Laurence

Advertising today is like trading in the 1980s pits: archaic, manual, and iterative guesswork. Humans update bids and keywords using noisy, delayed data with fixed rules that both break at scale and ignore stochasticity. We've built quantitative systems that advertise the way modern hedge-funds trade markets: continuous decision-making under stochastic outcomes and explicit profit constraints. We use customers' existing ad copy and use reinforcement learning and train our own LLMs to run ads on autopilot, harvesting profits when we’re confident and borrowing signals from similar keywords when data is sparse. Starting with brands on Amazon, we are automating tens of millions in live ad spend, increasing gross sales while decreasing advertising cost of sales by 40%.