Today an article by CB Insights caught my attention when it commented on the recent bankruptcy of Thrasio. Thrasio is an e-commerce startup that was valued at $10B after raising around $3.4 billion from top-tier investors including BlackRock, Silver Lake and Goldman Sachs.
Thrasio's bankruptcy stood out because for years our AI models estimated Thrasio's valuation as nearly 30X smaller than the whopping $10B it had successfully fundraised at.
The AI also detected Thrasio's decline long before it was publicly announced.
In other words, Thrasio is an example of AI being better at valuing a startup and avoiding a hype-cycle than even some of the world's most sophisticated human investors with intimate knowledge of the business itself.
Thrasio is an "Amazon aggregator", which is a business that acquires third-party sellers on the Amazon marketplace (or on other online markets). Aggregators try to identify profitable e-commerce brands and roll them up, with the goal of optimizing and scaling their operations while sucking up free cash flow. A throng of aggregators got drunk on capital during the Pandemic when online shopping surged, but the post-Pandemic world left many with a hangover. According to PracticalEcommerce and CB Insights, global funding of Amazon aggregators declined 88% between 2021 and 2022.
In the case of Thrasio, our AI estimated the business's valuation as being somewhat in line with its actual market valuation (as disclosed from funding rounds) until the first quarter of 2021. Then there was an abrupt divergence as investors surged Thrasio's valuation upward.
The AI did, indeed, see Thrasio continuing to grow throughout most of 2021, but not nearly to the extent of its huge valuation leap.
Our AI doesn't use fundraising data or publicly disclosed valuations (from funding rounds) when estimating the value of startups. Instead it uses a host of other signals and complex modeling to independently infer a business's valuation, mathematically. In other words, our AI attempts to estimate business valuations independently of fundraising or other exogenous factors. It also doesn't sit on the board, get confidential business updates, know the personalities involved or otherwise have insider knowledge of what's going on in a business. Still, it would seem the AI was better at keeping a cool head, accurately assessing the value created by Thrasio and detecting when things began to turn sour.
This pattern was similar to that of Benitago, another Amazon aggregator, just a couple months ago.
For nearly two years Benitago, another Amazon aggregator, had been showing an approximately 30X difference between its estimated dollars raised (from HSBC and others) and our AI's estimate of the company's valuation.
As in the case of Thrasio, the AI was more or less in line with Benitago's valuation at first. Then, in 2021, Benitago raised around $625 million in funding. From the AI's vantage point, this coincided with a brief increase the business's valuation before it began to rapidly decline. This September it was announced that Benitago had filed for bankruptcy.
Before going bankrupt Benitago claimed to have developed ten brands, more than 300 products and to have acquired nearly a dozen third-party brands.
We continue to see large discrepancies between the valuations of most Amazon aggregators and AI estimates.
We weren't alone in hearing the bubble pop for Amazon aggregators. As the bodies began to pile up there have been many analyses of what went wrong. Our AI shows a small number of aggregators doing okay while most of their roughly 80 rivals show varying signs of distress.
While tragic as a business story, it's uplifting as an AI story.
As AI shines more light into the relatively data-poor, dark, opaque worlds of startups and private companies, it allows us (humans) to better spot big opportunities and to avoid big mistakes.
It reminds me of a similar case last June when our AI had persistently estimated a startup's valuation at 10X below its fundraising valuation, only to see the startup fail when it was revealed that 95% of the app's users had been faked by the CEO. In that case, the CEO was able to defraud investors and coworkers but he didn't fool our AI. In the case of Thrasio and Benitago it wasn't an issue of fraud, but similarly there was a perception of the business's value held by investors and insiders that didn't square with the facts on the ground - at least in an AI's opinion.
While neither AI or humans will ever be perfect, having better tools let's us all do a better job. At a minimum, in the case of Amazon aggregators, AI was able to navigate a $16 billion dollar bubble in ways that even seasoned insiders and experts weren't able to.
To the extent that this AI and others like it continue to evolve and bring visibility to private markets, I'm hopeful for a future with less failure and more success for investors and entrepreneurs alike.