AI Pet Portrait: GTM for a Niche AI Product
Context
AI Pet Portrait launched in October 2023 as a deliberate niche experiment. The product used generative AI to transform user-uploaded pet photos into styled portrait series: Pawtrait (classic portrait), Master Chef (costume series), Wizard (fantasy), and Birthday Party (celebration). The visual quality was strong; the design team had produced effect outputs that were genuinely charming, and the product had a clear emotional hook that other MiniApp products lacked.
The commercial question was specific: could a product with a smaller audience — limited to users who owned pets and wanted to photograph them — still generate meaningful revenue and engagement within the MiniApp ecosystem? And if so, what GTM approach was right for a product where most users, by definition, weren't the target?
My role throughout was GTM strategy, promotional channel management, and funnel analysis. The effect quality and creative direction were owned by the design team.
The Challenge
- The audience ceiling was real and known from the start. Unlike AI Filter or AI Portrait, Pet Portrait required the user to own a pet, have suitable photos of that pet, and want to spend money on a novelty portrait product. Our existing BeautyPlus user base skewed heavily toward young women interested in self-photography. We had no way to precisely target pet owners within the full user base, which meant promotions had to be pushed broadly rather than surgically.
- Cold-start signal was hard to read. Because the addressable pool was smaller, early funnel data was noisier. A low click-through rate in the first days could mean the concept wasn't working, or it could simply mean we were reaching a lot of non-pet-owners who self-selected out quickly.
- The upload requirement created friction for a novelty product. Like AI Portrait, Pet Portrait required users to upload 10–15 photos — a significant commitment for a product they hadn't tried before. For a professional headshot, users had a clear motivation to invest the effort. For a pet portrait, the motivation was more emotional and impulse-driven.
Pre-launch
Accepting the niche and planning for it
The strategic decision before launch was to stop treating Pet Portrait's smaller addressable audience as a problem to solve, and start treating it as a design constraint to optimise within. We were going to reach the right subset as efficiently as possible, then convert them at a higher rate than a mass product would.
This meant the promotional plan prioritised precision over volume. Rather than spending banner impressions broadly, we front-loaded pop-up exposure during the first days to generate the early signal we needed to calibrate — the same cold-start approach that had worked for AI Filter. If click-through from pet-owner-likely users was strong, we'd know the concept was landing. If it was weak even with aggressive early promotion, we'd know earlier and could adjust spend allocation away from Pet Portrait toward the Halloween AI Filter campaign running in parallel.
Phase 1 — October launch
Aggressive cold-start, then read the data
Pop-up windows were the primary cold-start channel: 22% click-through rate on the in-app pop-up surface, consistent with other MiniApp launches. Banner placement at 0.83% CTR served a different function — self-navigating banner users showed significantly higher purchase conversion (2.01% post-click), so banner impressions were treated as the high-intent channel even at lower volume. The MiniApp Tab entry point (0.39% CTR) was the lowest-volume channel but generated the most committed users.
Of the ~846,000 users who saw the Pet Portrait entry across all channels, only 6.5% clicked through to try the product — significantly lower than AI Portrait (19% Western, 13.5% Japan/Korea) and much lower than AI Filter's 30%+. This gap was expected and consistent with the niche hypothesis.
The steepest drop-off was at the photo upload step — only 39% of users who reached the photo selection page completed the upload. Two hypotheses: users didn't have 10–15 suitable pet photos available, or the photo quality guidelines felt too strict for casual pet photography (pets don't hold still; most phone photos have motion blur, partial frames, or inconsistent lighting).
One finding from style performance worth flagging: all four styles showed remarkably even purchase distribution — Pawtrait (29.9%), Master Chef (27.2%), Wizard (21.6%), Birthday Party (21.4%). No single style dominated, which meant users were genuinely exploring the catalogue. It also reduced promotional complexity — we didn't need to identify a "lead style" to feature in banner creative, because all four performed comparably.
Phase 2 — Christmas campaign
Scaling the niche, and what breaks when you do
By December, Pet Portrait had earned the highest promotional priority across all MiniApp products — more banner and pop-up inventory than any other product in the portfolio. New styles launched in waves: a Disney-inspired style on December 7th, two Christmas-themed styles and a street photography style on the 11th, and kimono and celebrity styles on the 26th. The catalogue had grown from 4 styles to 12.
The Christmas campaign produced the portfolio's strongest revenue month. Pet Portrait contributed approximately 40% of all MiniApp GMV in December, and December 25th became one of the highest single-day revenue days across the entire MiniApp ecosystem.
But scaling also exposed a problem that hadn't been visible at launch. Overall conversion rate fell from 0.3% in October to 0.1% in December. Promotional volume had tripled, but the incremental users arriving through that expanded reach were less qualified than the early adopters who had self-selected in October.
Results
| Metric | Phase 1 (Oct, 14 days) | Phase 2 (Dec, Christmas) |
|---|---|---|
| Share of MiniApp revenue | 51% of Halloween period | ~40% of December GMV |
| Overall conversion rate | 0.3% | 0.1% (dilution from broader reach) |
| Top-of-funnel click-through | 6.5% | 4.9% (post UI update) |
| Upload step completion | 39% | Declining — flagged for iteration |
| Bottom-funnel conversion | 21% of users reaching payment | Consistent |
| Style distribution | Even across 4 styles (~21–30%) | pixie_paws dominant at ~31% |
| Paying users | ~2,200 | Highest single-day revenue on Dec 25 |
| Promotional rank | Mid-tier | #1 across all MiniApps |
What I Learned
A small addressable audience isn't a problem if your bottom-funnel conversion is strong.
Pet Portrait's top-of-funnel rate looked weak compared to other products, but users who self-selected through the niche filter converted at the payment step at a meaningfully higher rate than AI Portrait users did at equivalent funnel depth. The niche nature of the product did part of the qualification work — by the time a user had decided to upload 15 photos of their cat, they were already committed.
Upload friction is a bigger barrier for impulse-driven products than for functional ones.
For AI Portrait, users had a clear professional motivation to invest 10 minutes finding good photos. For Pet Portrait, the motivation was emotional and novelty-driven — the decision to start was more impulsive. This means every additional second of friction at the upload step cost us a higher proportion of the audience. Reducing upload requirements or adding a photo quality preview before submission would have been worth testing.
Scaling a niche product requires a different strategy.
The move from October to December was a transition from precision targeting to broad reach, and the conversion rate drop from 0.3% to 0.1% was a direct signal of that shift. Early adopters who found Pet Portrait in October were high-intent by self-selection. Users arriving through December's expanded promotional inventory were not. The lesson isn't that scaling was wrong — promotion volume alone doesn't maintain conversion quality once you've moved beyond your core audience.
All metrics and performance data referenced in this case study reflect the 2023–2024 period. Sensitive figures have been anonymised or aggregated, and no proprietary or commercially confidential information is disclosed.