Miros
Miros is an AI product discovery layer for online retailers. It reads shopper intent from browsing behaviour rather than typed queries, then feeds one live preference profile to search, browse, recommendations and shop-the-look. Pricing is quote-based.
What is Miros?
Miros is an AI product discovery platform for e-commerce sites, published by the Estonian company Miros OU in Tallinn. Its premise is that shoppers in visual categories cannot describe what they want: the vendor cites 85% of fashion shoppers never using the search bar, 90% of queries being too vague to return the right product, and 86% of apparel purchase intent being visual, leaving text to cover the remaining 14%. Instead of asking for a query, Miros infers intent from behaviour.
The technical foundation is called Intelligent Feed, described as a real-time intent extraction layer that sits underneath every discovery surface. It works in four stages: an AI assigns thousands of visual and semantic attributes to each product image without manual tagging; behavioural signals such as scroll velocity, click patterns, dwell time and ignored items are interpreted per shopper; contradictory signals are synthesised into a coherent preference profile; and that single profile then feeds every surface at once.
Six surfaces sit on top of it. Wordless Search lets shoppers discover products by browsing, with no query and no filters, and is included in all three plans. Discovery Bar adds natural-language search in a claimed 48 languages, enriched by the live profile. Inline Recommendations, Shop The Look, Find by Image and Style Collections complete the set. The vendor claims 50-300ms response times, more than a billion items indexed, over 500 million signals processed monthly, five patents and ownership of the underlying IP, developed over four years with the University of Tartu iCV lab.
Reported outcomes come from randomised A/B split tests across six retailers, with 140,000 to 2 million sessions per test: 7 to 19% GMV uplift, 296% higher conversion than Algolia, and 30 to 56% more repeat purchases. Named customers include momox fashion, Ravensburger, Carousell, ON24, Swap.com, PERCENTIL, Revolve and Jomashop. Integration is through JavaScript widgets or a GraphQL API, on headless or traditional stacks, in roughly two weeks. The company was founded in 2019 as FIND.fashion by the team behind Fits.me, and employs thirteen people across Tallinn, Lisbon and Belgrade.
What it does
- Read product images and extract thousands of visual and semantic attributes, with no manual tagging
- Decode live behavioural signals - scroll velocity, clicks, dwell time, ignored items - for each shopper
- Resolve contradictory signals into one coherent preference profile that updates in real time
- Feed that single profile to search, browse, recommendations and shop-the-look simultaneously
- Return visually similar products from a photo the shopper uploads, ranked by that profile rather than by pixel similarity alone
- Generate style collections automatically from real demand signals instead of manual curation
- Run natural-language search enriched by the live profile, in the 48 languages the vendor claims
When to use Miros / When not to
A quick filter to help you decide if Miros is the right fit.
When to use Miros
- Online retailers in visual categories - fashion, furniture, footwear, decor, toys - where shoppers cannot put into words what they are looking for
- Marketplaces and second-hand platforms with large one-of-a-kind catalogues, where keyword search collapses because no two items repeat
- E-commerce and merchandising teams whose own analytics show that most visitors never touch the search bar
- Retailers who want to add an intent layer without ripping out the search engine they already run, such as Algolia or Elasticsearch
- Product and growth leaders who want a head-to-head test on their own catalogue before any budget is committed
When not to use Miros
- Small merchants who expect to sign up, pay online and go live the same day: every plan goes through a sales conversation
- Buyers who need a published price before starting that conversation, since no amount, currency or billing unit appears anywhere on the site
- Catalogues described by part numbers or technical specifications rather than by images, where the visual-intent premise does not hold
- Teams looking for a permanent free tier or a self-serve trial: the free demo is offered only to selected businesses
- Organisations that require a signed data processing agreement, a subprocessor list or a published security certification before piloting anything
How to use Miros
A typical end-to-end flow, from setup to results.
- Check that the catalogue fits the premise: Miros targets visual categories such as fashion, furniture, footwear, decor, stock imagery and toys
- Open the contact page and complete the demo request form
- Provide the store URL, the search solution currently in place and the difficulties experienced with it
- Wait for Miros to build a working instance trained on the real catalogue, at no charge, before any commitment is made
- Compare that instance head to head against the search engine already running on the site
- Agree a plan and commercial terms with the sales team, since Starter, Premium and Enterprise all carry a Contact Sales button
- Connect the catalogue by feed on Starter, or in real time on Premium and Enterprise
- Integrate the discovery surfaces through JavaScript widgets, or through the GraphQL API on the Enterprise plan
- Start with a single surface and add others later: the platform is presented as modular and able to coexist with the current stack
- Follow the results in the dashboard included from Starter upwards, or in the custom monthly reporting offered on Enterprise
Pros & Cons
Pros
- Reported results come from randomised A/B split tests on named retailers rather than from selected cohorts
- No manual tagging is required: the AI reads the product images, so relevance does not depend on metadata quality
- Coexists with an existing search engine such as Algolia or Elasticsearch instead of forcing a replacement
- Short integration: around two weeks to production and less than one engineering man-week on the customer side
- A working instance is built on the prospect's own catalogue before any commitment, which makes the claims testable
- Owned technology: five patents, IP retained by the vendor, co-developed with a university computer vision lab
- Legal identity is fully verifiable: Miros OU, Estonian registry code 14677640, VAT number confirmed by the EU VIES service
Cons
- No published price at all: Starter, Premium and Enterprise each lead to a Contact Sales button
- No permanent free plan and no self-serve trial; the free customised demo is limited to selected businesses
- No terms and conditions and no legal notice: the Privacy Notice points to terms that do not exist on the site
- No public API documentation, although GraphQL API access is sold as an Enterprise feature
- No data processing agreement, no subprocessor list and no security certification are published
- The vendor's position on training models with customer data is never stated anywhere on the site
- Every performance figure is the vendor's own, and the 48 supported languages are counted but never named
Pricing & Plans
There is no free plan and no published price. The pricing page presents three tiers - Starter, Premium and Enterprise - and each one leads to a Contact Sales button, with no amount, currency or billing unit shown anywhere on the site. The only pricing statement made is that the tiers are calibrated to deliver a twelvefold return on investment. Commercial terms are therefore agreed case by case. Before any commitment, Miros offers a free customised demo consisting of a working instance built on the prospect's own catalogue; the contact page restricts that offer to selected businesses, so it should not be read as a general free trial.
- Wordless Search
- standard widget design
- feed-based catalog sync
- email support
- standard dashboard
- everything in Starter
- plus Inline Recommendations
- brand-configured widget design
- real-time catalog sync and dedicated support
- everything in Premium
- plus Discovery Bar AI search
- GraphQL API access
- custom widget design
- custom monthly reporting and dedicated support
Data, GDPR & hosting
A consolidated view of how Miros handles your data.
GDPR overview
Implementation is visible and specific, but never presented as a compliance claim. The Privacy Notice names Miros OU as controller, cites its legal bases article by article - Article 6(1)(f) legitimate interest, 6(1)(b) contract, 6(1)(c) legal obligation - separates essential cookies from optional ones placed only with consent, lists the categories of recipients, and frames transfers outside the EEA under the European Commission's standard contractual clauses. It enumerates the full set of data subject rights and names the competent supervisory authority, the Estonian Data Protection Inspectorate. What is missing is just as clear: the site never states that Miros is GDPR compliant, designates no data protection officer, publishes no data processing agreement and names no subprocessor. Being established in Estonia, the company needs no Article 27 representative and designates none.
Who owns the data?
The site publishes no terms and conditions, so no clause states who owns the data Miros handles for a retailer. The only legal document is the Privacy Notice, and it is explicit about its own scope: it covers visitors to the miros.ai website and the representatives of business partners and customers, not the shoppers whose behaviour the product analyses. Within that scope the named controller is Miros OU, Estonian registry code 14677640, Tehnika 16, Tallinn 10149, reachable at info@miros.ai. Anything concerning ownership of catalogue data, behavioural signals or the preference profiles the platform derives would have to come from a contract that the company does not publish.
Reuse rights
Nothing on the site tells a customer what they may reuse or do with the data afterwards, because no terms are published. The Privacy Notice describes only what Miros does with the personal data it collects for its own account: establishing contact with prospects, marketing its services, managing the commercial relationship, answering enquiries, defending legal claims and meeting accounting and tax obligations. The legal bases invoked are legitimate interest, performance of a contract and legal obligation. Recipients are given as categories rather than names: tax and other authorities, IT, payroll and accounting providers, analytics partners, banks, internal and external auditors, and investors or acquirers in the event of a corporate transaction. Transfers outside the European Economic Area are allowed and framed by the European Commission's standard contractual clauses. Rights over catalogue data, behavioural signals and the preference profiles built by the platform are simply not addressed.
Data retention & training
Hosting summary
The site names no hosting country and no hosting region, for the marketing site or for the platform. The Privacy Notice states only that Miros may transfer personal data outside the European Economic Area from time to time, and that such transfers are protected by appropriate safeguards, specifically the standard data protection clauses adopted by the European Commission. That is a transfer clause, not a declaration of where data is stored, and it should not be read as one. No subprocessor is named: the notice lists only categories of recipients, such as IT, payroll and accounting providers and analytics partners. The publisher itself is established in Estonia, and the marketing site resolves to a hosting address in France, but neither fact says anything about where the discovery platform stores catalogue data or shopper signals. A buyer who needs a hosting jurisdiction committed in writing will have to obtain it during contract negotiation.
Things to keep in mind
Risks and trade-offs to weigh before adopting Miros.
- Behavioural profiling is the product itself: scroll velocity, clicks, dwell time and even ignored items are read from the first visit, with no login and no history required, and shoppers are unlikely to grasp how much is inferred about them
- Nothing on the site explains what a retailer must disclose to its own shoppers; the Privacy Notice covers only visitors to miros.ai, leaving that compliance burden entirely with the customer
- Discovery driven by inferred taste narrows what a person sees, and a shopper who wants to be surprised may be shown an ever tighter reflection of their own past behaviour
- The vendor never states whether customer data or shopper signals are used to train its models, so a retailer cannot know what leaves its own perimeter
- Every performance number comes from the vendor; the method is described as randomised A/B testing, but no independent source corroborates it on the site
- The absence of terms, of a data processing agreement and of a subprocessor list means a buyer negotiates without any published baseline to work from
- Two different Tallinn addresses appear on the site, only one of which matches the register, and the site still serves an llms.txt describing a previous website whose pages have largely disappeared
Setup & Integrations
Technical difficulty
Low to moderate on the customer side. The vendor quotes around two weeks to production and less than one engineering man-week of effort, with two integration routes: JavaScript widgets or a GraphQL API, on headless and traditional platforms alike. No manual product tagging is required, since attributes are read from the images. The catalogue is connected by feed on the entry plan or in real time on the higher ones. The platform is modular, so a first surface can be deployed and others added later. The real effort is commercial rather than technical, as everything begins with a sales conversation.
Deployment
Behind Miros
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Miros.
Frequently asked questions
How does Miros differ from a conventional site search engine?
Do I have to replace the search engine I already use?
How long does integration take?
Will it slow my website down?
How much does Miros cost?
Is there a free plan or a trial?
Is there an API, and is it documented?
Is there a mobile application?
Who is behind Miros?
Which languages does it support?
Should you pick Miros?
Miros addresses a real and well-defined gap: shoppers in visual categories who cannot describe what they want, and who therefore never use the search bar at all. The answer it proposes - one behavioural intent profile built from the first visit and pushed to every discovery surface at once - is coherent, and the engineering behind it is credible, with five patents, a university research partnership and figures on scale that fit a platform serving large catalogues.
The evidence is unusually concrete for this category. Six named retailers, randomised split tests with 140,000 to 2 million sessions, and outcomes reported both as GMV and as repeat purchase rates. It should still be read for what it is: material published by the vendor, with no third-party corroboration offered on the site.
What is missing is contractual rather than technical. There is no pricing, no terms and conditions, no legal notice, no data processing agreement, no subprocessor list and no statement on whether customer data is used to train models - and the Privacy Notice itself points to terms that are not online. For a platform that profiles shoppers in real time and is sold to retailers who carry their own GDPR obligations, that silence is the main reservation to raise. The company's identity, by contrast, is entirely verifiable in the Estonian business register.
Miros is therefore best considered by retailers large enough to negotiate a contract and to run a proper head-to-head test on their own catalogue. The free instance built before any commitment makes that test easy to obtain. Smaller merchants looking for a published price and a self-serve signup will not find one here.
- Choosing a selection results in a full page refresh.
- Opens in a new window.