Matching buyer and vendor catalogs across a B2B flower network
How a B2B flower network improved buyer-to-vendor catalog matching at platform scale
Current phase: CUSTOMER DEPLOYMENT — completed matching program with expansion and integration options evaluated
A B2B network connecting flower buyers with growers and vendors faced a direct growth constraint: every buyer had to map its own product catalog to each vendor catalog before inventory could become visible and transactable.
Buyer catalogs ranged from hundreds to thousands of products, while individual vendor catalogs could exceed 46,000 records. One large buyer had more than 20,000 unmapped products and dedicated staff performing the work.
Claro worked with the network to implement and evaluate a many-to-many matching workflow with confidence tiers, human review, and domain-specific logic for flower attributes.
At a glance
Industry: B2B flower and perishables network
Region: International
Operating model: Transaction platform connecting buyers with growers and vendors
Scale: Buyer catalogs of approximately 800–12,000 products; vendor catalogs exceeding 46,000 products
Primary Claro workflow: Many-to-many B2B catalog matching
The challenge
Flower products have limited global identifiers and significant semantic variation. Color labels such as salmon and peach may need to map to a broader orange family, while stem length, weight, grade, and range overlap influence whether two records are commercially interchangeable.
The relationship is not always one-to-one. One buyer product may validly connect to several vendor products, improving availability and supply resilience.
Unmapped inventory was commercially invisible. Catalog matching therefore directly affected order opportunities, onboarding speed, and transaction revenue.
Why the existing approach was not enough
Manual mapping was slow and depended on operators with specialist product knowledge. New partners could abandon onboarding because the volume of mapping work was overwhelming.
Exact rules did not capture acceptable near-matches, while an unconstrained similarity model could ignore commercially important differences in grade, size, weight, and stem length.
The matching and review workflow also needed to fit naturally into the platform’s existing operating model. A separate external tool would create additional adoption friction for buyers and internal teams.
The solution
Claro combined semantic product matching with attribute penalties and flower-specific domain logic. Buyers received ranked vendor candidates, confidence tiers, and a confirm-or-deny review workflow.
The system accounted for commercially relevant relationships that could not be captured through exact identifiers alone. Color families, stem lengths, weights, grades, and acceptable ranges were normalized and incorporated into the matching score.
Measured iterations improved the quality of high-confidence suggestions. In one evaluation, opportunity accuracy increased from 59% to 83%. With vendor-product filtering, the true-positive rate improved from 25% to 77%, while opportunity accuracy reached 90%.
Earlier large-catalog testing showed that high-confidence results covered 53% of products at 83% accuracy, while the correct result appeared within the top ten candidates for 90% of products.
The engagement also identified a strategic next step: a shared global catalog and taxonomy that could reduce repeated buyer-to-vendor mapping across the network.
How the workflow works
Ingest buyer and vendor catalogs — Product names, grades, colors, sizes, weights, and available identifiers are loaded from both sides of the network.
Normalize domain-specific attributes — Color families, length units, grade ranges, weights, and other flower-specific conventions are prepared for comparison.
Generate ranked vendor candidates — Each buyer product receives a ranked set of relevant vendor suggestions rather than one forced match.
Apply semantic and attribute scoring — Similarity is adjusted according to commercially important differences such as grade, size, weight, and range overlap.
Route suggestions by confidence — High-confidence opportunities can progress automatically or be pre-approved, while uncertain matches enter a review workflow.
Feed confirmed mappings back into the network — Buyer decisions improve future suggestions and expand the amount of vendor inventory visible and transactable across the platform.
Results and current status
The customer engagement demonstrated measurable improvements in matching quality and delivered a branded review interface with opportunity highlighting and analytics.
Opportunity accuracy increased from 59% to 83% in one evaluation. After applying vendor-product filtering, the true-positive rate improved from 25% to 77%, and opportunity accuracy reached 90%.
On a larger catalog, high-confidence results covered 53% of products at 83% accuracy. The correct vendor result appeared within the top ten candidates for 90% of products, giving reviewers a much smaller and more relevant set of options to inspect.
The work also showed that matching quality alone was not enough. Adoption depended on embedding the workflow into the platform, reducing reviewer effort, and improving upstream data governance.
Following the completed matching program, the customer and Claro evaluated a broader global-catalog and AI-enabled transaction strategy.
Why this workflow matters
Many-to-many platforms compound the catalog problem: every new participant creates another set of mappings.
Without a shared product identity layer, each buyer and vendor relationship requires repeated manual work. A canonical catalog and taxonomy can turn matching from an onboarding task into reusable network infrastructure.
Key takeaway
Claro demonstrated that flower catalogs with weak standards and limited identifiers could be matched with useful confidence, ranking, and domain-specific controls.
The larger opportunity was not only faster mapping. It was the creation of a shared product identity layer that could make more vendor inventory visible, reduce onboarding work, and support transactions across the entire network.
Frequently asked questions
Why is flower catalog matching difficult? Products often lack global identifiers, while commercially important differences such as grade, color family, stem length, and weight are expressed inconsistently.
Can one buyer product match several vendor products? Yes. The workflow can preserve valid one-to-many relationships when several vendor products satisfy the buyer’s commercial specification.
What do top-five or top-ten results mean? They measure whether the correct vendor product appears within the ranked candidate set presented to a reviewer.
Why are confidence tiers important? Confidence allows high-quality matches to progress quickly while uncertain records remain available for human review.
Why is a shared taxonomy valuable? It reduces the need to solve the same buyer-to-vendor mapping repeatedly and creates consistent attributes for onboarding, search, availability, and transactions.
Does matching accuracy alone guarantee adoption? No. The review experience, integration into existing workflows, and quality of upstream product data also affect adoption.
