Stand wins a $1 billion valuation for pricing wildfire risk by simulating each home

If the $1 billion is a post-money figure, $70 million sells 7% of the company ^[7]. Set against the more than $10 billion of property Stand says it insures ^[9], the valuation is at most 10% of insured value ^[8]. Insured value is not premium, so the ratio measures scale.

The price is paid for growth. The growth figures do not quite fit together. Insured property went from $1 billion to more than $10 billion in a year, a rise of at least tenfold ^[14]. CEO Dan Preston says the business is growing about 15% a month ^[10]. Applied to insured property, 15% a month compounds to about 5.35 times over 12 months ^[11], or $5.4 billion from a $1 billion start ^[12], and a tenfold year takes roughly 21% a month on average ^[13]. Either growth ran faster earlier in the year and has since slowed, or the two numbers measure different things.

The value case is the backtest. The 85% is the headline figure, or rather the second most interesting one; the first is the survival share of the 310 homes ^[15], which Fortune's account does not give. A model that called every home a survivor would score exactly that share, so 85% against 50-55% means little until the split is known. In our view there is a second soft spot: homes hit by the same fire share wind and weather, so 310 homes are fewer than 310 independent tests. The 50-55% for incumbents comes from running industry-standard location-based models against the same cohort ^[16]. Both figures are chief science officer Matt DiStefano's, as reported by Fortune.

The approach itself has a real argument behind it. Stand says insurers price wildfire by looking backward at statistics and sorting homes by zip code ^[4]. DiStefano described the alternative this way: "We're not looking at the neighboring home. We're looking at this specific home, the specific materials and the specific arrangement of trees." ^[5] The product that follows is a price that moves when the owner acts. In the demo, one deleted tree changes the premium on screen ^[3], and Stand says an in-house neural network, the Stand World Model, runs the physics simulations instantly ^[6].

We think pricing each house is the right direction, because a score an owner can act on tells the owner where to spend. But $1 billion pays for growth in two states, California for wildfire and Florida for hurricanes ^[10]. The backtest has not yet shown that the valuation is justified. Two outcomes would move us. A loss ratio across a full fire season that matches the 85% would support the valuation. A Stand-insured home that the model rated safe and that burned would cut against it.