What’s the next “big thing” in agentic commerce? It’s probably not what you think.

Every merchant has a rich picture of what happened after a shopper found a product: clicks, add-to-carts, conversions, returns. Almost none have a reliable picture of what the shopper actually wanted before that, the precise thing, in the precise color, in their head.

For twenty years the storefront’s front door was a keyword box, at best paired with a color filter offering a handful of arbitrary swatches, and neither one could record intent it was never built to receive. Two capabilities now arriving on storefronts, trimodal search and virtual try-on, change that. They turn the moment of desire into structured, first-party data, and one dimension of that data has never existed in mainstream analytics at all.

𝗔 𝗻𝗲𝘄 𝗱𝗶𝗺𝗲𝗻𝘀𝗶𝗼𝗻: 𝗲𝘅𝗮𝗰𝘁 𝗰𝗼𝗹𝗼𝗿, 𝗻𝗼𝘁 𝗮 𝗱𝗿𝗼𝗽𝗱𝗼𝘄𝗻

Start with color, because it is the clearest example of something genuinely new. In a conventional store, color is a low-cardinality label the shopper picks from: red, blue, green, a dozen swatches if the merchandiser was generous. Analytics inherits that coarseness. You can learn that “red” sold. You cannot learn which red.

Now give the shopper a color wheel instead of a dropdown. They are not choosing from six presets; they dial in the exact shade they have in mind, one point out of roughly sixteen million. Not “red,” but this red, a specific value like #E53935. That alone is a resolution of intent no storefront has ever captured, because until now nothing at the front door could receive it.

Then let them add a sentence to it. The same search bar takes a natural-language prompt, “mini dress,” “relaxed linen blazer,” “something for a summer wedding,” and runs it together with that specific color shade they are interested in. Precise color and plain language resolve as a single query. That combination is the new dimension. The shopper stops flattening a rich want into whatever a keyword box and a swatch grid could accept, and expresses it in full, while the storefront records all of it.

The output is data that did not exist a moment earlier: this shopper wanted this exact red, in a mini dress. Multiply it across a season and a merchant holds demand at the resolution of a hex value and a sentence, not a six-item color menu. It is the signal closest to what the customer actually meant, and it is finally measurable.

And it is simpler than it sounds. The shopper does nothing technical. Under the surface it works the way good natural-language search already does: the platform reads the whole query, the words and the exact color value, works out what the shopper actually means, and matches it to the closest products in the catalog. The person just describes and dials a shade; the engine does the interpreting and the matching. When I architected this color and text-plus-color process at visualAI, and the pipeline that creates these capabilities, I thought it was so important and unique that I filed for a patent on it.

𝗥𝗲𝗰𝗼𝗺𝗺𝗲𝗻𝗱𝗮𝘁𝗶𝗼𝗻𝘀 𝘁𝗵𝗮𝘁 𝗺𝗮𝘁𝗰𝗵 𝗶𝗻𝘁𝗲𝗻𝘁, 𝗻𝗼𝘁 𝗸𝗲𝘆𝘄𝗼𝗿𝗱𝘀

Recommendation engines have historically worked off behavior proxies: what co-viewed, what co-purchased, what a similar cohort clicked. Useful, but downstream of meaning. When the input carries an exact shade and plain language, recommendations move to the level of true intent. A shopper who dialed in that specific red and typed “mini dress” can be shown the neighboring shades and cuts that satisfy the same desire, and steered off the near-misses that quietly kill a session. The signal also improves the model itself, because the training data now includes the precise thing people asked for, not only what they happened to click.

𝗣𝗹𝗮𝗻𝗻𝗶𝗻𝗴 𝗮𝗻𝗱 𝗯𝘂𝘆𝗶𝗻𝗴 𝗮𝗴𝗮𝗶𝗻𝘀𝘁 𝗿𝗲𝗮𝗹 𝗱𝗲𝗺𝗮𝗻𝗱

This is where it gets material. Merchandising and buying teams plan against history and instinct: last season’s sell-through, a trend forecast, a color story locked months ahead. Exact-shade intent data puts a live demand curve underneath those decisions. If shoppers keep dialing in a particular coral weeks before it appears in your buy, that is a signal you can act on, in assortment, in depth, in the specific colors you stock, at a precision a named-color report could never give you. It also trims the two most expensive tails in retail: demand you never saw, and colors you overbought into markdown. Analysts already put the retail analytics market near $20.65 billion by 2031 as retailers chase this kind of edge. Precise-intent data is the input that makes those tools sharper.

𝗧𝗮𝗿𝗴𝗲𝘁𝗶𝗻𝗴, 𝗿𝗲-𝘁𝗮𝗿𝗴𝗲𝘁𝗶𝗻𝗴, 𝗮𝗻𝗱 𝘁𝗵𝗲 𝘁𝗿𝘆-𝗼𝗻 𝗹𝗮𝘆𝗲𝗿

For advertisers and agencies, this is a new audience primitive. Traditional retargeting fires on a viewed product or an abandoned cart, blunt instruments built from what someone did. Exact-shade, natural-language intent is sharper and, crucially, it is zero-party: the shopper told you, in their words and to the exact value, what they were after. You can build an audience of people who dialed in a specific rose and searched a sleeveless midi, and re-engage them with that precise shade when it lands, instead of chasing them with the whole category. Analysts expect zero-party data inside customer platforms to grow around 37% a year, precisely because pairing what a shopper did with what they said reveals why they hesitated, compared, or bought.

Virtual try-on adds a behavioral layer on top. Every try-on is a high-intent event: which item, which size, how long, whether they were added to the cart. That is a confidence signal and a fit signal at once. It predicts conversion, flags likely returns before they happen, and gives targeting a segment that has literally pictured itself in the product. Industry surveys still put regular visual-search use around 10% of US adults, with roughly 42% interested, which tells you this stream is early, not niche. The merchants instrumenting for it now are capturing a signal most competitors cannot yet see.

𝗪𝗵𝗼 𝗮𝗰𝘁𝘂𝗮𝗹𝗹𝘆 𝗰𝗮𝗽𝘁𝘂𝗿𝗲𝘀 𝗶𝘁

None of this data exists unless the storefront can receive the input and log it. A keyword box and a swatch grid cannot record an exact shade and a full sentence, and a static gallery cannot record a try-on. The capability has to be there first. That is the layer we build at visualAI: the discoverGPT platform exposes trimodal search and virtual try-on as APIs an agency or developer can put on a merchant’s own storefront, and the apps built on it, shopperGPT for describe-and-dial-a-shade search and studioGPT for try-on, are the productized version of the same engine. The point is not the tools. It is that the moment a merchant turns on this kind of discovery, they stop throwing away the most valuable signal they generate and start keeping it.

𝗧𝗵𝗲 𝗺𝗼𝗮𝘁 𝘁𝗵𝗶𝘀 𝗯𝘂𝗶𝗹𝗱𝘀

Every query and every try-on adds to a proprietary corpus of intent, exact color, and fit that no competitor can copy and no platform hands over. That corpus makes the next recommendation better, the next buy tighter, the next audience more precise, which drives more usage, which deepens the corpus again. Search and try-on endpoints will commoditize; plenty of vendors will offer a version. What does not commoditize is the accumulated first-party record of what your customers actually wanted, described in their own words and dialed to their own colors. That moat widens every day the capability is live, and it belongs to whoever captured it first.

For two decades, retail analytics has been a story about what happened after the click. The merchants who add exact-color, natural-language discovery and virtual try-on are about to have a story about what the customer wanted before it, at a resolution the industry has never had, and the ones who instrument for it first will spend the next decade compounding a data advantage the rest cannot buy back.