Introduction

Intershop Copilot for Buyers is an AI-powered purchase and service assistant for digital commerce. It provides a conversational interface in the storefront and helps buyers find products, obtain contextual information, and complete supported service interactions.

The Copilot combines commerce search, customer and account context, approved knowledge sources, and customer-specific tools. It acts as the conversational control layer; the connected commerce, search, content, and customer systems remain authoritative for their respective data.

This concept is intended for implementation partners, solution architects, developers, product managers, and technical administrators.

References

• Guide - Copilot for Buyers Starting Guide

• Concept - Security for Intershop Copilot for Buyers

• Azure AI Search indexers

• Azure Blob Storage indexing with Azure AI Search

• Chunk documents for vector search

• Integrated vectorization in Azure AI Search

• Azure Blob Storage documentation | Azure Storage Explorer

• Intershop Copilot for Buyers in the PWA

Scope and Design Principles

• Commerce-aware: Product availability, prices, categories, customer context, and other transactional information are obtained from the connected commerce services.

• Retrieval-enabled: Approved documents and other knowledge sources can be indexed and used to provide contextual answers.

• Extensible: Partners can adapt flows, tools, retrieval logic, and customer-specific integrations within the agreed service and security boundaries.

• Source-conscious: Answers can include visible source references or provenance links when suitable source metadata is available.

• Environment-aware: Development, testing, and production configurations are kept separate. Environment-specific credentials and variables are not transferred implicitly.

Architecture Overview

The Copilot for Buyers solution consists of the following logical layers:

1. Copilot client: The Progressive Web App extension provides the buyer-facing chat interface and renders supported storefront actions.

2. Conversation and orchestration layer: Flowise Chatflows or Agentflows interpret the request, invoke the required tools, combine results, and prepare the response.

3. Commerce services: Intershop Commerce Management APIs and the configured commerce-search service provide authoritative product, price, availability, account, and order information.

4. Knowledge and retrieval services: Intershop provides Azure Storage and Azure AI Search for customer-specific document ingestion, indexing, vector retrieval, and supported advanced search configurations.

5. Customer-specific integrations: Custom Flowise tools can connect additional APIs, retrieval services, or external knowledge sources.

Copilot Client and PWA Boundary

The Copilot client is deployed through the regular PWA development and release process. Flowise deployment does not deploy or validate PWA changes. Changes that affect both the conversational flow and the storefront user interface must be coordinated across both release processes.

Commerce Search

The Copilot can use the commerce platform's configured search provider, such as Solr or SPARQUE. Commerce search remains responsible for authoritative product filters and commerce-specific relevance signals. SPARQUE uses graph-supported search capabilities internally. This does not expose a general-purpose Knowledge Graph for configuration through the Copilot.

Knowledge Storage and Indexing

Intershop provisions a customer-specific Azure Storage Account and Azure AI Search service as the supported knowledge and vector-search baseline for Copilot for Buyers.

Ingestion Workflow

1. The implementation partner places source documents or prepared chunks in the provided Storage Account. Content can be uploaded manually using Microsoft Azure Storage Explorer or automatically through Azure Storage APIs.

2. The partner configures the required Azure AI Search data source, index, indexer, and optional skillset.

3. For a standard PDF workflow, an Azure AI Search indexer reads the files from Blob Storage, extracts indexable content and metadata, and writes the results to the configured index.

4. Built-in Azure AI Search capabilities can be used for common chunking and vectorization scenarios. Customer-specific preprocessing or complex chunking is implemented and operated by the partner.

5. The partner validates indexing results and configures the Flowise retrieval tool. The Copilot for Buyers blueprint contains an example for querying the provided Azure AI Search service.

Source of Truth and Lifecycle

The provided Storage Account is an ingestion and indexing store, not the customer's source of truth. Authoritative documents, document governance, and source-document versioning remain in the customer's source system.

The implementation partner maintains the customer-specific knowledge base. This includes adding, replacing, and deleting content; defining retention; maintaining indexes and indexers; triggering or scheduling reindexing; and monitoring ingestion, capacity, and search quality. Intershop operates and backs up the provisioned infrastructure.

Capacity

The commercial Copilot size defines the included indexed-document volume. For capacity accounting, each indexed chunk is counted as one document.

A source document, such as a web page or a PDF file, can be up to 10 MB. For example, one PDF split into 40 indexed chunks consumes 40 documents of the included volume. The technical limits of the provisioned Azure services also apply.

Retrieval Capabilities

Supported Standard

Azure AI Search is the vector-search component provided and supported by Intershop. The standard blueprint provides a straightforward vector-search starting point that partners can adapt to the customer's indexed content and retrieval requirements.

Partners can configure customer-specific indexes, indexers, skillsets, query behavior, and Flowise retrieval tools. Intershop support covers the provided Azure AI Search infrastructure. The partner is responsible for the functional quality of the customer-specific retrieval solution.

Advanced Retrieval Pattern

Intershop also provides an optional advanced example that demonstrates a deterministic multi-stage product-search pattern:

1. Parse the buyer's request into search terms, filters, and intent signals.

2. Use text, semantic, and vector retrieval in Azure AI Search to identify product candidates and relevant descriptive context.

3. Apply authoritative commerce filters and relevance signals through the commerce-search service.

4. Fuse the retrieval signals, validate candidate suitability, and rerank the selected candidates with an LLM.

5. Apply configured quality and business rules before returning the final recommendations.

The available example uses Solr for the commerce-search step. An equivalent SPARQUE-based variant can be provided on demand for a customer project. The advanced pattern is an implementation example and is not enabled automatically in every Copilot deployment.

Alternative Vector Stores and Knowledge Graphs

A general provider abstraction for interchangeable vector stores is not part of the standard Copilot for Buyers product. Azure AI Search remains the mandatory component provided and supported by Intershop.

A partner may connect an external vector store through a customer-specific Flowise tool. The partner is then responsible for provisioning, licensing, security, ingestion, lifecycle, integration, monitoring, compatibility, and retrieval quality. The external service is outside the standard Intershop support scope unless separately agreed.

Copilot for Buyers does not provide a native, managed Knowledge Graph. External Knowledge Graphs can be connected as customer-specific retrieval or context sources. The partner is responsible for the integration, permissions, query mapping, result transformation, and quality assurance.

Sources and Provenance

Copilot for Buyers supports visible source citations or provenance links in answers. Partners must preserve suitable source metadata during ingestion and map it to retrieval results. A displayed reference must point to a source the buyer is authorized to access and must not reveal internal storage locations or administrative URLs.

Access Control and Security Boundaries

Intershop grants the implementation partner only the Azure roles required for the agreed implementation and operational tasks. Unrestricted administrative access to the Azure subscription is not provided. Exact role assignments depend on the actions the partner must perform and follow the least-privilege principle.

Credentials and secrets must not be embedded in Flowise code or exported flow definitions. They are maintained separately for each environment. Partners are responsible for protecting customer-specific credentials and for ensuring that connected source systems enforce the required authorization.

For further security requirements, see Concept - Security for Intershop Copilot for Buyers.

Environment and Release Model

Flowise Workspaces

Copilot for Buyers uses separate Flowise workspaces for INT, UAT, and PRD. Development takes place in INT; testing and approval take place in UAT; and the approved configuration is then promoted to PRD. Production should not be edited directly.

Current Flowise Promotion

The current process uses the standard Flowise export and import functions for complete Chatflows and Agentflows. Flow-level nodes, prompts, and tools are transferred as part of the flow definition.

Credentials, secrets, and global or workspace variables are excluded and must be configured separately in every target workspace. After import, the partner must validate environment-specific dependencies, endpoints, permissions, and credentials before activating the flow.

Knowledge Infrastructure Environments

By default, Intershop provides one central Azure Storage and Azure AI Search setup for the customer solution. Separate INT and UAT knowledge infrastructures can be provisioned on demand. The existence of three Flowise workspaces does not imply that three separate Storage Accounts and Azure AI Search services are provided automatically.

Azure AI Search configuration can be reproduced in another environment through Azure APIs or SDKs. Indexes, indexers, skillsets, data sources, and synonym maps are recreated with target-specific connections and permissions. Indexed content is rebuilt from the target source rather than treated as the authoritative deployment artifact.

Responsibilities

Limitations

• Azure AI Search is the supported vector-search baseline. Native multi-vector-store portability is not included.

• Customer-specific Flowise tools, external vector stores, external Knowledge Graphs, and advanced retrieval adaptations are implemented and operated by the partner unless separately agreed.

• The ingestion Storage Account is not a document-management system or source of truth. Product-level authoring, approval, versioning, and records-management workflows are not provided.

• Retrieval and generated answers depend on source quality, source availability, permissions, configuration, and AI-model behavior.

• AI-generated answers and recommendations must be tested for the customer use case. Business-critical decisions should retain appropriate human review and escalation paths.

• The current Flowise export/import does not clone credentials, variables, Azure knowledge infrastructure, or PWA configuration.