Prefixbox analyzed all agent conversations from a 50-day period in Spring 2026 to measure the impact.
Most incoming questions fell into predictable categories: "Where's my order?", "Can I return this?", "My candle arrived broken", "Is there a discount code?" Answering these individually, around the clock, was eating into the team's capacity for higher-value work.
At the same time, shoppers browsing the site often had questions about scents, product compatibility, or what to buy for a specific occasion. Without real-time help, those conversations simply weren't happening.
Types of Conversations
The agent handled 4 distinct types of conversations:
Product recommendation (38.9%) was the largest single category. In these conversations, the agent recommended specific products based on the shopper's preferences, from scent profiles to candle types. 41.6% of these conversations resulted in the shopper clicking through to a product page.
Support (29.1%) covered returns and refunds, damaged or missing items, shipping questions, billing issues, and account management. Order tracking alone accounted for 24% of all support conversations, with the agent handling over 600 order lookups since its introduction.
Discounts and promotions (17%) included questions about coupon codes, active sales, and checkout issues with discount application.
Product info (12%) grouped together scent advice, product care questions (burn time, wick trimming, wax type, pet safety), and availability or restock inquiries.
Support Automation
The agent handled around 6,000 support conversations over the 50-day period. At an estimated 3 minutes per message for a human agent to handle:
- 601 support hours reclaimed per month
- 20 hours of support time saved per day
- Equivalent to 2.5 full-time support employees
Product Discovery and Projected Revenue
Out of all conversations where products were shown:
- 41.6% resulted in at least one product click (Conversation-level CTR)
- 16.9% of individual products recommended were clicked (product-level CTR)
The difference reflects that the agent shows 2 products per conversation — while shoppers clicked in 4 out of 10 conversations, they were selective about which product they chose.
Key Takeaways
Even though the retailer was initially looking for a way to automate their support, the agent helped in product discovery as well: nearly 39% of all conversations were product discovery related.
Shoppers already expect conversational product discovery capabilities and use AI Agents as a way to find products, not just for addressing support questions.
On the support side, the agent absorbs the equivalent of 2.5 full-time employee's workload. Most of these conversations happen outside business hours, meaning questions that previously waited until the next morning now get resolved instantly.
The 41.6% conversation-level click-through rate on product recommendations is a strong engagement signal. When the agent suggests products, nearly half of shoppers click through to a product page. For this US Home & Decor on Shopify, Prefixbox's AI Agent isn't only automating support, it's also a powerful sales associate, available 24/7.