The problem
In 2020, during my MBA at Insper, I was planning a small online store for board games. The research showed a stable pattern: most retailers sell 8% to 10% of their inventory every month. If the stock kept its value, reinvesting every profit into more stock could become a steady business.
Three problems stood in the way.
Getting buyers was too expensive. Paid ads would eat most of the margin, and content marketing meant producing videos for several platforms at once.
Marketplaces solved traffic, but pulled everything toward price. On a marketplace my listing sits next to everyone else’s, sorted by price. Lowering prices kills the margin; keeping them high means not selling. And the cheapest seller is often the marketplace itself.
Marketplaces keep the customer. Buyers feel they bought from the marketplace, not from me. After thousands of sales I would still have no audience of my own.
In September 2021, Whatnot raised $150M in the US with live auctions, where sellers build a following like creators do. If it existed in Brazil, I would have used it. The question became: does that model work here, and can I find out without building anything?
What I decided
Test the model, not the software. No app, no code. The goal was to learn whether live auctions would sell, keep buyers coming back and escape price comparison, at almost no cost.
Choose the channel by delay and accountability. Instagram live had about a minute of delay between video and chat, which makes an auction impossible. Twitch and YouTube gave bidders too little accountability. WhatsApp had no video in groups and a cap of 256 members. Telegram had video inside private groups, much larger groups and a delay of seconds. Telegram it was.
Assemble everything else from what already existed. A simple website to bring people into the private group. Mercado Pago and other payment links, with Pix, cards and installments from day one. Google Forms for data and operations. Local couriers to bring products to a small studio inside the first seller’s store, and a host who already had streaming experience.
Start with one seller and the collectors themselves. I ran research with collectors first, then asked the sellers who answered if they wanted to join an auction. The first one said yes. The MVP was ready in less than a week, and the first live show happened on December 14, 2021.
The result
The first auction sold R$4,200, which meant R$336 in revenue at an 8% take rate. Over the following year:
- 320 live auctions and more than R$850K in sales.
- More than 20 sellers and more than 1,000 buyers; 60% of them bought in more than one auction.
- The Telegram community grew from 120 to more than 2,000 members, without paid acquisition.
- 97% of the more than 10,000 products offered were sold.
Price gravity eased, but did not disappear. Some buyers posted marketplace prices in the chat, and some sellers did it themselves to show a good deal. The auction kept prices honest and only sometimes pushed them above the market, depending on how many people were watching.
The relationship moved to the seller. 30% of buyers added more items from the same seller to share the shipping cost, and people complained when a host was replaced: part of the audience came for the person, not only for the products.
What I took from it
A business model can be validated with tools that already exist. The MVP proved demand, showed that a live auction needs a minimum audience to work, and gave me the numbers to decide what to build next. That became Moxen.
I was later invited to present the case in Insper’s postgraduate entrepreneurship course as an example of how to test a business idea before building it.