Walks through the Discovery Agent’s ranking, scoring, and recommendation capabilities alongside the Admin configurations that drive them — showing how merchandising teams adjust ranking weights, slot rules, seasonal boosts, concern taxonomy, and brand voice, and how each change is reflected immediately in the ACI Chat experience and its Reasoning panel.

• Ranking & Scoring Configuration – Defines how products are ranked using keyword match, personalization, fit, ratings and reviews, business value, and freshness. Adjusting these weights during the demo shifts the top ranking driver from relevance to social proof and then to margin, changing which product surfaces as the Hero recommendation each time.
• Slot Rules – Control the maximum number of Hero, Nudge, Complement, and Bundle products shown in the Reasoning panel. The walkthrough starts with three Hero, one Nudge, and one Complement product configured, then reduces Hero to one and shows the recommendation set re-composing accordingly.
• Seasonal Rules – Define available seasons, their boosted product categories, and the active months for each. With August falling inside the holiday season and Necklace configured as a boosted category, necklace products receive a seasonal boost that is called out directly in the Reasoning panel.
• Concern Taxonomy – Captures customer concerns such as breathability that influence recommendations whenever related phrases appear in ACI Chat, with associated ingredients available in the reasoning framework. In the demo, a comfort concern is identified from the taxonomy and reflected in the Reasoning panel alongside the ranked results.
• Brand Voice – Controls how ACI communicates with shoppers across four available tone options. Switching from direct and concise to friendly and conversational changes the presentation of the same recommendations without altering the ranking logic behind them.
• Knowledge Base & PDP Injection – Catalog data is imported and managed in the knowledge base across multiple configurable domains, while PDP Injection captures detailed product information including ingredients and product-specific attributes. Extracted concerns and ingredients are routed automatically to a review queue where reviewers approve or reject them, and approved concerns flow back into the concern taxonomy.
• Safety Guardrails – Displays the configured hard and soft guardrails governing agent behaviour before the demo moves into the chat experience.
• Discovery in ACI Chat – Prompts such as a comfortable and stylish dress for the season, a wedding-gift necklace under $200, highly rated dresses, and good quality dresses are routed to the Discovery Agent, with the Reasoning panel exposing the top ranking driver, identified intent, profile signals, query decomposition, the generated knowledge base query, per-product scoring breakdown, applied slot rules, and filters and exclusions showing catalog totals versus retrieved and ranked counts. Ambiguous requests trigger a clarification question before results are returned.