Lessons from Cybersyn
What I learned from running a DaaS startup for two years.
On September 11, 1973, Augusto Pinochet seized power in Chile, ending Salvador Allende’s regime and the Project Cybersyn initiative. Pinochet’s rule is infamous for its human rights abuses and brutal repression, but the economic legacy is a separate, more complex story. Under his regime, the "Chicago Boys," a group of market-oriented economists trained at the University of Chicago, implemented controversial reforms that, despite their divisiveness, were largely successful and continue to shape Chilean politics today.1
In a less dramatic turn, I recently shut down my data startup, Cybersyn. Named after the Chilean Project Cybersyn, our company shared the vision of using real-time data to measure the economy—though with a focus on shareholder value rather than Marxist ideals. Cybersyn will be a footnote in Snowflake’s history2, much like its namesake in Chile’s. However, for those interested in data businesses, I’ve captured my key learnings here for posterity.
When we started Cybersyn, we had a core thesis:
Use of third-party data would grow, and there would be a “moneyball-ization” of more industries driving demand for novel third-party data. This would be similar to what I’d seen in discretionary asset management.
Underutilized consumer data could be licensed from data owners, providing them with new, high margin revenue, and derived into useful products.
Cybersyn could adopt a capital-intensive model for licensing data, then scale by leveraging network effects and operating leverage (no revenue shares). Snowflake’s financial support, reputation, and distribution would enable this strategy.
The “10x” innovative data product would combine transaction, point-of-sale, and clickstream data at the consumer level. This product would offer insights into consumer behavior, combining the best of what the industry calls syndicate and panel data.
The core business model proved very difficult to execute. We learned that:
On the customer side, outside of asset management, insights datasets are most valuable for strategic decision making at the largest of companies. It may seem that digital-native DTC brands would be early adopters, but in practice, the time and complexity to realize a return on data-driven insights does not justify large contract values. I did not appreciate just how much longer actions and decisions take in operating business versus investing. Activation or performance marketing datasets likely would be easier to sell while subscale.
Consequently, it was difficult to find nimble, early adopters for whom the ROI for better data would make sense but who could move very fast. While we made genuine innovation on the product, facing existing insights providers for large CPG manufacturers’ business head-on was difficult without very broad data coverage.
On the supply side, large corporations (who have the most valuable data) were, in principle, interested in monetizing their data, and this trend was clearly accelerating. However, creating mutually beneficial deals proved very difficult. Data from any one company is narrow. We had to acquire multiple datasets and pay upfront, despite the time needed to generate meaningful insights (and therefore revenue). There was a tricky balance between convincing large organizations the opportunity of monetizing data was large, while simultaneously trying to negotiate a low initial price.
Alternatively, starting with commercially available proprietary data (and innovating on its processing) was far more affordable, but still challenging due to the time needed to build differentiated products, especially if the only distinction is in the derivative calculations. With already available data, the bar for building something differentiated was far higher.
In the end, the challenge with the Cybersyn business model could be summarized as having higher than expected capital-intensity, paired with slow and non-gradual offsetting revenue, due to the R&D timelines needed to build the “10x” product.
Beyond these difficulties, there were also exogenous factors:
The public market began to value profits over growth. Snowflake was not exempt from this scrutiny and our status as a consolidated entity added constraints.
Marginal venture dollars sought direct AI companies, while becoming tighter more broadly.
The above realizations led to the conclusions:
The money we raised for Cybersyn was not enough to accomplish our vision. We could not buy enough additional data types (point-of-sale, clickstream) beyond transaction data, while still maintaining a long enough runway to ensure we had a chance to complete and market with the “10x” product.
We could not raise more money because of the change in financing conditions (we were not, squarely, an AI company), our unusual capital structure, and most importantly, financial profile.
Continuing to build best-in-class transaction data products seemed unlikely to lead to the scale of outcome that excited us.
The public domain data products became very popular, even to the point where there is likely a company to be built around that alone.3 Demand from AI inference use cases and the improving ability of LLMs to clean and structure data both supported this business. I remain unsure whether any moats will protect first-movers in this space, but there is significant demand here. I predict an explosion of low-cost new options data from LLM-based data structuring.
Cybersyn had the opportunity to return significant capital to our investors based on our financial position and by selling the assets of the public domain business to Snowflake.
These realizations led us to shut down for the purpose of returning maximum capital. In another version of this story, a more reckless (or courageous) founder might have pushed ahead, spending the remaining capital on acquiring the necessary data, even with uncertain prospects for fast revenue growth. Snowflake, like other data technology companies, face tough capital allocation decisions in the new AI world. Personally, I still believe proprietary data content may still be a deserving strategic choice for companies.
On a personal note, I owe immense gratitude to a large number of people. First and foremost, the Cybersyn team took the risk to go on this journey. I am most proud of the talent density we assembled Second, this journey would not have been possible without Thomas Laffont, Christian Kleinerman, Mike Scarpelli, Lauren Reeder, and Mike Vernal. It was an immense privilege to work with Coatue, Sequoia, and Snowflake. Finally, a very large number of partners, customers, suppliers, and other investors were instrumental - I will not attempt a list for brevity and fear of omission, but you know who you are! Thank you.
I look forward to continuing to write and work on the topics in this blog, onto 2025!
See my synopsis of the Chilean Cybersyn story below, Cybernetic Revolutions, or the new podcast series, Santiago Boys
As most readers will be aware, Snowflake was Cybersyn’s largest backer, and we had a unique partnership with the company. See here for the background.
I published extensively on the topic of public data below. I think the rate at which LLMs can automate and assist in the core value propositions is understated.
Thanks for sharing the CyberSyn story. I really admire the vision and the work you & the team put into building data products. The data space offers big opportunity, and I’m excited to see the next chapters for you and the CyberSyn team members.
Really appreciate you taking the time to outline your learnings and educate others. Look forward to the next adventure Alex! As always, I'm only a call away if there's anything I can do to help.