Overview

The EdgeML "Flash Deal" feature enables kiosks to utilize Edge AI to enhance customer experience and operational efficiency. Models are trained and executed locally at each kiosk, allowing dynamic and responsive machine learning (ML) applications.

This initial phase focuses on two key models:

1. Product Suggestion Model (Model X)
   It predicts product suggestions based on historical buying patterns for the specific kiosk.
2. Dynamic Discount Model (Model Y)
   It suggests discounts based on stock levels, estimated restock time, and product sales velocity.

Together, these models aim to increase revenue, reduce waste, and boost the sale of items with short expiration dates, such as fresh foods, salads, and sandwiches.

How It Works

Workflow

1. Customer Action:
   A customer selects an item from the fridge during a purchase session.
2. Prediction Step 1 (Model X):
   Predicts other items the customer might want to buy based on historical purchase data.
3. Prediction Step 2 (Model Y):
   Assesses the predicted items (from Model X) against stock levels, restock timing, and sales velocity. Each item receives a score.
4. Suggestion Display:
   If an item's score from Model Y meets or exceeds a configurable threshold:
   • A suggestion is displayed to the customer.
   • The discount level is dynamically calculated based on the score, capped by a maximum allowable discount.

Key Criteria

Factors Affecting Suggestions and Discounts

• Model X: Product Pairing History
  • Higher frequency of co-purchases = Higher score.
• Model Y: Stock Dynamics
  • Stock Level: Higher stock = Higher score.
  • Days Until Restock: Fewer days = Higher score.
  • Sales Velocity: Slower sales = Higher score.

Focus on Expirable Items

The models heavily prioritize items with short expiration dates to reduce waste. However, the system can also suggest package deals for non-expirable items to boost overall sales.

• Items marked as IsExpireable are heavily favored in the algorithm.

Benefits

• Revenue Growth:
  Increased sales from cross-suggestions and targeted discounts.
• Waste Reduction:
  Improved metrics for items with short expiration dates.
• Customer Experience:
  Real-time, personalized offers create a more engaging shopping experience.

Technical Details

Model Training and Execution

• Training:
  Occurs during kiosk reboot in the background.
  • Scope: Limited to items in inventory.
  • Duration: A few seconds.
• Prediction:
  • Latency: <100ms.
  • Impact: Negligible effect on overall kiosk performance.

Settings and Configuration

Localizations

Reporting and Tracking

Receipts and Integration

• Price Type for Line Items:
  • Type 40: Individual line items.
  • Type 41: Reporting purposes.

Validation Metrics

• Monitor the impact through waste tracking.
• Metrics should reflect a reduction in waste due to sales influenced by the new Flash Deal pricing.

Requirements

• Software Versions:
  • DataService: ≥ 0.414
  • InventoryOS: 1.275
  • MailQueueSender: ≥ V147

Flash deal report:
Can be used to track Flash deal effects by showing revenue, waste and flash deal usage per kiosk and month

Flash deal transactions report:
Can be used to lookup transactions that includes a package deal to see what product a package deal occurred on and what was bought together with it