A B2B textile procurement platform connecting buyers with the right suppliers. Reckonsys built the Claude-powered layer that explains matches, resolves the messy language of materials, and turns activity into opportunities.
Fabyso
Textile & apparel sourcing, B2B marketplace
Claude API
AI Development
Custom Software Development
Data Visualization and Analytics
Reckonsys built a Claude-powered intelligence layer on top of Fabyso's existing platform:
Matching
Claude sits on top of the existing scoring engine rather than replacing it. The algorithm ranks; Claude writes why a supplier fits a requirement — and why an expected one didn't make the shortlist.
Strong fit. Holds certified stock in the exact 65:35 blend and GSM you specified, ships to your destination port regularly, and has fulfilled three orders above your volume in the past year.
Ranked low despite matching on material and price. Minimum order quantity is roughly four times your requirement, and lead time exceeds your stated delivery window by two weeks.
Plain-language reasoning for why a supplier fits a requirement is grounded in the attributes the engine actually weighed. Mismatch explanations are generated as well as matches, which is the harder generation problem and the one that builds confidence in the ranking. A ranked top-matches view makes a shortlist scannable rather than exhaustive.
Explaining an exclusion is what separates a recommendation engine users trust from one they second-guess. When a buyer can see why an expected supplier did not appear, the ranking stops looking arbitrary.
Materials
Claude validates free-text material entries against a canonical materials list. Spelling variants, trade names and regional terms resolve to one entity — so matching improves before the scoring engine even runs.
Textile terminology is regional, historical and inconsistently applied, which is precisely why a lookup table never solved this. The same fabric carries different names in different markets and different decades of the trade, and only a model that understands the language can tell a variant from a genuinely different material.
Supplier insights
Enquiry volume, search patterns and contact activity pass to Claude as structured signal. It writes dashboard cards that pair an observation with a suggested action — framed as an opportunity, not a passive statistic.
Eleven buyers searched for organic cotton jersey in your shipping region this week. Your catalogue covers it, but the listing does not name the certification they filtered on. Add it to appear in those results.
The distinction is the whole point of the feature. “Views are up 12%” is a number a supplier scrolls past. A prompt that names what changed and what to do about it is something they act on the same day.
“The algorithm ranks. Claude explains why.”
Anthropic partnership
All three features are language problems in different clothes. Claude was chosen for specialised domain vocabulary, consistent structured output at production volume, and copy that reads as considered rather than templated.
The scoring engine stays the ranking authority. Claude reasons about and communicates its output, so the AI layer can improve without touching ranking logic a live marketplace depends on.
Explaining a match means turning a numeric score and a set of attributes into a sentence a sourcing manager will act on. Normalizing material names means recognising that a trade name, a regional term and a misspelling all point at the same fabric, which defeats exact-match lookup and rule-based cleaning alike. Insight cards mean reading behavioural data and writing something a supplier finds worth their attention.
On a marketplace where users see generated text on every result, quality of expression is not cosmetic. It determines whether the feature gets trusted or ignored.
Claude powers all three intelligence features on the Fabyso platform. Fabyso's algorithmic scoring engine remains the ranking authority. That division of labour was deliberate: deterministic scoring where determinism matters and results must be reproducible, Claude where the work is language and judgment. It also keeps the AI layer independently improvable, since prompts and outputs can be refined without touching the ranking logic a live marketplace depends on.
Buyers see why a supplier was recommended and why another was not. Suppliers see what buyers are looking for and what to change. Delivered as an ongoing managed engagement, with Reckonsys continuing to operate and extend the layer as the platform grows.
A marketplace that shows its reasoning, on both sides of the transaction.
Let's collaborate to turn your business challenges into AI-powered success stories.
Get Started
