Havit
Can a "do i already own this?" app turn collector anxiety into a real business?
The make-or-break question is whether Havit can win a narrow niche where duplicate mistakes are expensive enough that people trust it, use it monthly, and pay for it.
If it stays a charming can app, it gets copied or ignored; if it becomes the collector memory for high-value niches, it has a shot.
Pushes for a tight wedge: high-frequency, high-value collectors who buy in person and feel the pain at the shelf.
Tests whether the niche is big enough, the retention is real, and the revenue can pay back acquisition fast enough.
Focuses on whether the current AI pipeline is good enough, defensible enough, and flexible enough to scale by niche.
The math that has to work
SARAH'S BACK-OF-NAPKINSarah’s bar is 500 active collectors with over 60% paid retention after three months before she’d talk check size.
That only matters if the niche is painful enough that users keep scanning after the first import.
This week on Tri Diligence: Havit, an app that answers one question at the exact moment it matters — standing in a shop, "do I already own this?" Photograph the item and a two-stage AI pipeline (embeddings to shortlist your likely duplicates, then a vision model for the final call) answers in about two seconds.
The founder's hook is real: they collect purple soda cans because the company is called Purpur, and keep forgetting in the store whether they already have that exact can.
Our three hosts fight over the one question that matters:
- Jake (the marketer) — the "do I already have it?" moment fires exactly when the wallet is out; start with high-frequency, high-value niches, not "every collector."
- Sarah (the backer) — is a generic collection app a hobby in a crowded market, or is there real money in niches where duplicates cost real money (cards, sneakers, vinyl, whisky/wine)? And which of three models wins: affiliate at point of purchase, freemium subscription, or B2B white-label to shops and auction houses?
- Ryan (the technologist) — the pipeline already works (serverless AWS plus embeddings plus a vision model); where AI is the product, and where a well-funded incumbent could bolt the same on.
Each host ends with a verdict — invest, wait, or pass — and one concrete first step. The honest bottom line: don't build an app for everything you could collect — build it for the thing you already buy every month.
Transcript
Welcome to Try Diligence, the show where three people kick the tires on one business idea and try not to dent the hood. I'm Jake, I do product and brand, and today I'm weirdly excited about purple soda cans.
I'm Sarah. I do money, risk, and the part where enthusiasm gets asked to bring receipts.
Ryan here. I build the machinery, and today that means photos, embeddings, vision models, and whether A-I is a moat or a chainsaw.
The idea is Havit. You're in a shop, you snap a photo of a collectible, and in about two seconds the app tells you, yes, you own this exact thing, or no, you're missing it. The founder's real trigger is perfect. Purpur collects purple soda cans, then stands in stores thinking, wait, is this the lilac one or the suspiciously similar mauve one?
As a can app, adorable. As a business, I immediately want to know which duplicates hurt enough that people pay. Missing a soda can is annoying. Accidentally buying a second rare card or whisky bottle can be real money.
I like that the hard part is already working. A progressive web app, serverless on A-W-S, image storage, embeddings, then Open A-I Vision for the final call. That isn't vaporware. It's vapor with a Git commit.
Customer segment first. I wouldn't chase every collector. I'd start with high frequency, high value buyers who browse in person or live auctions. Cards, vinyl, sneakers, wine, whisky, LEGO.
The top-down market is real. Grand View pegs the United States collectibles market at eighty-seven point one billion dollars in twenty twenty five. But Havit's reachable market isn't that. It's people with messy shelves and repeat purchases.
Exactly. The slogan isn't catalog your soul. It's do I already have this before I spend forty-five dollars again like a tired genius.
There are two users, though. Consumer collectors need instant duplicate checks. Shops and auction houses need batch cataloging and dedupe. Same engine, different workflow, different willingness to pay.
Value proposition is clean if accuracy is brutal. It saves money, saves embarrassment, and reduces inventory chaos. But if it's wrong twice, users stop asking. Trust is the product.
And the magic is timing. This isn't a Sunday spreadsheet chore. It speaks at the shelf, when the wallet is already stretching its hamstrings.
Technically, the two stage pipeline makes sense. Embeddings cheaply shortlist likely matches from your collection. Then a vision L-L-M compares the store photo against actual collection photos. That's smarter than asking one expensive model to search everything.
How clean is the answer? In cards, a wrong parallel or edition can be costly. In whisky, label changes are subtle. Thresholds by category aren't a nice feature. They're survival.
That's where A-I helps for us. Use category-specific prompts, store multiple angles, learn from user corrections, and ask for a second photo when confidence is low. Cheap automation handles most scans. Human patience handles the edge cases.
Channels are fun here. Start where collectors already brag and panic. TikTok, Reddit communities, Discords, card shows, bottle shops, record fairs. The purple can origin is charming content, not the market.
Paid acquisition worries me. A generic app download ad could be awful. Say C-A-C is fifteen dollars and monthly subscription is five dollars. Fine if retention is strong, painful if collectors churn after importing once.
Then do wedge marketing. Sponsor a checklist at one vinyl fair. Let shops put a Havit scan code by the register. Make the first imported collection feel like a tiny museum.
Delivery as a progressive web app is good for speed. No app store tax, quick installs, camera access, offline-ish caching. Later, native apps only if camera performance or push notifications demand it.
Customer relationship can't just be scan and leave. The collection becomes the lock-in. Wish lists, alerts, insurance exports, valuation history, maybe sale reminders. Otherwise Amazon or eBay can copy the camera trick.
Yes. Turn it into collector memory. You aren't using Havit because it recognizes cans. You're using it because it remembers your taste better than you do.
Notifications should be careful. Tell me a Discogs listing matches my missing pressing. Don't buzz me because a vending machine made eye contact. Relevance is retention.
Revenue has three lanes. Affiliate at purchase, freemium subscription, and business-to-business white label. Affiliate is elegant because intent is hot. But take rates are thin unless the category is expensive.
If the app says you're missing this, and then shows eBay, Tradera, Discogs, or Systembolaget where appropriate, that isn't an ad. That's a shopping assistant.
The app needs clean affiliate attribution and merchant data. For cards, matching against T-C-G-player style catalogs matters. For bottles, barcodes help, but photos catch vintage labels and imported weirdness.
Subscription pricing could be five dollars a month for serious consumers, maybe fifteen for unlimited niches. The real upside is shops at two hundred to one thousand dollars a month if it saves cataloging labor.
White label also makes the brand less cute and more useful. A vinyl store could offer, check your shelf before you buy. That's customer service with a revenue button.
For white label, I wouldn't fork the product. Multi-tenant accounts, brand themes, merchant catalogs, role permissions, and audit logs. Boring software, which is where money often hides wearing a gray cardigan.
Key resources aren't just the code. You need a trusted dataset, enough capital for model usage and support, niche credibility, and someone who can sign partnerships without sounding like a database with shoes.
The current A-W-S stack is a good base. Dynamo D-B for metadata, S-three for images, Lambda for scan jobs, CloudFront and W-A-F for delivery. I'd add analytics early, because every correction is training signal.
The founder build story is a resource too. Build in public. Purple cans become the memorable demo. Then bottles or cards become the serious wedge.
Key activities are accuracy, onboarding, catalog import, and partner sales. I'd add support. When a collector says the app confused two editions, that isn't a bug report. That's a trust emergency.
Import is the sleeper. If users must photograph eight hundred items one by one, goodbye. Use C-S-V imports, marketplace exports, barcode scans, receipt parsing, and bulk photo ingestion. The first hour decides the business.
Partnerships should be embarrassingly practical. One card shop, one bottle shop, one vinyl seller. Give them a dashboard. Watch what they actually scan.
And respect platform gravity. eBay wins on scale and charges seller fees on broad transaction value. Discogs proves niche collectors accept structured inventory, with its nine percent fee model. Havit should complement them before trying to fight them.
Partners also include model providers. Open A-I Vision works now, but keep abstraction around it. If another model gets cheaper or better on product labels, switch without rewriting the house.
Cost structure is mostly variable early. Model calls, image storage, hosting, support, and payment fees. If each scan costs a few cents, fine. If heavy users scan thousands, subscriptions need fair-use limits.
Rough assumption. Ten thousand scans might cost a couple hundred dollars in model usage if you're selective with vision calls. The embedding step is cheap. The expensive mistake is sending every blurry photo to the smartest model.
This is why a narrow wedge matters. A whisky collector scanning twelve bottles a month can support better economics than a casual can collector scanning chaos for nostalgia points.
On market size, I wouldn't pitch the whole eighty-seven billion dollar United States market. Whatnot doing more than three billion dollars in twenty twenty four live sales proves collector commerce can be huge, but Havit still needs a beachhead. Say premium whisky collectors, then wine, cards, and vinyl.
Now the ugly A-I question. A competitor can use the same building blocks. Amazon, eBay, or Whatnot could add, check my collection, right inside checkout. They have catalog data, traffic, and money.
Then Havit needs a wedge they don't care about at first. Own the collector relationship, correction history, niche community, and the one thing big platforms hate, being obsessively specific.
Risk round. What has to be true? First, users must feel the duplicate pain monthly. Second, accuracy must be near perfect for the chosen niche. Third, affiliate or subscription revenue must pay back C-A-C within months, not years.
The technical risk isn't whether it can work once. It's whether it works in bad lighting, with glare, damaged packaging, partial labels, and users who photograph like they're escaping a building.
The focus risk is bigger to me. Don't build an app for everything you could collect. Build it for the thing your first users already buy every month.
My verdict is wait, but interested. I'd invest after proof from one paid niche. Show me five hundred active collectors, paid retention over sixty percent after three months, and scan accuracy users trust. Then we can talk check size.
My verdict is build the wedge, not the empire. Next step, instrument the pipeline, run a blind test on one thousand real shelf photos, and publish the confusion matrix in human language.
I'm an invest-small yes. Next step, launch the bottle version with a partner shop and an affiliate flow. Purple cans got us the story. Expensive duplicates get us the business. This has been Try Diligence.
Havit only works if it stops being a cute collector toy and becomes the thing people already buy every month.