Sign In
Agentic product classification — header illustration.

Agentic Product Classification, Built for Scale and Precision

  • High-Precision Classification

    ProductHub is a next-generation, agentic product classification solution that delivers high-precision GS1 GPC brick-level classification at scale. Designed for enterprises managing complex and fast-changing product catalogues, ProductHub combines leading AI models with coordinated agent workflows to achieve classification accuracy that rivals and in many cases exceeds human performance.

  • Contextual Understanding

    ProductHub classifies products using rich contextual understanding, leveraging product name, description, brand, and company information as core inputs. These signals are automatically enriched through a combination of intelligent search and structured data enrichment agents, building a comprehensive product profile before classification occurs.

  • Agentic Reasoning Engine

    At the heart of ProductHub is an agentic reasoning engine that evaluates enriched product context against the GS1 GPC hierarchy, selecting the most appropriate brick with clear confidence scores and explainable reasoning for every decision. This ensures classifications are not only accurate, but transparent and auditable.

  • Scalable & Integrated Platform

    The platform supports single-product classification as well as high-volume batch processing, making it suitable for both operational workflows and large-scale catalogue remediation. ProductHub is available through a modern web UI and API, allowing seamless integration into existing master data, ERP, PIM, and e-commerce platforms.

  • Real-World Validation

    ProductHub has been validated using human-labelled products from the Australian GS1 dataset, with formal evaluations demonstrating strong alignment to expert-assigned bricks. This provides confidence that classifications are grounded in real-world standards, not theoretical mappings. Based on the assessment of expert LLM judges, agents perform significantly better at correctly classifying products than humans.

  • Extensible Framework

    Designed as an extensible agent framework, ProductHub can easily be expanded with additional agents to enrich product attributes, enhance brick-level detail, or support advanced use cases such as regulatory tagging and SNAP eligibility determination. As product, regulatory, and market requirements evolve, ProductHub evolves with them.

How ProductHub Classifies Products Using Agentic Intelligence

ProductHub uses a coordinated set of specialised AI agents to transform raw product inputs into high-confidence, explainable GS1 GPC brick classifications.

The process begins when product details are provided, including Product ID, product name, product description, brand, and company. These inputs form the initial context for the agent workflow.

Illustration of specialised agents enriching product context.

Context Enrichment Through Specialised Agents

Before classification runs, coordinated agents enrich raw catalogue inputs with trusted external context so every decision is grounded in real-world product and category data.

  • Vendor Agent — anchors products to verified company and brand information from authoritative sources.
  • Product Agent — expands and reconciles product descriptions using structured signals retrieved across the web.
  • Hierarchy Agent — injects GS1 GPC hierarchy context so bricks are evaluated against authoritative category logic.

Each agent contributes structured signals so the classifier never works from bare identifiers alone.

Illustration of agentic classification with confidence scores.

Agentic Classification with Confidence

The reasoning engine evaluates the enriched profile against the GPC taxonomy and returns ranked brick recommendations with explicit confidence for every output.

  • Top 3 bricks — the most likely GS1 GPC brick assignments for each product.
  • Confidence scores — clear probability indicators on every recommendation.
  • Ranked outputs — ordered results suited to automated pipelines and human review.

Operators can accept, override, or escalate classifications with full context at hand.

Diagram illustrating explainable reasoning across enrichment steps.

Explainable Reasoning

Every classification is accompanied by explicit rationale tied to the enriched product evidence and hierarchy rules applied during the decision.

  • Transparent — reasoning references the inputs and GPC logic used.
  • Auditable — decisions are traceable for compliance and quality review.
  • Trustworthy — suitable for regulatory and governance-critical catalogues.
  • Consistent — the same enriched context is applied on every run.
Illustration of scalable product data operations.

Scalable and Extensible by Design

ProductHub supports single-SKU classification through high-volume batch remediation, exposed via a modern web UI and API integrations into ERP, PIM, and master-data platforms — without reworking the core agent framework.

Agentic flow — coordinated agents enriching and classifying product data.

Agentic Flow

ProductHub uses a coordinated, agentic workflow to transform raw product information into high-confidence, explainable GS1 GPC brick classifications.

Product inputs are progressively enriched by AI agents, each contributing domain-specific intelligence to build a comprehensive and contextualised product profile before classification.

This layered enrichment approach delivers significantly high accuracy, while generating explicit reasoning at every stage to ensure transparency, auditability, and suitability for regulatory and governance-critical use cases.

Agentic flow flowchart — product enrichment and classification workflow.
Evaluation of performance — ProductHub vs human-labelled GS1 classifications.

Evaluation of Performance

ProductHub vs. Human-Labelled Products

ProductHub was evaluated against 10,000 human-labelled products from the Australian GS1 dataset, comparing agent-generated GPC brick classifications with those assigned by humans.

  • Top-1 Match: The agent and human agreed on the top brick 72% of the time.
  • Top-3 Match: The correct human-assigned brick appeared within the agent's Top-3 recommendations 84% of the time.

Independent LLM Judge Review

Where the agent and human disagreed, independent LLM judges reviewed the classifications using all available product context and GPC hierarchy definitions:

  • In Top-1 disagreements, LLM judges found the agent's classification correct 96–100% of the time, with human labels judged correct only 0–4%.
  • In Top-3 evaluations, LLM judges confirmed the agent's classification as correct 98–100% of the time.

Key insight: Humans make material classification errors in approximately 20–30% of cases, particularly when product descriptions are incomplete, ambiguous, or outdated. ProductHub's agentic approach dramatically reduces these errors by consistently applying enriched context, hierarchy knowledge, and structured reasoning.

Evals Overview

The following section outlines the evals overview used to assess the performance of the ProductHub agent solution. It describes how classification accuracy, confidence, and consistency are evaluated across different scenarios to validate the agent's effectiveness.

This framework ensures the solution performs reliably and at scale before being deployed into production workflows.

Evals overview flowchart — agent evaluation workflow.

Solution Demos

ProductHub fits the way you work today — classify one product at a time in the platform, upload a CSV to run batch remediation across your catalogue, or integrate through our API so your ERP, PIM, or data pipeline calls the same agentic engine. Each path delivers consistent GS1 GPC brick output, confidence scores, and explainable reasoning. Select a demo to see it in action.

Extending ProductHub beyond classification — agentic platform illustration.

Extending ProductHub Beyond Classification

ProductHub is designed as an agentic platform, not a single-purpose classifier. Once a product is classified at the GPC brick level, additional agents can be seamlessly introduced to extend capability without reworking the core system.

Attribute Agents

Using the classified brick as a foundation, Attribute Agents can automatically search, infer, and enrich brick-level and product-level attributes. These agents leverage multiple data sources, including open product databases, images, and targeted search, to progressively build a richer and more accurate product profile.

Specialised Enrichment Agents

This enriched context can then be passed downstream to specialised enrichment agents, such as:

  • Detailed product attribute enrichment (e.g. ingredients, form factors, nutritional indicators, pack types)
  • Enhanced product descriptions and metadata for downstream systems
  • Regulatory and eligibility agents, including SNAP eligibility determination with confidence scoring and reasoning

Each agent operates independently but cooperates through shared context, allowing ProductHub to scale horizontally across new use cases while maintaining consistency, explainability, and auditability.

The result: ProductHub evolves from a high-precision classification engine into a modular product intelligence platform — capable of supporting regulatory compliance, analytics, commerce, and policy-driven decisions, all powered by coordinated AI agents.

Extending ProductHub flowchart — downstream enrichment agents.
Democratising Product Data since 2021
Agentic Product Classification | ProductHub | ProductHub