AI Ecommerce Agent for a Fashion Retailer: 22% Increase in Average Order Value

Background

A mid-size direct-to-consumer fashion brand launched in 2019 had grown to $15M in annual revenue by focusing on elevated basics with a distinctive design sensibility. The brand served a primarily female customer base aged 28–45, with an average order value of $86 and roughly 30,000 active customers. Margins were healthy on paper but eroded dramatically by returns—fashion ecommerce's perennial nemesis. By late 2024, the CFO had calculated that returns cost the brand the equivalent of 18% of gross margin, a figure that forced the executive team to rethink their entire post-purchase strategy.

Challenge

A mid-size D2C fashion brand generating $15M in annual revenue was battling a 32% return rate—nearly double the industry average—driven almost entirely by sizing issues. Customers ordered multiple sizes of the same item and returned what didn't fit, inflating fulfillment costs and tying up inventory. The brand's cross-sell conversion sat at a dismal 2.4%, well below the 7-8% benchmark for fashion ecommerce, because product recommendations relied on static rules that a merchandising coordinator updated manually. Changing a single recommendation rule—say, pairing a new jacket with complementary accessories—required updating spreadsheets, syncing with Shopify, and QA testing, a process that took 2-3 days per change. During seasonal launches, the merchandising backlog meant new arrivals often went live with no cross-sell pairings at all, leaving revenue on the table during the brand's highest-traffic windows.

Solution

The brand deployed an AI ecommerce agent integrated directly with their Shopify storefront across three capabilities:

Personalized product recommendations. The agent analyzed each visitor's browsing behavior, purchase history, and real-time cart contents to surface recommendations that went beyond "customers also bought." It identified style affinities—a customer who gravitated toward oversized silhouettes and neutral tones received different suggestions than one who favored fitted cuts and bold prints—and dynamically assembled outfit-level pairings rather than individual product pushes.

AI size advisor. Using the brand's return data (which items were returned, in which sizes, and for what reason) combined with the customer's own purchase and return history, the agent recommended the best size for each item. For new customers without history, it asked three quick questions about height, preferred fit, and a reference garment from another brand to generate a size recommendation with a confidence score.

Dynamic bundling. The agent created personalized bundles at checkout—"Complete the look" suggestions with a small discount—assembled in real time based on what was already in the cart, current inventory levels, and margin targets set by the merchandising team.

Merchandising rules that previously took days to update were now generated automatically and refreshed in minutes as new products were added or inventory shifted.

Implementation timeline

• Weeks 1–2: Shopify integration and product catalog mapping. The agent ingested 2 years of purchase history, return reasons, and size conversion data.
• Weeks 3–4: Size advisor model training. The team added a one-question return survey ("Why are you returning this item?") to capture the signal that the size model needed.
• Weeks 5–6: Shadow mode on recommendations and bundles. The team A/B tested AI-generated vs. rule-based recommendations on traffic segments.
• Week 7: Full production. Merchandising coordinator shifted from rule-maintenance to rule-tuning role, working with the AI to optimize edge cases.

Results

• Average order value: +22%, from $86 to $105
• Return rate: Reduced from 32% to 20.8% (a 35% decrease), saving an estimated $1.2M annually in reverse logistics and restocking costs
• Cross-sell conversion: Tripled from 2.4% to 7.3%
• Merchandising update time: From 2-3 days per rule change to under 5 minutes for a full catalog refresh
• Size-related support tickets: Down 48%, freeing two part-time support agents to focus on post-purchase experience
• Bundle attach rate: 18% of orders included a dynamically generated bundle

Lessons learned

• Return-reason data was the unlock. Without understanding why items were returned, the size advisor would have been generic. The one-question return survey cost almost nothing to implement and transformed model accuracy.
• Outfit-level recommendations beat item-level. Customers who received outfit-level recommendations (top + bottom + accessory pairings) added 2.3 items to cart on average vs. 1.4 for item-level recommendations.
• Confidence scoring on size recommendations built trust. Displaying "95% confident" vs. "likely fit" on the size recommendation increased conversion on recommended sizes by 18%.
• Discounting wasn't the main driver on bundles. The 5% bundle discount was small; the personalization and "complete the look" framing drove most of the attach rate.

Takeaway

The return rate reduction delivered the largest financial impact—more than the AOV increase—because sizing-related returns had been silently draining margin through shipping, restocking, and inventory carrying costs. The AI size advisor worked best when it had return-reason data, not just return volume, which required the brand to add a one-question return survey ("Why are you returning this item?"). That small data investment unlocked the accuracy needed to move the needle. For brands evaluating similar implementations, the lesson is that recommendation engines and size tools amplify each other: better size confidence increases willingness to add items, and better recommendations increase basket size without inflating returns. Explore the full capability set at AI Ecommerce Agent. To evaluate platforms and estimate ROI, visit Solutions.