How Local Retail Chains in SG Use AI for Customers

Perishable Forecasting Inside FairPrice DRiVE

FairPrice Group runs roughly 160 FairPrice supermarkets and Finest outlets in Singapore, and its in-house DRiVE engine refreshes recommended order quantities for every store every night. The model blends five years of store-level POS history with calendar events — Chinese New Year, GST voucher payout dates, school holidays — at the granularity of individual SKUs.

Store managers don’t receive a black-box final number. They get a shortlist of suggested adjustments ranked by forecast confidence, plus the reason code behind each one. DRiVE replaced the old “ordering clerk eyeballs yesterday’s sales” routine. On fresh-chicken and ready-meal lines, the result is measurable: stockouts during Saturday peak hours are now rare enough that weekend shoppers at heartland stores aren’t silently shifted to the nearest Cold Storage at Bukit Timah Plaza.

Personalised Vouchers From One CDP

Watsons Singapore and Guardian both run member apps with over a million active profiles each locally. Their recent shift is structural: promotions move from demographic segment batches to per-customer coupon sets generated overnight by a recommendation engine that scores each active member against purchase history, past redemptions, and current inventory overhang.

The workflow is concrete. At 3 AM, the engine writes a unique voucher bundle into each member’s app feed. At checkout, the cashier POS validates the offer through a promo API call that cross-checks both the member ID and the physical basket contents. The metric that matters in regional retail meetings is redemption rate and basket uplift per voucher, not “engagement” or “reach.”

Vision AI Tracks Aisle and Queue Length

The most boring AI deployment in Singapore retail is also the most operationally useful. Chains are running computer-vision models on their existing CCTV feeds at checkout zones. The model detects three specific failure states: queue length per lane, self-checkout machines stuck in “unexpected item in bagging area” loops, and the time it takes for floor staff to respond to a beeping machine.

Output pushes to a floor-leader dashboard in the store, not to a head-office BI portal. One 22-store Singapore grocery operator cut average checkout dwell time from 4.1 minutes to 3.3 minutes across its pilot cluster by tying that dashboard to a simple alert: if a lane queue exceeds three people for over 90 seconds, the duty manager gets a push notification.

RAG Chatbots Handle Post-Purchase Complaints

The two busiest customer-service lanes in local retail — delayed delivery and refund status — are now running on retrieval-augmented generation. These bots don’t pattern-match canned phrases. They pull live order status from the warehouse WMS and from courier APIs (Ninja Van, Grab, Lalamove) and then generate a specific answer about the actual parcel.

When the RAG pipeline fails confidence checks, it escalates to a human agent with the customer’s full message history attached — no “please repeat your enquiry” handoff. Chains running this stack for a full quarter report post-purchase ticket volumes down 60–70%, with the remaining tickets being genuinely unusual edge cases.

A/B Testing AI Per Store Cluster

The adoption playbook in Singapore has settled on one pattern: no company-wide AI rollout. The standard approach is a single cluster — three heartland stores versus three mall stores — run for three sales cycles, benchmarked against the same cluster’s previous-year baseline. After the test window, store managers vote on whether to keep the system at their location. This kills two problems at once: it avoids forcing a system onto a store manager who won’t use it, and it produces a clean, store-level before-and-after comparison.