The natural instinct of many founding engineers is to make a hacky MVP first and iterate on that as they get more feedback and learnings. But in the case of building data analytics and ML applications, starting with the data first usually leads to less wasted effort and better insights.