{"slug": "how-garcia-got-its-ai-try-on-to-lift-conversion-7", "title": "How GARCIA got its AI try-on to lift conversion 7%", "summary": "Faslet reports that GARCIA's AI virtual try-on layer lifted conversion 7.06% and cut returns 5.54% in a live-traffic test where both arms already used Faslet's size-me tool, with more than 5,000 ordered items per group. Faslet did not publish visitors per arm, test dates, baseline rates, randomisation details or statistical significance, and the percentages appear to be relative changes rather than percentage-point movements. The test isolated the visual fitting-room layer from size advice, asking whether seeing a garment on a body like her own adds commercial value once a shopper knows her size.", "body_md": "## Experiment summary\n\n[GARCIA](https://www.wearegarcia.com/about-us) ran a virtual try-on test that layered Faslet’s AI fitting room on top of the sizing tool both groups already had. Faslet reports 7.06% higher conversion and 5.54% fewer returns. Visitor counts and statistical uncertainty have not been published.\n\n- GARCIA tested an AI virtual fitting room against its normal product journey on live webshop traffic.\n- Both groups already had Faslet’s size-me tool, so the experiment measured the extra visual try-on layer on its own.\n- Faslet says each group produced more than 5,000 ordered items and reports 7.06% higher conversion plus 5.54% fewer returns for the treatment.\n- The public case does not disclose visitors per arm, test dates, baseline rates, randomisation, statistical significance, return-maturation time, implementation cost or market split.\n- The useful operating idea is to test visual confidence separately from size advice and judge the layer on conversion and kept-order economics together.\n\nIn Faslet’s account of the test, a GARCIA shopper could choose a garment and ask to see it on a body like her own. The product page supplied the clothing image. The shopper could upload or take a photo, and an AI system produced the visualisation. Size advice remained underneath the experience.\n\nBecause both groups already had the same size-me tool, the test asked a narrower question than most try-on pilots: once a shopper knows her size, does seeing the garment on a body like hers add anything commercially?\n\n## The test isolated the picture from the size\n\nGARCIA began in Italy and grew in the Netherlands. Its parent, JOG Group, says the brand now has more than 2,500 points of sale across Europe and North America. For an apparel business at that scale, online uncertainty has two separate parts. A shopper can wonder which size to order and still wonder whether the cut, colour or shape will suit her.\n\n[Faslet describes](https://faslet.me/cases/business-case-garcia/) a live-traffic split between shoppers who could access virtual try-on and shoppers who followed the standard journey. Both arms kept Faslet’s size recommendation feature. According to Faslet, the only switched variable was access to the virtual fitting room.\n\nThe public case says more than 5,000 items were ordered in each group. That is useful scale for the post-purchase measure, but it is not the experiment’s full denominator. Conversion needs the number of eligible or exposed visitors in each arm, and Faslet does not publish it. Nor does it say whether assignment was random, whether a returning visitor stayed in the same arm, or whether concurrent promotions and experiments were balanced.\n\nThe intervention itself is relatively bounded. Faslet’s [technical documentation](https://docs.faslet.net/docs/virtual-try-on/) says a merchant supplies a shop ID, brand, product ID and product image. An existing customer can add the virtual try-on capability without a separate installation. The shopper can upload or take a photo, while the product image provides the garment to render.\n\n## One result arrived before the other\n\nFaslet reports that the treatment converted 7.06% better and produced a return rate 5.54% lower than the control. It also says 80% of shoppers who rated the experience gave it a positive response.\n\nThose percentages appear to be relative changes, not percentage-point movements, although the case does not state that explicitly. Without absolute rates, an editor cannot reconstruct the number of extra orders or avoided returns. A move from 5.0% to 5.35% conversion is a 7% relative lift. A move from 20% to 18.9% returns is a 5.5% relative reduction. Both examples are arithmetic illustrations, not GARCIA’s unpublished baselines.\n\nConversion can be read as soon as an order is placed. Returns need time. The case does not say how long the team waited for returns to mature, whether exchanges were counted, or whether the calculation used returned items, returned orders or customers. That missing window is especially important when the sample is described in ordered items rather than people.\n\nFaslet supplied the product, ran or supported the test and published the result. GARCIA has not released the analysis independently. The public record includes no confidence interval, significance level, pre-registered stopping rule, device split, country split or raw count of returns. A controlled comparison on live orders still tells a team more than click or engagement data would.\n\n## What AI virtual try-on can and cannot show\n\nFaslet’s product page says the system can work from a shopper photo and the merchant’s product image. It generates a view of the garment on a matching body. That can answer a styling question quickly. It cannot prove how the fabric will stretch, drape or feel after delivery.\n\nA 2026 [CHI study](https://doi.org/10.1145/3772318.3791811) helps draw that boundary. Researchers randomly assigned 24 shoppers to a journey with or without virtual try-on and followed the purchases after delivery. The tool changed exploration and expectations, but both groups returned the same number of items. Participants also split on whether the rendered fit resembled the garment they received. The sample was small and Korean-only, so it does not overturn GARCIA’s field result. It shows why a visual tool still needs a post-purchase guardrail.\n\nEarlier randomised field experiments on [virtual fit information](https://pubsonline.informs.org/doi/10.1287/msom.2017.0686) found higher conversion and lower fulfilment costs from returns and home try-on behaviour. That evidence concerns fit guidance rather than a generated image. GARCIA’s design raises the bar by keeping sizing help active in both arms and testing only the extra visual layer.\n\n## The margin test is still missing\n\nFor GARCIA’s ecommerce team, shopper enjoyment is the easy part. The layer earns its place if extra contribution from converted and retained orders exceeds model inference, integration, support and image-processing costs.\n\nIn a separate JOG Group ecommerce discussion, Head of E-commerce and Marketplaces [Mark Vliegenthart](https://nl.linkedin.com/in/markvliegenthart) described the broader operating choice as a balance between conversion, cost and service to the customer. He was speaking about ecommerce generally, though the same three factors are what this test still needs to report.\n\nFaslet does not publish GARCIA’s price, implementation time, generation latency or the share of eligible shoppers who completed a try-on. Its documentation says only that an existing integration needs minimal additional configuration. The [data terms](https://faslet.me/data-processing-terms/) add another operating constraint: uploaded images are processed for the requested visualisation and deleted within 72 hours, while the merchant remains responsible for the legal basis and shopper notice.\n\nA complete readout would calculate contribution margin per exposed visitor. Revenue from extra orders sits on one side. Cost of goods, returns, discounts, payment fees and inference sit on the other. If virtual try-on attracts more low-margin orders or slows the product page, the conversion headline can overstate the gain.\n\n## The next test needs a mature return window\n\nThe most useful follow-up keeps the same clean boundary. Randomise eligible visitors, preserve assignment across sessions and leave size advice unchanged in both arms. Publish visitors, orders and ordered items per arm, baseline rates, confidence intervals and the minimum detectable effect chosen before launch.\n\nReturn analysis should wait until every order has passed the same return window. It should separate size-related returns from style, quality and changed-mind returns, then report exchanges and bracketing. Segmenting by device and product type would show where a generated image helps and where it merely adds latency.\n\nGARCIA changed one thing and measured orders and returns. It did not bundle sizing, visualisation and a redesigned product page into one treatment. It held the fit tool constant and asked whether the next uncertainty was worth solving. Faslet’s numbers suggest it was, and the unpublished statistics and margin figures will show whether that holds at scale.\n\n| Experiment detail | Public evidence | \n|---|---|\n| Commercial problem | Reduce hesitation and returns after size advice was already available. | \n| Control | Standard journey with Faslet size-me active. | \n| Treatment | The same journey plus access to AI virtual try-on. | \n| Scale disclosed | More than 5,000 ordered items in each arm. | \n| Reported result | 7.06% higher conversion, 5.54% lower return rate and 80% positive ratings. | \n| Not published | Visitors, dates, randomisation, baseline rates, significance, return window, price and implementation cost. | \n\n## Sources and disclosure\n\n- Faslet [GARCIA business case](https://faslet.me/cases/business-case-garcia/) and[public experiment summary](https://www.linkedin.com/posts/faslet_does-virtual-try-on-actually-sell-more-or-activity-7495461084254892032-jChi)\n- Faslet [integration documentation](https://docs.faslet.net/docs/virtual-try-on/) and[data processing terms](https://faslet.me/data-processing-terms/)\n- GARCIA [company history](https://www.wearegarcia.com/about-us)\n- CHI 2026 [virtual try-on user study](https://doi.org/10.1145/3772318.3791811)\n- Manufacturing & Service Operations Management [randomised field experiments on virtual fit information](https://pubsonline.informs.org/doi/10.1287/msom.2017.0686)\n\nFaslet supplied the system and published the GARCIA result. GARCIA has not independently released the analysis. Industry Contents found no public test dates, visitor denominator, baseline rates, statistical uncertainty, return-maturation window or implementation cost.\n\n## FAQ\n\n## What did GARCIA test?\n\nGARCIA tested access to Faslet’s AI virtual try-on on live webshop traffic. Both treatment and control retained the same Faslet size recommendation tool.\n\n## What result does Faslet report?\n\nFaslet reports 7.06% higher conversion and a 5.54% lower return rate in the treatment. It says each arm produced more than 5,000 ordered items.\n\n## Was the GARCIA virtual try-on result independently verified?\n\nNo. Faslet supplied the system and published the case. GARCIA has not released the analysis independently, and important statistical details remain unavailable.\n\n## What can another retailer copy?\n\nKeep existing sizing help in both groups and test the visual try-on as a separate layer. Judge it on contribution margin per exposed visitor after a complete return window.\n\n## Does virtual try-on prove physical fit?\n\nNo. It can help a shopper judge visual suitability, while fabric feel, drape and precise physical fit still need post-purchase measurement and return guardrails.\n\nCompare more measured experiments in the [Signal Index](https://industrycontents.com/ic-lab/growth-signal-index/), browse the [Growth Signal archive](https://industrycontents.com/category/growth-signal/), or use the benchmark tools in the [IC Lab](https://industrycontents.com/ic-lab/).", "url": "https://wpnews.pro/news/how-garcia-got-its-ai-try-on-to-lift-conversion-7", "canonical_source": "https://industrycontents.com/garcia-virtual-try-on-test/?utm_source=rss&utm_medium=feed&utm_campaign=editorial", "published_at": "2026-11-16 08:00:00+00:00", "updated_at": "2026-11-16 08:23:06.038113+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "generative-ai"], "entities": ["GARCIA", "Faslet", "JOG Group"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/how-garcia-got-its-ai-try-on-to-lift-conversion-7", "markdown": "https://wpnews.pro/news/how-garcia-got-its-ai-try-on-to-lift-conversion-7.md", "text": "https://wpnews.pro/news/how-garcia-got-its-ai-try-on-to-lift-conversion-7.txt", "jsonld": "https://wpnews.pro/news/how-garcia-got-its-ai-try-on-to-lift-conversion-7.jsonld"}}