Amplify Brands in Real Time for Retail Growth

Schema Design Process

To prevent schema and performance issues as your app grows, start with sound data modeling.

In this solution, you model a unified data layer around four core collections:

• products: Product catalog (names, descriptions, categories, identifiers).

• inventory: Stock levels, restock data, predictive metrics.

• stores: Store metadata (name, location, operational details).

• brandAmplification: Per-store brand/category boost rules.

Follow this process to implement brand amplification and scale reliably.

1. Identify Your Application Workload

   Identify your most frequent operations.

   Use your top queries to choose effective indexes and reduce database calls.

   Assume both e-commerce and stores consume the same operational data.

   Connect customer intent with brand strategy in real time and keep inventory consistent across channels.

   Use the following table to estimate Leafy Associate workloads and treat this estimation as your baseline for read/write priorities:

   Action	Query Type	Target Collection	Frequency	Priority	Notes
   In-store product discovery. Find products that match what the customer asks for (text / intent / hybrid search), showing if it’s in stock in this store right now. (50 stores)	Read	Products, inventory	~500K/day	Critical	Main associate flow. Embed per-store inventory summary flags in each product to avoid $lookup
   E-commerce search. Search by specification, intent, or hybrid (6K products)	Read	Products	~1M/day	Critical	Drives online discovery and must remain low latency. Consider dedicated or read-isolated search nodes
   Apply brand boosts for ranking	Read	brandAmplification	~300k/day	High	Store-level brand boosts applied at ranking time; no product updates required
   Update live inventory and predictive metrics (ingest)	Write	inventory	~700k/day	Critical	Single view from inventory systems of record (stores and warehouses). Small, frequent writes
   Find nearest alternative store with stock	Read	inventory, stores	~100k/day	Low	Geospatial rank by distance
   Configure brand amplification (update boost strategy)	Write	brandAmplification	>500/day	Low	Per-store brand boosts with optional category scope; runtime adjustable, no product rewrites.

2. Map Your Schema Relationships

   The workloads you identify influence how you design your schema.

   In brand amplification, your relationship model determines read latency in the Leafy Associate and e-commerce apps, and affects the cost of write operations in inventory systems.

   Optimize Queries

   To improve in-store product discovery, embed inventorySummary[] within products using the embedding pattern. This approach returns product details and availability in a single read operation and avoids expensive $lookup operations. The code snippet shows an example document in the products collection.

   { "_id": "685bfe2b3d832cf7e1614577",
     "productName": "Onion",
     "brand": "Fresho",
     "price": { "amount": 31.02, "currency": "USD" },
     "category": "Fruits & Vegetables",
     "subCategory": "Potato, Onion & Tomato",
     "aboutTheProduct": "Onion is a versatile vegetable used in salads and curries.",
     "embeddingText": "Onion | Fresho | 5 kg | Fruits & Vegetables | ...",
     "multimodalEmbeddingVector": [0.03, 0.12, 0.07, "..."],
     "inventorySummary": [
       {
         "storeId": "store-019",
         "sectionId": "S01",
         "aisleId": "I11",
         "shelfId": "SH111",
         "inStock": true,
         "nearToReplenishmentInShelf": false
       },
       {
         "storeId": "store-027",
         "sectionId": "S02",
         "aisleId": "I21",
         "shelfId": "SH213",
         "inStock": true,
         "nearToReplenishmentInShelf": true
       }
     ]
   }

   For optimized operational inventory single view queries, keep detailed inventory data in a separate collection and reference the product. Store one document per product to maintain a single view across channels and stores, and update stock in one place. The following code snippet shows an example document in the inventory collection.

   { "_id": "6863fc8057105b3a8ccb57cd",
     "productId": "685bfe2b3d832cf7e1614577",
     "storeInventory": [
       {
         "storeId": "store-015",
         "storeName": "Trikasemmart - Chiang Mai",
         "coordinates": [98.985583, 18.7929],
         "shelfQuantity": 22,
         "backroomQuantity": 0,
         "inStock": true,
         "predictedStockDepletion": "2025-10-23",
         "nextRestock": "2025-10-19"
       },
       {
         "storeId": "store-006",
         "storeName": "Vinyuvanichkul - Nakhon Ratchasima",
         "coordinates": [102.104829, 14.964336],
         "shelfQuantity": 27,
         "backroomQuantity": 49,
         "inStock": true,
         "nextRestock": "2025-10-21"
       }
     ],
     "updatedAt": "2025-10-17T00:00:01Z"
   }

   To find nearby stores with available stock, save store coordinates in a GeoJSON Point to run geospatial queries. The following code snippet shows an example document in the stores collection.

   {
     "_id": "684aa28064ff7c785a568aca",
     "storeId": "store-001",
     "storeName": "Chaihirankarn, Bunlupong and Chomsri - Surat Thani",
     "location": {
       "type": "Point",
       "coordinates": [
         99.317989,
         9.133786
       ],
       "address": "3597 Methavorakul Tunnel Suite 886",
       "city": "Surat Thani"
     },
      …
   }

   Data Consistency

   Use Atlas Triggers to sync the needed fields from inventory and keep products.inventorySummary[] consistent.

   Optimize Brand Ranking Rules In Each Search

   Apply brand boosts at ranking time and store rules in brandAmplification. Promote strategic brands without rewriting product data.

   {
     "_id": "68e43c6865c4a374ea78b4b0",
     "storeId": "store-015",
     "brandName": "Teamonk",
     "categories": ["Beverages", "Gourmet & World Food"],
     "boostLevel": 1,
     "effectiveFrom": "2025-10-01",
     "effectiveTo": "2025-10-31"
   }

3. Indexing

   Create indexes on your most common query fields. View the ‘Build the Solution’ section for details. Monitor and adjust index usage as you scale.

Architecture Flow

View the architecture diagram to understand how the brand amplification solution operates.

Figure 2. The architecture illustrates how MongoDB’s ODL integrates the Leafy Associate application, Advanced Search Microservice, and Voyage AI embeddings. It runs full-text, vector, and hybrid searches within a single Aggregation Pipeline, combining real-time brand configuration and customer intent in a unified, modern data platform.

0. Siloed data

Operational data often lives in silos such as Point of Sale (POS), Enterprise Resource Planning (ERP), and marketing. Unify it to deliver consistent, real-time experiences.

1. Store unified data in MongoDB’s Operational Data Layer (ODL)

The ODL uses the full power of MongoDB’s flexible document model, to store structured, semi-structured, and unstructured data together in one place.

The modern data platform delivers performance, horizontal scalability, and advanced capabilities. It enables fast integration, low-latency queries, and real-time operations at scale.

This unified approach keeps every system—from e-commerce to in-store apps—running on the same trusted, up-to-date information.

2. Store associate enters a search query

The store associate uses the Leafy Associate application to enter a query. They select full-text (MongoDB Atlas Search), intent (Vector Search), or hybrid (rank fusion or score fusion).

In parallel: Real-Time brand amplification configuration

Store managers and authorized personnel use the Leafy Associate application to create per-store boost rules in the brandAmplification collection, for back end use.

As they configure brand amplification, they can observe in real time the metadata that shows the number of products and how many of them are in each category. This information comes from the $searchMeta pipeline stage.

3. and 4. Leafy Associate builds the API request

The Leafy Associate application composes a search request with full context. It builds runtime parameters directly in the front end and retrieves the active brand configuration for that store from MongoDB. Send the request to the Advanced Search microservice.

POST /api/v2/search
{ "query": "green tea skin care",
  "storeObjectId": "684aa28064ff7c785a568aca",
  "option": 4,
  "page": 1,
  "page_size": 20,
  "weightVector": 0.5,
  "weightText": 0.5,
  "fusionMode": "rrf",// you can also use "scoreFusion"
   "brandAmplification": [
    { "name": "Innisfree", "boostLevel": 1 },
    { "name": "Olay", "boostLevel": 2, "categories": ["Face Care", "Skincare"] },
    { "name": "The Body Shop", "boostLevel": 3 }
  ]
}

5. Embeddings service with Voyage AI

When the search type is vector or hybrid search, the microservice generates customer query embeddings with Voyage AI.

6. and 7. Advanced Search microservice builds the aggregation pipeline

The Advanced Search microservice builds the aggregation pipeline at runtime and executes it within MongoDB. The microservice uses a single pipeline and a single round trip to improve performance and efficiency.

For text relevance using $search, the pipeline applies a compound structure that combines must, should, and filter clauses with fuzzy matching and field boosts to refine results.

For semantic intent using $vectorSearch, the pipeline compares the customer query against precomputed embeddings of product data.

In hybrid search, the microservice fuses text and vector results using either $rankFusion, which applies rank-based reciprocal rank fusion (RRF), or $scoreFusion, which blends normalized scores with tunable weights at runtime.

Brand Amplification improves result relevance through two complementary paths:

1. Pre-fusion text boost (inside $search)

   Inject rules as compound. should clauses with score. boost for each brand and (optional) category.

2. Post-fusion multiplicative using $set + $multiply

The microservice preserves store context without merging other collections using $lookup by filtering directly on top Products.inventorySummary[] with $filter. Finally, it produces a UI-ready response with $facet, and returns both the resulting documents and the total count.

8. Reliability, Scalability, Security and Real-Time Consistency

MongoDB maintains high availability with replica sets and scales horizontally with sharding. Use dedicated Search Nodes for low-latency reads. Keep data consistent in real time with Change Streams and Atlas Triggers. MongoDB provides comprehensive, end-to-end security through built-in access control, network isolation, and encryption across data at rest, in transit, and in use.