All case studies

Customer story · Snoonu

How Snoonu handles 14,581 driver chats a quarter inside its rider app

Open’s AI agent is embedded in the Snoonu Rider app in Qatar, answering drivers in English, Arabic, and Roman Urdu. It releases them from stuck orders on its own, and every case it cannot close is measured and categorised rather than quietly dropped.

14K+
Driver chats
Qatar
Live market
3
Languages

Snoonu delivers across Qatar, and the people Open supports are not the customers ordering. They are the riders. A driver messaging support is standing next to a scooter with an order that will not move, and every minute costs them the next job.

Support Where the User Is Already Losing Money

The single largest thing drivers write in about is an order the merchant has not finished. One rider’s message, verbatim: “Now order not ready.” That one situation accounts for 1,591 of the escalations in a three-month census — nearly a quarter of everything handed to a person.

The stakes are not symmetrical either. Releasing a driver from an order before pickup is a routine correction. Releasing one after pickup means an undelivered order in the system and a customer waiting on food nobody is carrying.

And they do not all write in the same language. Snoonu’s riders message in English, Arabic, and Roman Urdu, often switching mid-conversation.

An Agent That Releases Orders, Under Guardrails

Open is embedded as a widget inside the Snoonu Rider app, backed by a microservice that reads live order and driver state. It went live on order-release flows in June.

The release itself is deterministic, not conversational. The agent checks the order state against the API and releases only before pickup. Post-pickup release is barred outright and goes to a human, as does anything touching an order already in delivery.

Where a driver is genuinely unable to continue — a vehicle problem, an ended shift, a health issue — the agent can take them offline. Accidents, conduct reports, and anything contested never trigger that automatically; they route to supervisor review.

Everything it cannot resolve posts a structured escalation into the responsible Snoonu ops group, so a merchant-delay case reaches the merchant team and a driver-side problem reaches supply, rather than a single queue absorbing both.

The Results

Measured across a full three-month census of the rider-support org:

MetricImpact
Automation rate44%
Handoff rate46.1%
Channels coveredIn-app chat
Handoff destinationSnoonu live ops

Open did not estimate the gap; it counted it. All 6,723 handoffs over three months were read and categorised: nearly three quarters happened because an instruction told the agent to hand off, and roughly a quarter because no tool existed for the action the driver needed. That is a roadmap, not a rating.

The Impact

A rider-support line that runs in three languages inside the app the driver already has open, with order release handled end to end and the boundaries drawn deliberately rather than discovered in production.

For the driver standing at a counter with an order that is not coming, the difference is that the release happens in the chat, and they are back on the road instead of waiting for someone to pick up.