Miros

Miros is a behavioural product discovery layer for visual retailers. It builds a real-time intent profile from scroll, click and dwell signals, then feeds that profile to search, browse and recommendations. Vendor tests report 7 to 19 percent GMV gains and a two week deployment using JS widgets or a GraphQL API.

Miros is a product discovery layer for online retailers whose catalogues resist text search. The company argues that shoppers in fashion, furniture, footwear and decor largely cannot articulate what they want, so keyword boxes, filters and chatbots leave most of that intent unserved. Miros replaces the prompt with observation: it watches how a visitor scrolls, clicks, dwells and skips, then turns those signals into a live preference profile for that individual. The mechanics run in four stages. The AI first extracts thousands of visual and semantic attributes from each product image, which removes manual tagging from the setup. It then interprets behavioural signals in real time, including scroll velocity, clicks, dwell time and deliberately ignored items. Contradictory signals are reconciled into one coherent profile, and that single profile is pushed to every discovery surface at once: search, browse, recommendations and shop-the-look. Products are surfaced in under 60 seconds according to the vendor, with response times quoted between 50 and 300ms across catalogues of a billion items and above. The fit is mid-market to enterprise commerce teams in visually driven categories, plus marketplaces and second-hand platforms with large, messy inventories. Miros is positioned as a layer rather than a replacement, and the company states it works alongside Algolia, Elasticsearch or an existing stack. Published results come from randomised A/B splits across six retailers: GMV gains of 7 to 19 percent, conversion up 296 percent against Algolia in a 100K session test, and repeat purchase lifts of 30 to 56 percent. Named users include Revolve, Ravensburger, Carousell, momox fashion and PERCENTIL. The build sits between custom in-house personalisation work and prompt-based search tools such as Algolia or Google Vertex. Deployment is quoted at two weeks, with JS widgets or a GraphQL API for headless and traditional storefronts, and the vendor says the integration costs under one engineering man-week on the client side. Modules can be adopted one at a time. A working demo on your own catalogue is built before any commitment, and the average sales cycle from demo to production is given as 36 days.