How an Indonesian quick commerce brand started keeping its delivery promise

About the brand

An Indonesian quick commerce retailer delivering FMCG groceries, fresh produce, chilled goods and household care across Greater Jakarta and a second metro area. It operates a dense network of neighbourhood convenience stores and has converted a growing share of them into fulfilment nodes, alongside a smaller number of purpose-built dark stores.

Each node serves a two to four kilometre radius. In normal conditions customers are promised delivery in under thirty minutes. Riders are a core in-house fleet, topped up with gig capacity at peak.

Background

In quick commerce the product and the promise are the same thing. Nobody chooses the channel for range or price. They choose it because they need milk and cooking oil within the half hour, and they abandon the app the third time it fails them.

The company had one advantage most quick commerce operators lack. It already owned hundreds of small shops in exactly the neighbourhoods it wanted to serve, already stocked and already staffed. What it did not have was any way to make them fulfil an online order.

The promise was not holding either. The demand side worked well, but the layer between an order arriving and a rider reaching the door did not exist. Supervisors assigned riders by phone and group chat, oldest order first, with no view of rider position or which drops sat near each other. Ten kilometres across Jakarta takes more than twenty-five minutes on average and far longer in rain, and much of the city is navigated by kampung and gang rather than street number, so riders lost minutes to phone calls at the end of unfamiliar trips.

Inside the dark store, pickers worked from paper lists, expiry discipline depended on the shift, and stock records drifted far enough that the app kept selling items that were not on the shelf.

The retailer partnered with Fynd to rebuild how orders are promised, picked, batched and assigned.

Challenges, solutions and impact

Challenge #1: Dispatch decided by hand

A supervisor cannot solve an assignment problem across thirty live orders and a dozen riders every minute. Orders went out oldest first, so a drop two hundred metres from an idle rider waited behind one across a toll road. Batching, the biggest single lever on delivery cost, happened only when someone noticed two nearby addresses. When volume spiked after payday, the queue lengthened and every order in it broke its promise together.

Solution. Fynd TMS replaced manual dispatch with continuous automated assignment, re-evaluating every pending order against every available rider on a rolling cycle and solving for the assignment that keeps the most orders inside their windows. Clustered drops are batched, typically two or three, constrained by what fits in a rider box alongside chilled items. Estimated times come from the company's own rider GPS history rather than car-derived traffic feeds, because motorcycles filter through gridlock and use alleys absent from road graphs. Delivery points are stored as confirmed pins, so a location found once is found instantly again.

Impact. Delivery timelines fell 23%, and 85% of orders now arrive inside the promised window. Last-mile cost per order dropped 28% as batching and reduced idle time raised drops per rider hour. Supervisors moved from running dispatch to managing exceptions.

Challenge #2: A promise made before anyone knew it could be kept

Every customer saw the same countdown regardless of distance, time of day, pick queue depth or weather. It was a marketing number, not an operational one, so orders were accepted that could not be delivered on time and the failure surfaced only once the customer was already waiting.

Solution. Fynd OMS computes the promise at checkout from live inputs: the pick queue at the serving dark store, rider availability, distance, current conditions and actual stock for every item in the basket. Where catchments overlap, orders route to whichever site can genuinely serve them fastest rather than the one that is nominally closest. During surges the quoted window widens rather than silently breaking, and capacity limits stop a site absorbing more orders than it can dispatch.

Impact. Broken promises fell sharply and refunds declined. A customer offered an honest thirty-five minutes in a downpour proved far more tolerant than one promised twenty and delivered in fifty.

Challenge #3: The clock starts inside the store

Roughly a quarter of the elapsed time on a late order was gone before a rider touched it. Pickers walked inefficient paths from paper lists. No rule governed which batch of milk to pick, so short-dated stock aged on the shelf while fresher units went out first. Infrequent cycle counts let recorded and actual stock diverge across a week.

Solution. Fynd WMS put structure into dark store operations. Picking follows an optimised path with handheld guidance. Batch and expiry data is captured at receiving, and FEFO rules direct pickers to the shortest-dated acceptable unit automatically. Chilled lines are picked last and staged cold. Short picks suppress the item in the app immediately, and a rolling cycle-count cadence keeps records close to physical stock.

Impact. Pick time per order fell materially, giving the last mile more of the thirty-minute budget to work with. Write-offs on short-dated stock declined. Substitutions caused by phantom stock became rare.

Challenge #4: Turning shops into fulfilment nodes without breaking them

Building dark stores from scratch is slow and expensive, and the company already owned the property. But a shop is not a warehouse. Its stock sits on display for walk-in customers, its aisles are sized for browsing, and its two or three staff are already serving a till. Early attempts had pickers competing with shoppers for the same aisle at the same hour, online orders drawing down stock the shelf label still promised to the person standing in front of it, and store managers judged on walk-in sales quietly deprioritising online picks.

Solution. Fynd runs converted stores and purpose-built dark stores on one model. A single inventory pool per site serves the till and the app, so a unit sold at the counter disappears from the app in the same moment. Each site carries its own capacity ceiling by hour, so a small shop with two staff is never handed a queue sized for a dark store, and overflow routes to a neighbouring node. Pick sequences avoid the busiest aisles during peak trading hours. Store performance is measured on combined walk-in and fulfilled revenue, so managers are not penalised for picking.

Impact. The company added fulfilment capacity at a fraction of the cost and lead time of building dark stores, on property it already held. Converted stores reach useful order volumes within weeks of going live, and coverage now extends into neighbourhoods that could never have justified a dedicated site. Walk-in trade held up, because picking load is capped rather than absorbed.

Way forward

Next on the roadmap: converting the remainder of the store estate that qualifies, demand forecasting per node so assortment reflects the neighbourhood, predictive rider positioning ahead of lunchtime, payday and Ramadan peaks, and a tighter promise in the densest zones.

If your challenges resonate with those in this story, it's time to schedule a call with our experts. Whether you're automating dispatch and routing, making a delivery promise you can keep, or turning existing stores into fulfilment centres, our solutions are designed to streamline your operations and drive growth.

With over a decade of experience in omnichannel commerce and a proven track record with 2,300+ brands, Fynd is here to turn your challenges into success stories. Let's drive your growth together.

Book a demo