Skip to main content
RET-03 · Retail & E-commerce

Store Associate Knowledge Copilot

Outcome​

Answer a store associate's floor question in seconds from current policy, planogram, inventory, and service bulletins, so the customer is served without leaving the aisle or escalating to a manager.

Portfolio attributeValue
Primary personaStore associate
Primary aggregateAssociateAssist
Cosmos DB containers2
Grounding datamanuals, policies, inventory, service bulletins
Development estimateUSD 60-220 monthly dev estimate
Runtime optionsPython FastAPI and C# ASP.NET Core

Productivity hypothesis​

These measures reflect the workflow, evidence, decision boundary, and industry risk for this use case.

Impact typeMeasureConservative internal assumptionCustomer measurement
ProductivityAssociate answer time20-35% reduction(baseline median answer-preparation minutes - assisted median minutes) / baseline median minutes * 100
Quality or riskAccepted answers5-10 percentage-point improvementassisted percentage of answers accepted as correct and useful - baseline percentage

Important: Conservative internal pilot assumptions only; not public benchmarks, commitments, or expected customer results. Validate against a comparable baseline or matched control cohort with mandatory human review.

Rationale: Manuals, policies, inventory, and service bulletins support faster point-of-service answers.

Pilot guardrail: Require staff authority for returns, exceptions, safety matters, and inventory or account changes.

Customer validation plan​

  1. Baseline the current workflow with representative users and tasks.
  2. Run the accelerator on the same task mix with human review retained.
  3. Measure associate answer time and accepted answers with the documented formulas.
  4. Replace internal assumptions with customer-observed results before a business case.

Why Azure Cosmos DB​

This AI solution needs an operational data plane, not only a vector index:

  • Hybrid vector and full-text retrieval combines semantic recall with exact identifiers, codes, names, and domain language.
  • Hierarchical partition keys and metadata predicates keep tenant and domain boundaries in the normal query path.
  • Operational JSON, source chunks, workflow and agent state, feedback, and audit records remain together for low-latency reads and governed writes.
  • Change feed reacts to source and operational updates without polling.
  • Autoscale throughput, configurable consistency, and multi-region distribution support variable demand and enterprise availability.

Domain workflow​

Domain workflow

StageActorInput → outputApproval
1. Ask a floor question on the handheldStore associateSpoken or typed question with scanned SKU → Normalized question bound to store contextPolicy controlled
2. Retrieve policy, bulletins, and stockRetrieval agentQuestion, storeId, and region scope → Ranked evidence with effective-date filterPolicy controlled
3. Resolve bulletin override against standing policyPolicy reasoning agentConflicting policy and bulletin evidence → Single authoritative instruction with precedence notePolicy controlled
4. Deliver a step-by-step floor answerAssociate copilotAuthoritative instruction and stock location → Cited answer with aisle and backroom locationPolicy controlled
5. Escalate exception to the manager on dutyManager on dutyProposed exception such as an out-of-policy return → Approved or denied exception with reasonHuman required

Technical architecture​

Technical architecture

Cosmos DB data model​

ContainerHierarchical partition keyDocumentsSearch paths
associateAssists/tenantId, /storeIdAssociateAssist, StoreInventorySnapshot, TaskHandoffOperational fields only
storeOperationsKnowledge/tenantId, /documentIdPolicyArticle, ServiceBulletin/bodyVectorEmbedding, /instructionVectorEmbedding, /bodyVector, /instructionVector

Query contract​

QueryContainerScopePurpose
Active bulletin check for a SKUstoreOperationsKnowledgecross-partition-boundedKnowledge-container query against storeOperationsKnowledge (bounded fan-out across the /tenantId + /documentId partitions named by the shortlist): confirm whether an unexpired recall or correction bulletin overrides standing policy for the scanned SKU, over the bounded bulletin id set returned by the SKU-to-bulletin index.
In-store stock and locationassociateAssistssingle-partitionTell the associate where the item sits on the floor or in the backroom.
Region-scoped policy retrievalstoreOperationsKnowledgecross-partition-boundedKnowledge-container query against storeOperationsKnowledge (bounded fan-out across the /tenantId + /documentId partitions named by the shortlist): pull the in-force chunks of the policy documents shortlisted by vector search, narrowed to the store's region so a Texas associate never sees a California-only rule, as a bounded fan-out across the at most ten policy document ids returned by vector shortlisting.
Open store task boardassociateAssistssingle-partitionShow tasks created from assists that are still open for this store.

Python and C# services route each document type to its declared container and hierarchical partition key. ETag-based writes protect mutable aggregates, while append-only records and bounded cross-partition queries preserve scale.

Azure deployment​

Azure deployment

Clone the source repository, authenticate with Azure, and run:

Set-Location accelerators/retail-e-commerce/store-associate-knowledge-copilot
az login
./Deploy.ps1 -Runtime python -Location eastus2

Change python to csharp to deploy the .NET implementation.

Explore the implementation​

Return to all accelerators