Meta Ad Library screenshot of RYZE's Halloween offer. Visible artwork and destination description display different discount figures; checkout conditions were not verified.
Can we talk about the real AI video cost?
Everyone shows you the clip that worked. Show me what it cost to get there.
On October 13, Brian Ng said his team was spending over $1,000 a day on Higgsfield for AI drama and song ads. His screenshot shows six entries totaling $8,720, including one pending $1,600 payment.
We don't know how many usable ads that bought. That's exactly the missing number. Brian's post: nitter.cf/briannjho/status/21101…
"Price per second × seconds in my final ad" is a terrible production budget.
Even before you throw anything away, the settings matter. Higgsfield's current table lists standard Seedance 2.0 at about 22 credits per five seconds in 720p, versus 60 in 1080p. Its monthly Plus plan is $59 for 1,200 credits.
That puts a 15-second generation at roughly $3.25 versus $8.85. Same model. Same length. Different resolution. Those are estimates using the plan's credit value, before tax, assuming you use the credits. They are not finished-ad quotes. Official pricing: higgsfield.ai/pricing
And the developer API has a different meter. For Seedance 2.0 reference-to-video, it bills the reference video's duration as well as the generated duration. It also charges a 40% lower token rate when a video reference is supplied. Image and audio references don't count as video input.
So "they charge for the input!" without mentioning the discount is only half the story. Read the actual billing rule for the route you're using. API terms: open.higgsfield.ai/models/by…
Then comes the work the render price doesn't describe. The take where the bottle changes shape. The version you reject after watching it. Fixing the voice. Cutting the shots together. Getting someone to approve it.
I'm not going to invent a universal retry rate or pretend those hours are free. Your generation history and your editor's time are the receipts that settle this.
The human UGC comparison needs the same honesty.
Billo currently starts at $99 for a basic creator video. That includes creator pay, two revisions within the original brief and usage rights. Product and shipping are extra. The basic footage excludes captions, music and graphics; editing and creator posting are separate add-ons. Billo's October pricing explanation: billo.app/blog/how-billo-pri…
Comparing that with one AI render and doing a victory lap about AI savings tells me you haven't matched what you're buying. Paying somebody to film your product also doesn't automatically buy their audience.
What would I change first? Get the script and product reference approved before generating variations. Test motion in a shorter, lower-resolution shot where the job allows it. Check whether the mistake needs a new generation or an ordinary edit. And stop sending a whole reference video when the useful bit is a few seconds.
At QuickAds, our published generation rate for 15-second Seedance 2.0 at 1080p is $1.95. I still wouldn't call that the cost of your finished ad. Plan costs, rejected takes and the remaining production work belong in the comparison too. Our rate comparison: quickads.ai/quickads-vs-higg…
Show me the approved ad, everything spent making it and what still needs paying for.
Then we can have a useful conversation about cheap.
Prices checked October 20, 2026.
An AI ad can sell the fantasy. Your returns team gets the reality.
Before scaling the prettiest creative, check margin after refunds and fulfilment. Compare repeat orders at the same time since purchase.
A beautiful demo is a terrible substitute for that second column.
ALT Proposed next-brief decision framework: adjust sales for discounts and refunds; subtract fulfilment and other variable order costs; check traffic-source spend and attribution scope; compare repeat orders at equal cohort maturity. Not an exhaustive accounting model.