On this page
- Why it matters
- The scale of the fit-returns problem
- The agentic-specific failure mode
- Key terms
- Vendor landscape (as-of 2026-09)
- How agentic sizing works
- Technical integration (MCP layer)
- Advanced garment modelling
- Review-derived fit signals
- Platform infrastructure
- Benchmarks (as-of 2026-09-07)
- Returns reduction by approach
- Market size
- What practitioners report
- Regulatory tailwinds
- Frontier links
Agentic Sizing
Agentic Sizing
A class of fit-recommendation capability where an AI shopping agent resolves clothing or footwear sizing on behalf of — or in direct service of — a shopper, either within a conversational interface or as part of an autonomous purchase flow. The term was named by Bold Metrics as a distinct product category (Agentic Sizing Protocol) in early 2026 and adopted by multiple vendors and analysts in the same period.
The defining characteristic that separates agentic sizing from first-generation fit-tech (static size charts, survey-based recommenders) is machine-readability: for an AI agent completing a purchase autonomously, fit data must be queryable via API or Model Context Protocol (MCP) — a human-readable size chart on a PDP is invisible to the agent. (True Fit BusinessWire, 2026-02-17; YourSizer, 2026-07-29)
Why it matters
The scale of the fit-returns problem
Apparel and footwear carry the highest return rates of any ecommerce vertical. NRF and Happy Returns reported US retail returns reached $890 billion in 2024, with online return rates near 19–20%; apparel return rates run between 24% and 40% by category (as-of 2026-04-27, citing 2024 data). (Size AI, 2026-04-27)
The $890B NRF figure reflects 2024 data. The 2025 and 2026 equivalents had not been published at time of harvest (2026-09-07).
True Fit stated that with nearly $850 billion in projected returns for 2025 and almost one in five online purchases coming back, retailers need an agent built on structured fit data to solve the problem at its source (as-of 2026-02-17). (True Fit BusinessWire, 2026-02-17)
Fit-returns share: Coresight Research (2023, n=100 US apparel decision-makers) puts fit-driven returns at 53%. McKinsey is cited by Size AI at "closer to 70%". Bold Metrics cites industry data at 67%. Tredence cites 52%. All figures circulate in the same conversation without a single reconciled source; variation likely reflects category mix, survey methodology, and whether bracketing returns are included. (Size AI, 2026-04-27; Bold Metrics, 2026-02-17; Tredence, 2026-06-29)
The Coresight 53% figure is from a 2023 survey. McKinsey and Bold Metrics do not provide primary publication dates for their figures.
The agentic-specific failure mode
Without access to structured fit data, an autonomous shopping agent has a clear failure mode: order two or three sizes and let the human sort it out on delivery. This scales Bracketing (Fashion Returns) at machine speed and machine volume — a materially different cost profile than occasional human bracketing. (YourSizer, 2026-07-29)
True Fit, in its analysis of real customer conversations with AI shopping agents, found that up to 70% of fashion-related questions directed at AI shopping agents were about fit and sizing (as-of 2026-02-17). (True Fit BusinessWire, 2026-02-17)
True Fit described uncertainty around fit and sizing as the number one reason for purchase hesitancy among fashion shoppers, citing Mintel. (True Fit BusinessWire, 2026-02-17)
Key terms
| Term | Meaning |
|---|---|
| Agentic Sizing Protocol | Bold Metrics' named product for exposing fit data to AI agents via machine-queryable interface |
| Fit Intelligence Layer | True Fit's API/MCP endpoint that shopping agents, copilots, and merchant personalisation systems can query for real-time fit recommendations |
| Living Style Profile | Continuously updated record of shopper preferences, body data, purchase history, browsing behaviour, and contextual signals; described by Tredence as the data foundation for every agentic commerce interaction |
| Shop-by-fit | Discovery model upstream of the PDP where shoppers reference a garment they own and the catalog surfaces all dimensional matches |
| Generative Engine Optimization (GEO) | Catalog optimisation for machine-readable discovery by AI agents; differs from traditional SEO ranking logic |
Vendor landscape (as-of 2026-09)
| Vendor | Positioning | Agentic capability |
|---|---|---|
| True Fit | "First AI shopping agent built on 20 years of real fit data"; Fashion Genome covers 82M shoppers, 29K+ brands, $616B+ in transactions, 91K+ brands | Fit Intelligence Layer via MCP; available from April 2026 |
| Bold Metrics | Agentic Sizing Protocol as named product | Machine-queryable fit API alongside Smart Size Chart, Virtual Sizer, Virtual Tailor |
| Size AI | LiDAR-based per-SKU garment measurement | 200+ data points per garment in 0.92s / iPhone; ~5mm accuracy; 15 user-facing measurements per item |
| YourSizer | Fit schema for machine-queryable size data | Focuses on converting human-readable size charts into agent-callable fit schemas |
| Fit Analytics | Decoupled from Snap ~May 2025; independent again | Historical body measurement approach; current agentic positioning not confirmed from primary source |
| 3DLOOK | Body measurement and virtual try-on | Named by Tredence as major provider; primary agentic positioning not confirmed |
| Virtusize | Garment-to-garment comparison (shopper owns an item → compare to new item) | Named by Tredence; primary agentic positioning not confirmed |
Tredence (2026-06-29) lists True Fit, Fit Analytics, 3DLOOK, Virtusize, and Bold Metrics as major providers in the broader fit-tech space. (Tredence, 2026-06-29)
Named smaller/newer providers in 2026 include YourSizer, SIXFIT AI, Visoryx, SAIZ, Size AI, Sizekick, Popsize. Status of these providers not independently verified from primary sources.
How agentic sizing works
Technical integration (MCP layer)
True Fit's Fit Intelligence is available via Model Context Protocol (MCP), allowing shopping agents, copilots, and merchant personalisation systems to query real-time fit and size recommendations, eliminating doubt at the point of decision. (True Fit BusinessWire, 2026-02-17)
YourSizer argued that brands treating their size chart as a "static, low-priority asset" are most exposed to agent-driven mis-sizing, distinguishing a human-readable size chart from a machine-queryable fit schema an agent can call programmatically. (YourSizer, 2026-07-29)
Advanced garment modelling
Tredence described advanced agentic sizing systems that simulate how a specific fabric (e.g. ponte knit vs woven satin) behaves on a 3D body model constructed from the shopper's measurements, and that predictive fit scoring can flag when a shopper's proportions suggest sizing up even if their standard size in that brand has historically fit. (Tredence, 2026-06-29)
Review-derived fit signals
Bloomreach's Loomi Shopping Agent can analyze customer reviews to determine that a garment "runs large" and automatically recommend the shopper size down before they add the item to cart. (Bloomreach Blog, 2026-04-20)
Baymard Institute's quantitative survey study of 1,922 US online apparel shoppers (2026-06-12) found that 48% of shoppers use reviews primarily to resolve fit uncertainty — ahead of quality/durability. Size accuracy was the number-one thing sought in reviews. (Baymard gated; cited at baymard.com/blog/apparel-and-accessories-quantitative-ux-insights-2026)
Platform infrastructure
Shopify launched a dedicated "Agentic Storefronts" admin page (May 2026) that automatically makes merchant product catalogs — including size/variant data — accessible to external AI channels including ChatGPT, Microsoft Copilot, and the Shop app via Shopify Catalog, with performance tracking and product data improvement recommendations built in. (Shopify Changelog, 2026-05-11)
Tredence stated that for AI agents to reliably recommend and sell products, fashion brand catalogs require comprehensive attribute coverage (fabric, construction, sizing notes), semantic tags for occasion-based discovery, verified sustainability claims, and real-time inventory/pricing APIs — describing this as Generative Engine Optimization (GEO). (Tredence, 2026-06-29)
Algolia's Agent Studio enables fashion brands to build conversational AI assistants capable of answering size-specific queries such as "Will these jeans be too long if I'm 5 foot 1?" and redirecting the shopper to a better-fitting alternative if the answer is yes. (Algolia Blog, 2025-12-18)
Algolia blog post is dated 2025-12-18. Agent Studio capabilities may have changed since.
Benchmarks (as-of 2026-09-07)
Returns reduction by approach
Size AI compiled the following benchmark table based on its own analysis. These are vendor-reported figures; no third-party audit was cited:
| Approach | Returns reduction (vendor-reported) |
|---|---|
| Traditional size chart | 0–5% |
| ML behaviour recommender (M/L style) | 15–24% |
| Survey-based | 3–7% |
| Body-scan + brand size chart | 5–12% |
| Shop-by-fit discovery model (PDP-level) | 15–24% |
| Shop-by-fit discovery model (discovery stage) | 25–40% |
| Per-SKU garment measurement (Size AI claimed) | 77% (24.4% → 5.5% fit-related return rate across 1M+ garment captures) |
First-gen fit-tech efficacy: Size AI states the industry-wide promised 30–40% reduction from first-generation fit-tech delivered only 2–8% in real-world deployment, attributing this to cold-start blindness on new SKUs, survey friction, and outdated size charts. The same source's benchmark table shows ML recommenders at 15–24%. The 2–8% reflects actual deployment outcomes vs. the 15–24% upper-bound lab result; Size AI's article does not make this distinction explicit. (Size AI, 2026-04-27)
Bold Metrics' Helly Hansen case study: Smart Size Chart users in North America achieved +1.8× conversion lift and +$50.47 AOV increase; European users achieved +3.8× conversion lift and +$65.36 AOV increase, with approximately 18–19% of purchases using the tool (as-of 2026-02-17). (Bold Metrics, 2026-02-17)
Bloomreach reported Isadore (athletics apparel brand) using Loomi Shopping Agent's proactive sizing/fit guidance achieved a 29% reduction in potential returns (as-of 2026-04-20). (Bloomreach Blog, 2026-04-20)
Market size
The e-commerce apparel market is projected to reach $808.62 billion in 2026, growing at 8.8% annually (as-of 2026-02-17), citing ResearchNester. (Bold Metrics, 2026-02-17)
Virtual try-on market size (2025): Bloomreach (citing Business Research Company + eMarketer) places the virtual try-on market at $12.09 billion in 2025 projected to $38.92B by 2030 at 26.3% CAGR. Mordor Intelligence places the virtual fitting room market at $8.21 billion in 2025 / $9.81 billion in 2026. Both figures likely reflect different market scope definitions (VFR only vs VFR plus AI try-on). Neither figure was fetched from its primary report page. (Bloomreach Blog, 2026-04-20; Size AI citing Mordor Intelligence, 2026-04-27)
What practitioners report
YourSizer observed that Google folded a "Try it on" visual experience directly into Search in 2026 while winding down its standalone Doppl try-on app, interpreting this as a signal that fit and visualization are consolidating into discovery and purchase surfaces rather than standalone tools. (YourSizer, 2026-07-29)
Tredence cited Google's rollout of AI-powered virtual try-on features across billions of apparel listings as repositioning try-on "as a styling tool rather than a novelty," placing it within the agent conversation rather than as a standalone module. (Tredence, 2026-06-29)
Bloomreach (2026-06-10) lists Manifest AI as a leading AI shopping assistant for mid-market Shopify brands, offering a dedicated "fit predictor" agent as one of 500+ prebuilt specialized AI agents that brands can deploy without a custom engineering project. (Bloomreach Blog, 2026-06-10)
Nosto's 2026 ecommerce trends content identifies "reducing purchase anxiety through sizing clarity and reassurance messaging" as a key conversion strategy for first-time buyers, with AI recommended to adjust size recommendations post-return (when return reason is "fit") to feed back into pre-purchase personalisation. (Nosto Blog, 2026)
Regulatory tailwinds
YourSizer stated the EU's Ecodesign for Sustainable Products Regulation (ESPR) introduces a prohibition on destroying unsold apparel and footwear for large companies taking effect around mid-2026 (European Commission source), coinciding with the EU removing its €150 customs de minimis exemption and adding a per-parcel handling fee — making cross-border fashion returns structurally more expensive at the same time as agent-driven order volume grows. (YourSizer, 2026-07-29)
Around 7 in 10 US retailers already charge for at least some returns (as-of 2026-07-29); YourSizer noted Zara and ASOS charge in the $4–5 range per return, but a fee on the shopper does not offset reverse-logistics, restocking, and inventory-drag costs carried by the brand. (YourSizer, 2026-07-29)
Frontier links
Virtual Try-On · Generative Engine Optimization (GEO) · Living Style Profile · Model Context Protocol (MCP) · Agentic Sizing Protocol · Fashion Returns · Footwear Sizing · Catalog API