For people who watched good products disappear
Founded in Sydney in 2023. We saw recommendation logic work at 5,000 products, then become a liability at 500,000. We built the engine to solve that.
Category ranking was never made for large catalogs
Showing 48 products from 350,000 makes 349,952 invisible. Something at position 200,000 may be exactly right for this shopper, while the product at position 1 may be irrelevant.
The usual answer is a merchandising team writing rules: boost new arrivals, penalise slow-movers, pin seasonal picks. That works at small catalog sizes. At 100K products and above, rules conflict, age quickly, and need constant care. Last week's curated pages may not suit today's traffic.
ParticularAudiance replaces the rules layer with a model that reads each session and returns a ranked list for that shopper. No rules to write or pages to maintain. The right product, in the right place, for each person.
Three people, one shared obsession
Built inside the ecommerce infrastructure problem we set out to solve.
Recommendation infrastructure work showed him the same pattern repeatedly: quality products get buried once a catalog passes 50K SKUs. Catalog volatility was a structural problem, not a settings issue. That understanding led to ParticularAudiance.
Works on ML systems for high-volume event streams and live feature pipelines. She designed the ranking engine to process each session signal and return a scored product list in under 50ms, while keeping results sound at real retail catalog volumes.
Brings ecommerce product management experience across platform types and catalog sizes. He connects ranking and discovery to live storefronts, owns the connector layer, and shapes onboarding through the first ranked result in under a day.
Our principles
Particular, not general
We optimise each session, not a segment. Even 50,000 shoppers are still a segment. Our ranking is for the shopper in front of us.
Real-time or nothing
Yesterday's model cannot know what today's shopper wants. Delay in the ranking signal leaves revenue behind. Sub-50ms is required, not optional.
Rules are the issue
Sorting rules point to a system that cannot learn. Each rule bets that judgment beats session data. At scale, it rarely does.
Integration, kept simple
A ranking engine taking three months to connect is not useful. Integration belongs in a working day, not a project. We design for that whenever we work on the connector layer.
Angel-stage, AU-based, product-focused
We closed a $500K angel round in early 2026 for the initial engine and first retailer integrations. We are not seeking more funding now. The business is built to show measurable revenue lift per session before asking a retailer to pay more than a month's subscription. That constraint keeps us honest.
Sydney is home, and AU/NZ is our first market. Three early UK-based retailers are in beta. We are not rushing into ten markets before the product is ready. The catalog-scale problem is consistent across geographies, but we address each new market carefully.
Talk to our team
We answer every genuine inquiry. If you have a large catalog and a sorting approach that is no longer scaling, we want to hear from you.