6 November 2026International edition
Vol. I · No.
6 November 2026
AI IN FASHION
DAILY
The daily briefing on AI in the fashion business
Where fashion meets artificial intelligence.
Commerce & Marketing · Stories

Returns cost fashion $743bn globally in 2025: how AI assistants reduce them

Fit guidance, size recommendations and styling advice powered by artificial intelligence are addressing the root causes of returns that drained three quarters of a trillion dollars from the industry last year.

brown wooden cross with ribbon
Photo: Nizzah Khusnunnisa / Unsplash

KEY TAKEAWAYS Summary by the editors

  1. Product returns cost the global fashion industry $743 billion in 2025, driven primarily by size and fit issues.
  2. AI assistants that provide fit guidance, size recommendations and styling advice target the leading causes of returns before checkout.
  3. Retailers using AI size recommendation tools report return rate reductions of 20 to 35 per cent within months of deployment.
  4. Implementation barriers include integration complexity, customer trust in AI advice, and the need for high-quality product data.

Product returns drained $743 billion from the global fashion industry in 2025, according to estimates from supply chain analytics firms, with size and fit problems accounting for the largest share of unwanted items sent back. The cost extends beyond refunds to include reverse logistics, inspection, restocking and markdowns on returned goods that can no longer be sold at full price.

Artificial intelligence assistants embedded in e-commerce platforms are now addressing these root causes directly. By analysing body measurements, purchase history, product specifications and return patterns, these systems deliver personalised fit guidance, size recommendations and styling advice at the point of sale. Early adopters report return rate reductions of 20 to 35 per cent, translating to tens of millions of dollars in saved costs for large retailers.

The technology arrives as European and North American regulators tighten rules on return fraud and environmental impact, raising the stakes for fashion companies to prevent returns rather than simply process them.

Why are fashion returns so costly?

Returns impose direct and indirect costs across the value chain. Direct costs include outbound and return shipping, warehouse labour for receiving and inspecting returned items, customer service time, payment processing fees and the original inventory carrying cost. Indirect costs stem from lost sales opportunities when stock sits in reverse logistics, markdowns required to sell returned merchandise, and disposal expenses for items too damaged or out of season to resell.

Online fashion returns run significantly higher than in-store purchases because customers cannot touch fabric, assess fit or compare colours in person before buying. Industry surveys consistently show that size and fit issues drive 60 to 70 per cent of fashion returns, followed by style or colour not matching expectations, quality concerns and ordering multiple sizes or styles with intent to return most items.

The logistics of reverse flow compound the expense. A returned item may travel from customer to regional depot to inspection facility to warehouse, incurring handling at each node. Items returned after 30 or 60 days may miss peak selling windows, forcing steeper discounts. Some estimates place the total cost of processing a return at 1.5 to 2 times the outbound shipping cost alone.

What causes most fashion returns?

Size and fit mismatches dominate return reasons across apparel and footwear categories. Standard sizing systems vary by brand, country and even product line within a single retailer, leaving customers to guess whether a medium in one label matches a medium in another. Body shape diversity means that two people with the same chest measurement may require different fits depending on shoulder width, torso length or waist-to-hip ratio.

Style and colour discrepancies follow as the second most common return driver. Photography, lighting and screen calibration can distort how a garment appears online compared to in hand. A deep navy may look black on some monitors, or a slim-fit trouser may appear relaxed in a styled photograph. Customer expectations shaped by marketing imagery sometimes diverge from the physical product.

Quality issues, including defects, damage in transit and materials that feel different from descriptions, account for a smaller but still significant share. A fabric described as cotton may contain synthetic blends that change drape or breathability. Finally, deliberate over-ordering, sometimes called bracketing, occurs when shoppers buy multiple sizes or colours intending to keep one and return the rest, using the retailer as a fitting room.

How do AI assistants reduce size and fit returns?

AI-powered fit recommendation engines compare customer body data with garment specifications to predict which size will fit best. Some systems ask shoppers to input height, weight and a few key measurements. Others infer body shape from past purchase and return behaviour, tracking which sizes a customer kept versus returned across brands and styles. Advanced implementations use computer vision to analyse photos customers upload, extracting dozens of measurements without manual input.

The algorithms account for brand-specific sizing variance by building statistical models of how each manufacturer cuts garments. A customer who wears a size 10 in one brand's dresses but a size 12 in another's receives tailored recommendations for each. Machine learning continuously refines predictions as more customers buy and return, improving accuracy over time.

Natural language processing enables conversational interfaces where customers describe fit preferences,"I like a loose fit around the waist",and the system adjusts recommendations accordingly. Some platforms integrate real-time chat, letting shoppers ask whether a specific jacket will accommodate broad shoulders or if a trouser runs long, with the AI pulling from product data and aggregated feedback to answer.

Size and fit issues drive most fashion returns
Return reasons
  • Size/fit problems65%
  • Style/colour mismatch20%
  • Quality issues10%
  • Other reasons5%

Source: Industry surveys, 2025

How does styling advice prevent returns?

AI styling assistants analyse product catalogues, trend data and individual customer preferences to suggest complete outfits or complementary items, reducing the likelihood that a purchase will disappoint when worn. By showing how a garment pairs with items the customer already owns or other pieces in the same order, these tools help shoppers visualise the final look before checkout.

Colour and pattern matching algorithms prevent clashes that lead to returns. If a customer adds a floral blouse to the basket, the assistant might warn that the pattern will compete with a striped skirt also in the cart, or suggest a solid-colour alternative. Seasonal and occasion-based recommendations ensure that weight, formality and style align with intended use.

Personalisation engines learn from browsing behaviour, wish lists and social media signals to refine suggestions. A customer who consistently views minimalist designs receives different styling advice than one drawn to bold prints. Over time, the system identifies style archetypes and predicts which recommendations will resonate, reducing exploratory purchases that often end in returns.

Which companies have deployed AI assistants for returns reduction?

Several major retailers and platforms have integrated AI-driven fit and styling tools into their e-commerce flows, publishing measurable outcomes. These implementations offer concrete evidence of impact, though results vary by category, customer base and execution quality.

One publicly traded European online fashion retailer reported in its 2025 annual report that deploying an AI size recommendation engine across its platform reduced return rates by 24 per cent year-on-year in categories where the tool was active. The company attributed cost savings of approximately €150 million to the initiative, calculated from lower reverse logistics expenses and higher full-price sell-through on items that were not returned.

A North American department store chain announced in early 2026 that a pilot programme using conversational AI styling assistants in its mobile app led to a 30 per cent drop in returns among users who engaged with the feature for at least three interactions before purchase. The retailer expanded the tool chain-wide in the second quarter of 2026, integrating it with loyalty programme data to personalise recommendations further.

An athletic footwear brand disclosed in an investor presentation in mid-2026 that its AI fit finder, which asks customers to photograph their feet and answers questions about arch support and toe box width, cut footwear returns by 18 per cent in markets where the feature launched. The brand plans to extend the technology to apparel sizing by the end of the year.

What are the technical requirements for AI fit and styling systems?

Effective AI assistants require rich, structured product data as inputs. Each item must include precise measurements,not just labelled size but actual garment dimensions such as chest width, sleeve length, inseam and rise. Fabric composition, stretch percentage, weight and care instructions feed into fit predictions. Photography from multiple angles and on diverse body types improves visual matching.

Customer data integration presents both opportunity and complexity. The system needs access to purchase history, return records, browsing behaviour and any measurements or preferences customers have shared. Privacy regulations in Europe and North America impose strict consent and storage requirements, so data pipelines must support granular permissions and anonymisation where personally identifiable information is not needed.

Real-time inference demands low-latency architecture. A shopper waiting more than two or three seconds for a fit recommendation may abandon the interaction. Cloud-based machine learning platforms can scale to handle traffic spikes during sales events, but integrating them with legacy e-commerce systems often requires middleware and API development. Frontend design must present AI advice clearly without overwhelming the purchase flow.

Data typeWhat AI needsCommon availability
Garment measurementsActual dimensions for every SKUModerate,often incomplete
Fabric propertiesStretch, weight, drape behaviourLow,rarely digitised
Customer body dataMeasurements, fit preferences, historyModerate,requires consent
Return reasonsStructured feedback on why items came backLow,mostly free text

What are the risks and limitations of AI return-prevention tools?

Customer trust remains a barrier to adoption. Shoppers accustomed to relying on size charts, reviews or trial and error may hesitate to follow AI advice, especially for high-value or special-occasion purchases. If a recommendation proves incorrect, the resulting return and negative experience can erode confidence in the tool and the retailer. Building trust requires transparency about how recommendations are generated and clear recourse when the system fails.

Data quality constrains accuracy. If product measurements are inconsistent or missing, or if fabric behaviour under wear is not captured, the AI cannot compensate for bad inputs. Brands that lack rigorous measurement standards across factories and suppliers will see poorer results. Similarly, sparse customer data for new or infrequent buyers limits personalisation, leaving the system to rely on population averages that may not fit individuals.

Bias in training data can lead to recommendations that favour certain body types or demographics. If the majority of return and purchase data comes from customers within a narrow size range, the model may underperform for those outside it. Ensuring diverse representation in training datasets and testing across segments is essential but resource intensive.

Legal and regulatory considerations add complexity. Some jurisdictions require disclosure when automated decision-making affects consumer transactions. If an AI assistant steers a customer away from a purchase or suggests a size that is out of stock, competition authorities may scrutinise whether the system unfairly disadvantages certain products or suppliers. Data protection officers must approve architectures that process personal information.

How do AI assistants compare with other return-reduction strategies?

Fashion companies employ multiple tactics to lower return rates, each with distinct cost, complexity and effectiveness profiles. AI assistants sit alongside improved photography, detailed size guides, customer reviews and return fees as part of a broader returns management programme.

High-quality product imagery and videos showing garments on diverse body types help set accurate expectations but do not provide personalised guidance. Augmented reality try-on features let customers visualise products on their own images, addressing some fit concerns, though technical limitations around fabric drape and body shape accuracy persist. Written size guides and measurement charts require customers to measure themselves accurately, which many find inconvenient or error-prone.

Return fees and restocking charges discourage frivolous returns but risk alienating customers and reducing conversion, especially in competitive markets where free returns have become expected. Customer reviews aggregate subjective opinions on fit and quality but vary in reliability and may not cover new or low-volume products. AI assistants offer scalability and personalisation that manual or static methods cannot match, though they require upfront investment and ongoing maintenance.

AI in fashion commerce milestones
  1. 2019First major retailers pilot AI size recommendation engines
  2. 2021Computer vision body scanning apps launch for consumer use
  3. 2023Large language models enable conversational styling assistants
  4. 2025Returns cost global fashion industry $743 billion
  5. 2026 Q2North American department store chain rolls out AI styling tool

Source: Industry reports and company announcements

How will AI assistants evolve in the next two years?

Advances in generative AI and computer vision are likely to enhance the sophistication and accessibility of fit and styling tools. Large language models already enable more natural, context-aware conversations, allowing customers to describe needs in everyday language rather than navigating menus or forms. As these models ingest more fashion-specific training data, their ability to interpret nuanced style preferences and fit concerns will improve.

Three-dimensional body scanning using smartphone cameras is becoming more accurate and faster, reducing friction in collecting customer measurements. Integration with smart mirrors and in-store kiosks may blend online and offline experiences, letting shoppers receive AI recommendations while browsing physical locations. Real-time inventory visibility combined with predictive analytics could enable assistants to suggest alternatives instantly when a preferred size is unavailable, keeping customers engaged rather than triggering returns.

Sustainability reporting requirements in Europe and proposed regulations in North America are pushing fashion companies to quantify and reduce waste, including returns that end in landfill. AI assistants that demonstrably cut return volumes will become not only a cost-saving tool but also a compliance and corporate responsibility asset, strengthening the business case for investment.

What should fashion companies do now?

  1. Audit product data quality. Conduct a systematic review of garment measurement accuracy, completeness and consistency across the catalogue. Identify gaps in fabric properties and size variance data, then establish governance processes to maintain standards as new products are added.
  2. Pilot AI tools in high-return categories. Select one or two product categories with the highest return rates and test AI fit or styling assistants with a subset of customers. Measure return rate changes, customer engagement and cost per implementation to build a business case before scaling.
  3. Integrate customer feedback loops. Design systems that capture structured return reasons and fit feedback, feeding this data back into AI models. Encourage customers to rate recommendation accuracy and provide comments that inform continuous improvement.
  4. Invest in transparency and trust-building. Clearly communicate how AI recommendations are generated, what data is used and how customers can override or ignore suggestions. Publish accuracy metrics and customer satisfaction scores to demonstrate value.
  5. Train customer-facing teams. Equip customer service, sales and merchandising staff to understand how AI assistants work, answer questions and troubleshoot issues. Ensure human support is available when automated tools fall short.
  6. Collaborate with suppliers on data. Work with manufacturers and brands to standardise measurement protocols and share fit data. Establish contractual requirements for product specifications that enable accurate AI recommendations.
  7. Monitor regulatory developments. Track evolving data protection, consumer rights and competition law in key markets. Engage legal counsel early to ensure AI implementations comply with disclosure, consent and fairness requirements.

Frequently asked questions

How much do returns cost the fashion industry globally?

Product returns cost the global fashion industry an estimated $743 billion in 2025, encompassing refunds, reverse logistics, inspection, restocking and markdowns on returned goods that can no longer be sold at full price.

What percentage of fashion returns are due to size and fit problems?

Industry surveys indicate that size and fit issues drive approximately 60 to 70 per cent of fashion returns, making them the leading cause of products being sent back.

By how much can AI assistants reduce return rates?

Retailers using AI fit recommendation and styling tools report return rate reductions of 20 to 35 per cent in categories and customer segments where the technology is deployed, though results vary by implementation quality and product data accuracy.

Stories