I recently revisited one of the most practical conversations I’ve had about scaling: insights from Ian Wong, co-founder and CTO of Opendoor. Before founding Opendoor, Ian was Square’s first data scientist, where he developed machine learning models and infrastructure for fraud detection. That trajectory—building from operational muscle to algorithmic advantage—maps closely to how I’ve led product and data investments in fast-growing environments.
As Ian puts it, in the early innings it might make sense for your startup to be operations heavy. But as you start to scale, data science becomes a critical component for running a business with longevity in mind. This mirrors my experience: hands-on operations create the learning loops you need early, while data science turns those learnings into scalable systems, better forecasts, and tighter risk controls.
We dive into how both Square and Opendoor approached this transition. The progression is instructive—start with pragmatic operational workflows to validate demand and unit economics, then shift toward machine learning and automation once the process is stable, the data pipeline is trustworthy, and the cost of manual work begins to drag margins and responsiveness.
Along those lines, we discuss some of the early considerations for your fledgling data science team, including the type of folks to hire for the early team, like whether to look for generalists or specialists, and how to set up your interview loops. In my playbook, I bias toward T-shaped generalists first—people who can partner with product, analytics, and engineering—then layer in specialists (ML ops, causal inference, experimentation) as complexity grows. For interview loops, I ensure we evaluate for problem framing, data intuition, model-to-product translation, and stakeholder communication—not just model accuracy.
Ian also dives into his lessons on structuring the data science function so that it’s deeply integrated with the rest of the technical org. I’ve found embedded pods work best early—data scientists sit with product teams to accelerate discovery, instrumentation, and iteration—paired with a light central platform group to standardize data quality, experimentation frameworks, and model deployment practices.
Next, we dive into some of his biggest lessons as a first-time founder and CTO, including his practice with Opendoor’s leadership team of doing pre-mortems to predict why something might not work. He also encourages founders to run through a bi-yearly exercise of re-writing their job rec. I’ve seen both rituals raise the bar: pre-mortems surface hidden risks before launch, and re-writing the job rec forces leaders to shed responsibilities, prevent role drift, and keep the org structured for the next stage—not the last one.
Practically, my guidance to founders and product leaders is straightforward: define the decisions data science will improve (pricing, risk, routing, personalization), instrument for leading indicators, and ruthlessly prioritize the smallest models that deliver outsized business value. Avoid premature optimization—let operations teach you where the algorithm belongs. Use clear success metrics to track outcomes, not just output, and revisit them as your market and product expand.
Finally, remember that the goal isn’t to replace operations—it’s to make operations smarter. Pair human judgment with machine learning in the workflows that matter most, invest in trustworthy data foundations, and build hiring and interview loops that reward interdisciplinary problem solvers. That’s how you turn early operational grit into durable, data-driven advantage.
