Delivery and mobility / Ride-hailing and delivery
Grab: Using an LLM to classify missing or wrong item reports on GrabFood orders and calculate partial refunds
- Company
- Grab
- Country
- Singapore
- Adoption stage
- In operation
- Source published
- Date basis
- The date the source was published. It can differ from the date adoption started.
- How the source was checked
- Read the full source text
The work problem
According to an article Grab published on August 14, 2026, missing or wrong item (MIWI) issues are among the most commonly reported support problems in its app. A meal combo is stored in the database as one bundled item, so an individual item such as a missing muffin has no price of its own. To err on the side of consumer satisfaction, the system often refunded the entire item, which created unfair costs for merchants, while sending every partial issue to human agents created long waits for consumers and backlogs for support teams.
Technology and data
Grab first added an entry point in the app for reporting partially missing items with photos and descriptions, and routed these tickets to human agents to test user behavior. An LLM now reads the user's description and classifies the problem, sending safety or quality risks such as mold, undercooked food or foreign objects straight to a human agent. For partially missing items, the LLM scans the merchant's full menu for a standalone price of the missing item, finds the closest equivalent to calculate a proportionate refund when the item is not sold separately, and classifies the case as a full refund when an essential component such as the broth for a noodle soup is missing.
Results
In the test phase, around 20 per cent of all missing item reports were for partially missing items, and consumers honestly reported the exact missing items. Grab said it has rolled out this LLM process in most of its markets, and that with improved resolution accuracy its support teams already handle roughly 32,000 fewer manual tickets per month. The company said merchants are not over-penalized while consumers get fast, accurate reimbursement without waiting for manual agent review.
Limits and open questions
The article does not give the accuracy of the refund calculations or any change in merchant disputes.
Sources
- How we taught LLMs to solve the "missing muffins" mysterygrab.com, Accessed
Compiled from public sources. These are not results from ATF Works customers.