Building on the rapid success of Adaptive Transformer Search (ATS), which virtually eliminated zero-result searches, Particular Audience now enables retailers to A/B test the AI models that determine relevance, margin, and retail media yield.
4th March, London, UK – Particular Audience today announced the launch of Search Model A/B Testing, a new capability that allows retailers to directly compare and govern the AI models driving on-site search and sponsored results.
This launch builds on the success of Adaptive Transformer Search (ATS), first introduced in 2023 to solve the $300bn product discovery problem. While the industry average for zero-result searches remains close to 20%, ATS clients consistently operate below 0.5%, proving that modern AI can reliably understand long-tail and conversational intent.
1% gains in relevance can mean hundreds of thousands of dollars in recovered revenue, with AI’s rapid evolution, the threat facing retailers is standing still.
From Finding Keywords to Capturing Intent
The majority of search technology in 2026 remains keyword-first. Even platforms claiming to use AI often rely on vectors only for early recall, before reverting to rules-based ranking. This works for simple queries, but breaks down when intent is nuanced, seasonal, or contextual.
True relevance—understanding what a customer means and selecting the right product in that moment—requires a different approach.
ATS uses a single relevance model to evaluate intent, product suitability, and commercial context (for a retailer’s KPIs) together. Search Model A/B Testing extends this foundation by unlocking componentry and allowing retailers to test different relevance strategies directly, rather than accepting a vendor’s default behavior.
For the first time the debate at a retailer shifts from “which search vendor is the best”, to “which model is best for us right now, and how might we constantly prove it?”.
At the same time, most multi-brand retail platforms still operate two separate systems: one for organic search and another for sponsored products. These systems are typically stitched together late in the process using rules or rank overrides. That approach may blend results visually, but it does not unify relevance and put the customer experience first.
Why A/B Test Search Models?
Because relevance is not static.
A retailer may want:
- Faster models for high-traffic promotional events
- Margin-weighted models during peak trading periods
- Brand-safe models for premium categories
- Multilingual models for cross-border expansion
Each of these goals involves trade-offs between speed, recall, margin, and localization. Search Model A/B Testing makes those trade-offs measurable and selectable.
Retailers can now test different model architectures side-by-side and choose the one that best aligns with their commercial strategy—using clear metrics such as conversion rate, average order value, margin contribution, and zero-result rate.
Particular Audience’s latest release allows a retailer to configure models based on unique sentence embedding structures, distinct foundational LLMs, and select reinforcement learning data—both genuine and synthetic—to tune their models. Expert customer support is 24/7 and takes the technical sting out of the tail making model manipulation just a request away.
Implications for Retail Media Networks
For Retail Media Networks, this shift only applies to the leading networks that have evolved from legacy keyword targeting approaches.
ATS turns previously unanswerable, long-tail queries into valid result pages, creating new monetizable inventory. Model-level testing then allows retailers to optimize how sponsored products are surfaced within true relevance—not as interruptions, but as contextually appropriate outcomes. Particular Audience’s Search Model A/B Testing allows yield optimization for Retail Media Networks for the very first time.
This moves retail media from placement optimization (keyword bidding) to intent-level optimization (automating high-yield targeting), improving both advertiser performance and network yield.
ATS unlocks the entire grid representing a force multiplier on sponsored search revenue.
From Black Box to Glass Box
Most vendors offer a single, opaque algorithm that decides what is relevant. Particular Audience exposes relevance as a governed decisioning layer.
“AI driven discovery is now table stakes,” said James Taylor, Founder & CEO at Particular Audience. “What matters is how relevance behaves under different business constraints, and how retailers can ensure their trust in their search model(s) is constantly positively reinforced. ATS ensures customers always find something. Model A/B Testing ensures retailers choose what kind of relevance best serves their strategy.”
About Particular Audience
Particular Audience is a retail AI company helping retailers and retail media networks compete through superior discovery, relevance, and decisioning. Its flagship product, Adaptive Transformer Search (ATS), has set a new standard for handling conversational and long-tail commerce queries, virtually eliminating zero-result searches.
PA’s platform unifies search, merchandising, and retail media within a single relevance framework—giving retailers control, transparency, and measurable outcomes.