Skip to main content
RET-02 · Retail & E-commerce

Merchandising Content Studio

Outcome​

Give merchandisers a governed studio that drafts channel-ready product copy and campaign assets from validated attributes, brand rules, and live performance signals, with brand and legal approval before publish.

Portfolio attributeValue
Primary personaMerchandiser
Primary aggregateContentWorkItem
Cosmos DB containers2
Grounding dataproduct data, brand rules, campaigns, performance
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
ProductivityBrief-to-approved-draft time10-18% reduction(baseline median draft-cycle minutes - assisted median minutes) / baseline median minutes * 100
Quality or riskContent rework5-10% reduction(baseline corrected or repeated content-task rate - assisted rate) / baseline rate * 100

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: Merchandisers can create grounded product content faster with fewer avoidable brand and factual corrections.

Pilot guardrail: Keep output editable and require compliance with current product facts, brand rules, and policies.

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 brief-to-approved-draft time and content rework 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. Select style set and channel targetsMerchandiserCategory assortment and launch calendar → ContentWorkItem with channel targetsPolicy controlled
2. Assemble validated attributes and brand rulesGrounding servicePIM snapshot and rule library → Constrained grounding bundlePolicy controlled
3. Generate channel copy variantsContent generation agentGrounding bundle and performance signals → CopyDraft per channel with claim citationsPolicy controlled
4. Run brand, claim, and accessibility checksCompliance checkerCopyDraft set → Rule check report with blocking findingsPolicy controlled
5. Approve and schedule publishBrand and legal reviewerChecked drafts and diff against live copy → Approved copy released to channel feedsHuman required

Technical architecture​

Technical architecture

Cosmos DB data model​

ContainerHierarchical partition keyDocumentsSearch paths
contentWorkItems/tenantId, /categoryIdContentWorkItem, ProductAttributeSet, CopyDraft, PerformanceSignal/attributeSummaryVectorEmbedding, /copyVectorEmbedding, /attributeSummaryVector, /copyVector
brandGovernanceRules/tenantId, /ruleDomainBrandRule/ruleTextVectorEmbedding, /ruleTextVector

Query contract​

QueryContainerScopePurpose
Category work queuecontentWorkItemssingle-partitionShow every style awaiting drafting or review inside one merchandising category.
Draft version historycontentWorkItemssingle-partitionCompare successive copy versions for one style and channel during review.
Blocking rule violationscontentWorkItemssingle-partitionList drafts held by severity-high brand or claim violations before a launch date. The offending ruleId, ruleDomain, and severity are denormalized onto ruleCheckResults when the check runs, so this reviewer query never joins to the brandGovernanceRules container.
Underperforming copy sweepcontentWorkItemssingle-partitionFind live styles whose conversion fell while returns rose, as refresh candidates.

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/merchandising-content-studio
az login
./Deploy.ps1 -Runtime python -Location eastus2

Change python to csharp to deploy the .NET implementation.

Explore the implementation​

Return to all accelerators