Claro

Claro (also referred to as Claro AI) develops software that ingests supplier feeds, spreadsheets, PDFs, drawings and existing catalog records and turns them into structured, validated product and supplier data. The platform decodes incoming files, resolves product identity across sources, classifies items into customer taxonomies or standards such as ETIM, ECLASS and UNSPSC, enriches missing attributes from source material, and writes approved records back into systems of record such as ERP and PIM. Each match, attribute and classification carries a source reference and a confidence score.

The product is organised in four layers: a context layer holding cited claims, catalog fingerprints, a domain glossary and ERP/PIM history; a set of six agents (Decoder, Classifier, Matcher, Enricher, Crosswalk and Monitor) each performing a single task; a governance layer that routes changes into three lanes — auto-apply for low-impact, already-trusted fix patterns, a review queue as the default, and mandatory human sign-off for merges, un-merges and reclassification regardless of confidence; and a grid view where catalog teams review records, sources and scores with grouped approval. The Monitor agent watches downstream changes and feeds them back into context, so the system is positioned as continuous governance rather than a one-time cleanup.

Claro positions itself as a complement to, not a replacement for, ERP, PIM, MDM, commerce, procurement and marketplace systems. Earlier public material described the company as building AI knowledge infrastructure that converts unstructured enterprise data into specialised Small Language Models for tasks such as text classification, extraction and data enrichment, selecting from a library of more than 100 custom SLMs.

Founding story

Claro was founded in 2024 by Matteo Fava and Tameesh Biswas, who met during an Antler founder residency in Berlin. Fava spent about ten years at Delivery Hero, joining as employee 300 and going on to build and lead its global data organisation, with earlier roles at Rocket Internet and Wolters Kluwer. Biswas is an NLP researcher whose work has been published in journals including MIT Press, and previously worked as a software developer at Yelp — where he built a data monetisation and geolocation ML platform — and Autodesk. Fava has said the company grew out of watching catalogs degrade at Delivery Hero whenever a second supplier, market or ERP was added.

Business model

Claro sells its data layer to businesses in four delivery modes: the Claro platform used directly by catalog teams, an API called per record or per batch by engineering teams, an embedded deployment inside vertical platforms that offer product matching to their own customers, and an MCP server that customer-side agents query directly. Published pricing is organised around pilot/value validation, continuous operations and embedded/API deployment, with separate self-serve pricing for the Research Agent.

Commercial engagements are structured around a paid pilot or value-validation phase, continuous operations, and embedded/API deployment, with self-serve pricing offered separately for the Research Agent.

Traction

The company reports more than 140 supplier feeds matched into a single catalog across active engagements and a 4,000-product supplier file processed into structured, validated records. Published operating guidance describes use by teams managing roughly 50,000 to 500,000 SKUs and multi-quarter enrichment backlogs compressed into weeks, presented as an anonymized range rather than a guarantee. Customer identities are withheld unless covered by a named case study.

Latest developments

Claro's current public product scope covers catalog matching, product deduplication, enrichment, classification, supplier onboarding, AI output validation, AI search optimisation, compliance documentation and a pricing monitor, alongside a self-serve Research Agent, with industry-specific offerings for distributors, industrial manufacturers, spare parts and aftermarket, and marketplaces. The company lists a five-person team and states it has no open roles posted while accepting open applications.