The decision layer for modern retail.
Built by the people who wrote the playbook.
AI-driven predictive models and control systems for online retail, travel, and media — the kind built inside Google and Amazon, deployable this quarter.
Causal price elasticity at the SKU level, with marketplace-wide cross-elasticity.
Causal ML estimates true price elasticity at the SKU level — not correlations, not category averages. Cross-elasticity reveals how a single price change ripples through your full catalog, so you can simulate marketplace-wide profit impact before committing. Optimize automatically or manually at SKU level.
Pricing, inventory, and supply chain — optimized as one system.
Pricing without inventory awareness guarantees stock-outs or markdowns — both destroy margin. Stellar integrates inventory levels, sales velocity, perishability, and supply chain costs into price optimization. Constrained inventory triggers higher prices; perishable goods get markdowns timed against carrying cost; sourcing decisions are scored against the prices they enable.
Substitutes mapping reveals the gaps in your selection.
The Substitutes Map projects every product into a space where proximity equals substitutability. Combine with elasticity and demand signals to spot high-traffic regions with no product mass — demand your catalog isn't capturing. Direct 1P merchandising to fill the gaps, or propose adjacent products to top 3P sellers.
Ad ranking and auction pricing for retail media networks.
Ranking and pricing are the two highest-leverage decisions in any retail media network — and the two most often left to vendor defaults. Our stack ranks and prices ads to maximize long term profitability, not just clicks x bids. Built by the team that scaled Amazon Sponsored Products 100×.
Multi-layered control systems for stable, predictable spend.
Maximizes revenue within whatever constraints your advertisers define — average CPC, CPA, ROAS, budget caps, or any combination. Multi-layered control systems deliver not just peak performance but stable, predictable spend. Less wasted budget. More value to your partners. More profit to your bottom line.
Click, conversion, and revenue prediction that outperform benchmarks.
Accurate predictions are the foundation of every high-performing ad system. Our click, conversion, and revenue models consistently outperform industry benchmarks — powered by deep neural networks with proprietary user and product embeddings on modern encoder/decoder architecture. More relevant ads, higher conversion, better returns.
Quantified substitutes and complements for any product.
How products relate is foundational to great recommendations and pricing. Our embeddings instantly identify the closest substitutes and complements for any product, with quantified substitutability and complementarity scores for any pair. They power carousels, recommendation engines, and serve as high-signal features in downstream ML.
Stop wasting impression space — surface what converts for each shopper.
Generic ranking treats every shopper the same and trades long-term value for short-term clicks. Our stack scores each impression against the probability that this shopper buys this product, in this moment — using user embeddings, the Substitutes Map, and our conversion models. Impression space stops being inventory and starts being a conversion engine.
Our founders have led ads, commerce, and ML systems at Google and Amazon — we've built these systems at a scale few teams in the world have matched.
Our products are designed for immediate deployment, not multi-year research engagements.
We measure our success by your revenue growth, not model accuracy in a sandbox.
Lars built the Sponsored Products auction, quality, and bidding systems at Amazon — one of the highest-revenue ad platforms in the world. As a leader across Product Management, Applied Science, and Engineering at Google and Amazon, he developed deep, hands-on expertise across the full ad tech and commerce stack. His work on targeting, bidding, and graph-cluster-based experiment design has been adopted broadly across the industry, delivering significant revenue growth at scale.
Ram architected the Sponsored Products ad server at Amazon, scaling it 100× to meet global demand. His expertise spans the full ad stack — auction dynamics, price optimization, and ML pipelines for relevance and conversion prediction — built across 20+ years designing Tier-1 distributed systems at Bloomberg, Amazon, Convoy, Flexport, and Avantus. He also led the architectural redesign of Amazon's A/B testing and analytics infrastructure, enabling safe, high-velocity experimentation at massive scale.
Yannet spent five years as a Data Scientist at Google. There, she built models to analyze ad quality, user segmentation, and lifetime value across YouTube and Google TV ads. She then moved into startups, building machine learning models as a senior data scientist and co-founder. She was a Professor of Data Science at the University of San Francisco, teaching Machine Learning and Deep Learning. Additionally, she's held visiting appointments at UC Berkeley's Statistics Department and Esade in Barcelona. Her research applies deep learning and transformers to problems in natural language, recommendation systems, and applied science.
Brian led product for Google's Search Ad Automation — automated bidding, targeting, advertiser-facing conversational experiences, the Ads AI GenAI/LLM partnership with Google DeepMind, and the Ads ML real-time inference platform. Before Google he was a Director at Amazon, where he incubated and scaled the Amazon Scholar Program and led cross-functional product, science, and engineering teams in demand forecasting, price elasticity, and large-scale experimentation — including Amazon's Weblab experimentation platform and clickstream analytics backplane.
30-minute conversation. We'll show you what's possible with your data.
hello@radicaltech.ai →