The most cited AR case study in commerce is a furniture app. IKEA Place launched in 2017, letting shoppers drop true-scale sofas and shelves into their rooms through the phone camera. Our database records two facts about it that frame the whole build-or-buy argument: IKEA built it in-house, and furniture placement was the use case. Both facts are true, and together they mislead almost everyone who copies them, because IKEA had a 400-product core catalog, a global brand budget, and a decade-long horizon. You probably have a different shape of problem.
How the 68 brands actually got their AR built

Counted across all 68 records, in-house is the minority route: 16 brands. Named external vendors built 39, and 13 records do not name a builder at all. No single vendor dominates. ModiFace is the largest with 5 brands, and all 5 are beauty, which is why it owns the virtual try-on conversation. The Marxent and 3D Cloud family accounts for 4, mostly large home retailers. Snap supplies 3. Perfect Corp, Fittingbox, Vyking and Wannaby hold 2 each, across eyewear and footwear, and 19 brands use a vendor that appears only once or twice in the whole set.
The in-house 16 cluster where fit is the product: Warby Parker, Ray-Ban and Lenskart in eyewear; James Allen, Blue Nile, Tiffany and Brilliant Earth in jewelry; Nike and Zara in apparel. IKEA, Wayfair and Houzz carry the furniture flag, which is 3 of the 12 furniture brands. The pattern is consistent: brands build in-house when the AR experience is the brand, and they buy when AR is a feature on a product page. A case study like IKEA's belongs to the first group. Most stores reading this are in the second.
The in-house route, honestly priced

In-house means a decent camera setup plus software ($0 to $150 a year for apps like Polycam Pro) plus labor. Throughput runs 5 to 15 products per day per person, including cleanup. Quality is inconsistent until the team learns, then consistent. In-house is best for 500 or more SKUs, ongoing new products, and brands where 3D is core to the offer. Below that scale, the salary is the budget and the models are the side project.
The SaaS route, honestly priced
Per-model SaaS inverts the shape. ModelSnap charges $9.99 per HD model: 100 SKUs cost $999, with no equipment and no hiring. Throughput is parallel, since you upload 100 products and get 100 models back. Quality is consistent and store-ready, GLB and USDZ out of the box. SaaS is best for under 300 SKUs, seasonal catalogs, and testing AR before committing to it. Sources: Shopify
The 13 records with no builder named are a finding too
Thirteen of the 68 records do not name who built the experience. Some of that is missing research on our part, and we flag it as unrecorded rather than guess. But part of it reflects how the work is actually bought now: platform natives, theme-level 3D viewers and small freelance jobs leave no vendor logo to record. Shopify itself appears once as the named implementation, for a store whose AR is simply the platform feature switched on. The practical read: the market's center of gravity has moved from named AR agencies toward a mix of in-house teams at fit-driven brands, try-on specialists in beauty and eyewear, and quiet platform defaults everywhere else. Your decision is less agency-versus-salary than it used to be. It is increasingly specialist-versus-default, and the default is free.
The hybrid most brands land on
SaaS for the back catalog, in-house capture for new weekly arrivals once AR proves itself. Start with the $999 experiment, not the $15,000 hire. If you want a zero-dollar first step before either, our free turns a single photo into an exportable relief model, which is the smallest possible test of whether your shoppers touch 3D at all. Deciding whether to buy or build your 3D pipeline? Buying gets you live in weeks; building pays off only at serious SKU scale. IKEA is the proof that in-house can be great. The other 67 records are the proof that it is not the only way anyone ships.