$250,000 a month from a wine-cellar app, with AI making the ads that sell it
A founder says 60-80% of his app's code is now written with AI, and a small in-house team ships many ad versions a week. The sharper lesson is what happened when he doubled the ad budget.
In this article
Somewhere in Toronto, a collector smashed a bottle of Caymus while hunting for it in his cellar. His son, an engineer who had studied AI, built him a simple app to log every bottle and show which rack and slot it sits in. Seven years later, as the founder tells Starter Story in an interview published on 20 October 2026, that app brings in around $250,000 a month.
The details matter more than the number. About 500,000 collectors now use it to manage more than $3 billion of wine, from a single fridge to cellars of 10,000 bottles. It has a sommelier-curated database of 2 million wines, a 3D replica of your cellar, and an AI sommelier you can ask what to drink tonight.
But the part that is actually copyable is not the wine. It is how the business buys customers, and how AI changed that machine.
Start with a problem someone already pays to solve
The first version did one thing: log a bottle and show exactly where it is. That was enough to test whether people would pay. The founder offered a free tier and a paid tier from the start, then put about $15,000 behind ads one summer. The first paying customer handed over roughly $50 for unlimited bottles.
That early plan became the paid tier, and the founder says the paywall converted at roughly twice the typical rate for an app. By the end of 2023 the app had 22,500 monthly active users. Paid users cost about $79 to acquire and paid that back within ten months.
You do not need a wine cellar to copy the shape of this. Pick a specific group with a specific annoyance, build the smallest version that fixes it, and charge from day one so you learn whether the problem is worth money.
The ad that worked became the problem
Growth ran on paid social. The best ad the company ever made was built around a single idea called “The $400 Mistake”: what it costs to lose track of a great bottle. It carried about 31% of the Meta spend.
That concentration was the risk. As the ad wore out, costs climbed, and the outside agency could not produce new concepts fast enough. So the founder ended the agency and moved creative in-house using AI.
The new loop is deliberately unglamorous. The team makes many versions a week, gives each a small budget, kills the losers quickly, and scales only the winners. The bottleneck, in the founder's words, is usually creative rather than budget.
Watch the cost of the last dollar
The most expensive mistake in the interview came from believing more spend would buy more growth. In March and April the company pushed ads to between $93,000 and $123,000 a month. Customer acquisition cost jumped to $148-$187, nearly double normal.
When the founder pulled back over the summer, acquisition cost fell and growth barely slowed. The last dollars of ad budget were the most expensive ones. Normal spend is now about $81,000 a month: 80% Meta, 17% Apple Search Ads, 3% Google, with a blended acquisition cost near $80.
The founder's predicted two-year value of a subscriber is $225-$270, roughly three times what he pays to get one. Those are his figures and they are not independently audited, but the ratio is the number to watch in any paid-growth plan.
Use AI where it compresses the work
AI shows up in three places in this story, and only one of them is a feature.
The visible one is the AI sommelier. Short videos of people asking it what to drink became the organic engine: three reels passed 1 million views each, and the founder says that attention made the paid ads convert better because people had already seen the product in their feed.
The quiet one is software. The founder says 60-80% of the app's code is now written with AI, and a team of about eight shipped more this quarter than ever before.
The commercial one is the ad factory. AI lets a small team produce and vary creative at a pace an agency could not match, which matters because the winning ad always dies eventually.
Fix the checkout before you scale
Payment was another lever. The app moved buyers from Apple's in-app purchase to a web checkout run through Stripe. That route now handles 40% of revenue, up from 17%.
The founder's stated reason is arithmetic: Apple takes 15-30%, Stripe about 3%, and web subscribers renew at roughly 91% versus 62-65% through the App Store. The trade-off is that you now own the checkout, the refunds and the fraud, so the saving is not free.
His other costly lesson was simpler: the company once ran a web trial test that produced about 9,000 trials and almost no payments. The flow after the click was broken. After fixing it, the web business had record months, with 356 new paid subscribers in July and 419 in August.
What to copy, and what to leave
Copy the order of operations: charge early, measure acquisition cost and payback, then scale only what you can afford to repeat. Copy the creative loop: many cheap versions, fast decisions, no attachment to the ad that is working today. Copy the payment review: compare what each store and processor actually keeps, including renewals.
Leave the assumptions. This is a seven-year-old product with 500,000 users, a curated database and a visual hook that happens to be fun to watch. A new app does not start with those advantages.
Also treat every figure as self-reported. The interview is a promotional case study, revenue is not profit, and the founder does not break out margins, salaries or the full cost of the ad machine. The method is real; the outcome is his, not a promise.
Sources & further reading
How I Built A $250K/Month App For Managing And Valuing Wine Collections — Starter Story
Our articles help you explore AI and build practical skills. For a correction or a question, contact us. Contact
