Nobody buys a dress. They buy an occasion.

There are around twenty shopper-facing virtual try-on apps for apparel live on the Shopify App Store. We went through them in August 2026.

They answer the same question: does this specific item look good on me?

Not one answers the question the customer actually arrived with: I have a wedding in October and nothing to wear.

That gap is the whole product.

What is the customer actually asking?

She isn’t asking about a dress. She’s asking about an event. The real question has a date, a place, a dress code, a social position, a budget, and a body she already knows well. “Midi dress, navy” is what’s left after she gave up trying to type the real thing into a search box.

The real thing looks like this:

Outdoor wedding, mid-September, in Provence. I’m in the wedding party but I’m not the bride and I shouldn’t outshine her. My sister will be there and she’s seen everything I own. It’ll be 30°C at four and cold at eleven. Around €250, and I’d like to wear it again.

Every one of those constraints changes the answer. Not one of them fits in a filter menu.

So she does what everyone does. She goes to Instagram, saves nine screenshots, opens six tabs, and orders three sizes of two dresses intending to send most of it back. The industry then calls the resulting returns a fit problem and sells her a size chart.

It was never a fit problem. It was a decision problem, and it happened long before she reached your product page.

Why doesn’t a try-on button solve this?

Because a button on a product page has no reason to be pressed. The customer has to have already found the item, already be interested enough to upload a photo of herself, and already be patient enough to wait for a render. Try-on arrives at the very end of the decision, where it can only confirm or deny a choice she made somewhere else.

The few numbers that exist support this. The most specific public figure we found comes from a merchant writing on Substack in February 2026 about his own store experiments: 15–25% of product page visitors engaging, after optimisation. One store, self-reported — and still more transparent than anything a vendor has published.

The category’s answer has been to make the images better. Three years of research effort has gone into drape, fabric and print fidelity. The images are genuinely excellent now.

It hasn’t moved the problem, because the problem was never image quality. It was that nobody had a reason to press the button.

What does occasion-first actually look like?

The customer states the occasion in her own words. The AI reads what she’s told it and what she’s shown it — silhouette, the colours that actually flatter her, what she already wears. It picks from your catalogue, never from stock imagery or a generic style database. Then it shows her the pieces on her own body. Try-on is the last step, not the first: the proof, not the pitch.

The order matters more than any single part of it:

She states the occasion. Not a filter. A sentence.

The AI reads the customer before proposing anything. Silhouette, the colours that suit her, the register she already dresses in. This is the step that separates a stylist from a search box, and it’s the step almost everyone skips.

It answers from your collection. Your pieces, your sizes, what’s actually in stock. Not a mood board, not a competitor’s dress, not a Pinterest grid.

Then, and only then, it shows her the result on herself.

Try-on stops being a widget and becomes the last line of an argument.

Why does this matter commercially — and why we don’t lead with returns

Every competitor in this category sells the same thing: fewer returns. It’s saturated, it’s contested, and it’s slow and awkward to prove on your own dashboard within a quarter. We audited the category’s returns claims and could not trace almost any of them to a source.

An occasion is a multi-item purchase by nature. A wedding is a dress, shoes, and something for the evening when it gets cold. A first day in a new job is three things, not one. When the entry point is the occasion rather than the item, the natural output is an outfit.

That has two consequences you can verify yourself:

Items per order and average order value show up in your own Shopify analytics within weeks. Not quarters. You don’t need a vendor to tell you whether it’s working.

Almost nobody in this category even mentions AOV. They’re all selling a returns number they can’t source.

We’d rather compete on the metric that’s provable.

Where does this matter most?

Occasionwear. Bridal party. Formalwear. Modest fashion. Resort. Festival. The segments where the purchase is an occasion, the basket is naturally multi-item, and the stakes are high enough that the customer wants advice rather than filters.

There’s a second reason those segments matter, and it’s worth being honest about.

Generative try-on has documented weaknesses on exactly the garments those segments sell.

Loose and draped garments break body-shape estimation, because the fabric occludes the body contour the model needs to read. That’s the finding of Wu, Shen and Igarashi at the University of Tokyo: existing methods “rely on human body semantic maps to align garments with the body, but these maps become unreliable when body contours are obscured by loose-fitting garments, resulting in degraded outcomes” (arXiv:2506.12348).

Layered garments are worse. The standard approach masks out the existing outfit to build a “cloth-agnostic” representation — which, as Feng, Chen, Shan and Kemelmacher-Shlizerman at the University of Washington put it, “irreversibly discards the essential contextual information — the very outfit upon which a new layer must be added” (arXiv:2607.22924). The inner layer simply disappears.

And Google’s own blog post on the extension says plainly that replacing a garment on a dress-length image “can smudge the person’s features or obscure important details of their body” — the company’s own description of why full-coverage garments are harder.

So an abaya over a dress, a kaftan over a slip, a jacket over a shirt — the layered silhouette that dominates Gulf occasionwear — is an open research problem as of August 2026, not a solved engineering task. Any vendor telling you otherwise either hasn’t tested it or is hoping you won’t. Ask for a test on your own pieces before you sign anything.

Which is precisely why the occasion framing matters more in those segments, not less. When the customer’s real question is “does this read right for the event?” rather than “is this exactly my size?”, the styling answer carries the weight and the render supports it. Get the order backwards and the technology’s weakest dimension becomes the whole product.

The thing nobody in this category is collecting

“Wedding guest, mid-September, outdoor, €200, midi length, sleeves.”

That isn’t a session. That’s merchandising intelligence.

Right now the try-on apps on Shopify collect email addresses and run share-for-discount loops. Not one of them captures what customers are actually dressing for — or, more valuable still, what they asked for that your catalogue couldn’t answer.

Every unanswered occasion is a gap in your next collection, told to you by someone who was ready to buy.

What we’re building, and what we’re not claiming

Jedid is an AI try-on and styling engine that runs inside a Shopify store. The customer describes the occasion, the AI reads her — silhouette, the colours that flatter her, the style she already wears — recommends from your collection, and shows her the result on herself. Install is a plugin, not a project.

What we’re not doing: we won’t publish a conversion multiple until we have one from a named brand, over a defined window, with a methodology we can show you in full. We audited this category’s statistics and most of them have no traceable source. We’re not adding to that pile.

We’re opening the beta to a small number of Shopify fashion brands in France and the Gulf. Free during the programme. What we want back is honest feedback, and permission to publish the real numbers — good or bad — with your name on them.

If that’s you, the details are on the beta page.

Sources. Shopify App Store listings, reviewed 24 August 2026 · Wu, Shen & Igarashi, University of Tokyo, “Real-Time Per-Garment Virtual Try-On with Temporal Consistency for Loose-Fitting Garments,” arXiv:2506.12348 · Feng, Chen, Shan & Kemelmacher-Shlizerman, University of Washington, “Layering Virtual Try-On,” arXiv:2607.22924 (v3) · Google Shopping blog, “Virtual try-on for dresses,” 5 September 2024 · “The Harsh Reality of Virtual Try-On,” Substack, 11 February 2026.

Last updated 24 August 2026.