'Silent killers': How AI start-ups are trying to solve one of the retail industry's biggest problems

15.8% of U.S. retail sales were returned in 2025 ($849.9bn) and online returns were 19.3%, a major drag on margins. AI virtual-try-on startups (e.g., Catches) claim ~10% conversion uplift and 20–30x ROI for brands, while ASOS reported a 160bps reduction in its returns rate contributing to improved profitability; major players (Zara/Inditex, Shopify, Google, Amazon) are rolling out or integrating similar tools alongside policy levers (return fees) to protect margins.

Analysis

Realistic virtual try-on that meaningfully lowers returns is an earnings driver disguised as a margin story: a 2–3 percentage-point fall in online return rates typically converts into ~50–200 bps of gross-margin expansion for apparel retailers within 6–12 months because reverse logistics and restocking consume outsized operating dollars per returned unit. That improvement is front-loaded to cash flow (lower shipping/processing, fewer disposals) and can free working capital by reducing inventory churn and markdown pressure, materially improving ROIC for brands with tight wholesale relationships.

The capital and services chain shifts are the key second-order effects. Expect persistent incremental demand for low-latency GPU inference (benefiting GPU vendors and cloud providers) and for professional capture rigs/3D asset pipelines (photogrammetry studios, middleware). Conversely, providers whose business model relies on high reverse-logistics throughput — third-party returns warehousing, cross-border couriers focused on reverse flows — could see volume declines that compress their growth multiple over 12–36 months.

Near-term catalysts are concrete: merchant integrations (platform-level installs), early conversion lift data from pilots, and rollout into large marketplaces — each can show measurable revenue/GM lift in sequential quarters. Tail risks that flip the thesis include privacy/regulatory constraints on body-scan biometrics, accuracy plateaus that disappoint conversion expectations, or consumer pushback on data collection; any of these could keep the net effect to incremental merchandising improvements rather than a structural margin shift.

The crowd’s bullishness on “AI solves returns” understates concentration risk. Luxury and mid-market premium brands get the highest ROI per customer because per-order economics justify heavy compute; mass-market fast-fashion economics cap how much brands will invest per SKU. That means public winners will be platform/infrastructure providers and select software integrators, not every listed retailer — strategy should therefore tilt toward enablers of the tech stack rather than downstream merchants indiscriminately.

Key Decisions for Investors

• Long NVDA (stock or 3–6 month call spread): target entry on any pullback of 5–10%. Thesis: sustained uptick in GPU inference demand from photoreal try-on drives incremental data-center utilization and partner certifications. Position size: 1–2% portfolio; upside: 30–60% IF enterprise adoption grows; downside: limited to premium paid on options or ~15% stop on stock.
• Long SHOP (6–12 month timeframe): buy into dips or purchase 12-month calls. Shopify is the fastest route for SMB/brand adoption of turn-key try-on apps; incremental ARPU and app-marketplace monetization can expand margins. Position size: 1–2% portfolio; risk/reward ~2.5:1 assuming 20–30% improvement in merchant take-rates vs 15% downside if marketplace monetization stalls.
• Long GOOGL (12-month call or stock): allocate modestly to capture search-integrated try-on adoption and higher-intent shopping queries monetization. Entry on weakness or after each major merchant integration PR; expect steady ad-ARPU lift. Position size: 1–2%; reward asymmetric if Google converts search sessions to purchases, downside tied to ad cyclicality ~10–15%.
• Pair: Long SHOP / Short AMZN (small size, equal notional, 3–9 months): shopify benefits from frictionless app installs while Amazon’s scale and logistics exposure make it a slower beneficiary of boutique try-on tech. Keep pair size limited (net market neutral) and cap drawdown at 8% per leg; target 20–40% relative outperformance over 6–9 months if platform adoption accelerates.