A friend of mine runs a sports card breaking business on Whatnot with two others. Breaking works like this: you buy sealed product — a case of basketball cards, a box of soccer — go live, and sell slots in real time. Each viewer buys a team or a player and receives whatever cards get pulled for them. Part lottery, part live sport, part community.

Since April 2025 I've been their data and strategy advisor. The business was clearly working in the ways you can feel — shows were busy, orders were landing, revenue was climbing. What nobody could answer was narrower and much harder: which individual shows actually made money?

What they could and couldn't see

They could see the top line. Revenue per show was obvious the moment a stream ended. What sat underneath it wasn't. Product cost was spread across cases bought weeks earlier and broken across several streams. Live hours varied wildly. Formats varied more. So the only profitability number anyone had was a monthly one, in aggregate, long after the decisions that produced it.

That makes a specific kind of mistake invisible. If a format loses money every time it runs, but it runs inside a month that was profitable overall, nothing in the numbers tells you to stop running it. It just quietly subtracts.

The model

I built a P&L and stream-level performance model that resolved cost down to the individual show, then tracked 20+ metrics per show on top of it — GMV per hour and profit per hour, AOV, ROAS, orders per minute. Applied across 37 distinct formats, it gave the three of them their first per-show view of the business.

The window it covered: $160K GMV, 12,470 orders, and 479 live hours across five months and 69 streams.

What it surfaced

Finding 01
One stream in three was running at a loss.
23 of the 69 streams lost money. Not marginally — structurally, and repeatedly, in formats that kept getting scheduled because nothing had ever flagged them. A third of the calendar was working against the other two thirds. Aggregate profitability had been hiding a third of the calendar.
Stream outcomes · 69 streams over five months
Profitable
46
At a loss
23
Finding 02
The cause was concentration, not volume — inventory was taking 69% of total spend.
Tracing the losing streams back through the cost side pointed at one line: inventory was consuming 69% of total spend. The problem wasn't that they were selling too little. It was how much capital went into product before a single slot had been sold, and which formats that capital was going into.
Share of total spend
Inventory
69%
Everything else
31%

What changed

The model turned the weekly seller review into where sourcing and scheduling decisions actually got made — rather than a recap of a week already spent. Formats that couldn't clear their own cost got cut. Others got repriced.

Over the same stretch, monthly revenue grew from $19.9K to a $54.1K peak. I want to be careful about what I claim there: three people running better shows is what produced that. What the model did was make the losing third visible, and give the weekly conversation something concrete to argue with.

A note on this work. This is unpaid advisory work for a friend-operated business, done with their permission to write about it. The figures here are the ones they've agreed to share. I'm not an owner, employee, or contractor of the business, and I take no compensation from it. Analytics didn't grow this business — the three people running it did. What the model contributed was visibility: a per-show P&L where there had only ever been a monthly one.

Why this one matters to me

The standard portfolio project uses clean data with a known answer and no consequences. This had the opposite properties. The data was messy and small. The question wasn't handed to me — "which shows make money" only became the question after a few weeks of looking at the wrong things. And the answer had stakes for people I know, who were going to act on it.

Telling three friends that a third of their calendar was losing money is a different exercise from writing it in a notebook. That's the part I'd want to be doing in a Summer 2027 internship — analysis that someone is actually going to decide something with. More on the broader plan here.