Hugo speaks with Sarah Catanzaro, General Partner at Amplify Partners, about investing in data science and machine learning tooling and where we see progress happening in the space.

Sarah invests in the tools that we both wish we had earlier in our careers: tools that enable data scientists and machine learners to collect, store, manage, analyze, and model data more effectively. As you’ll discover, Sarah identifies as a scientist first and an investor second and still believes that her mission is to enable companies to become data-driven and to generate ROI through machine and statistical learning. In her words, she’s still that cuckoo kid who’s ranting and raving about how data and AI will shift every tide.

In this conversation, we talk about what scientific inquiry actually is and the elements of playfulness and seriousness it necessarily involves, and how it can be used to generate business value. We talk about Sarah’s unorthodox path from a data scientist working in defense to her time at Palantir and how that led her to build out a data team and function for a venture capital firm and then to becoming a VC in the data tooling space.

We then really dive into the data science and machine learning tooling space to figure out why it’s so fragmented: we look to the data analytics stack and software engineering communities to find historical tethers that may be useful. We discuss the moving parts that led to the establishment of a standard, a system of record, and clearly defined roles in analytics and what we can learn from that for machine learning!

We also dive into the development of tools, workflows, and division of labour as partial exercises in pattern recognition and how this can be at odds with the variance we see in the machine learning landscape, more generally!

Two take-aways are that we need best practices and we need more standardization.

We also discussed that, with all our focus and conversations on tools, what conversation we’re missing and Sarah was adamant that we need to be focusing on questions, not solutions, and even questioning what ML is useful for and what it isn’t, diving into a bunch of thoughtful and nuanced examples.

I’m also grateful that Sarah let me take her down a slightly dangerous and self-critical path where we riffed on both our roles in potentially contributing to the tragedy of commons we’re all experiencing in the data tooling landscape, me working in tool building, developer relations, and in marketing, and Sarah in venture capital.