Toters delivers groceries and food across Lebanon and Iraq. Look at what customers actually type when they open a support chat and there is almost no variety: the opening message is, literally and repeatedly, “Where is my order?”
One Question, Tens of Thousands of Times
In delivery, the dominant contact reason is not a complaint. It is a status check from someone who is hungry and watching a map that has stopped moving. It arrives the moment the estimate slips, and the honest answer is a time, not an apology.
A generic acknowledgement makes it worse. Told “we are looking into it”, the customer asks again four minutes later, and one order becomes three conversations.
Toters also runs this across two countries with separate operations — a Lebanon business and an Iraq call centre — on the same platform and the same contact centre.
Answering With the Order, Not About It
Open sits inside Twilio Flex, Toters’ contact centre, on in-app chat. When the customer asks where their order is, the agent calls Toters’ order service and replies with the actual state and the current ETA — a time the customer can act on.
It writes back into Flex as well. Resolved conversations close the Flex task with a recorded reason, and anything the agent cannot finish is assigned to a human worker through Flex routing rather than left in the queue.
The boundaries are set by what is wired, and Toters is explicit about it. Cancelling an order, issuing a refund or credit, changing a delivery address, reaching the shopper directly — none of those are automated, because none of the write endpoints are exposed yet. They go to a person.
Two triggers force a handoff regardless of what the agent could say: an ETA that has slipped, and a customer who is visibly frustrated. Both are treated as human moments by policy.
The Results
Measured across the two production organisations, Lebanon and Iraq combined:
| Metric | Impact |
|---|---|
| Automation rate | 44% |
| Markets | Lebanon and Iraq |
| Channels covered | In-app chat |
| Handoff destination | Toters agents in Flex |
Over six thousand sessions a week resolved without a person, on a channel where the question is nearly always the same one. The remaining volume is not a mystery either: it is concentrated in the actions that have not been wired yet, which is a build list rather than a limitation of the answering.
The Impact
Six thousand AI-resolved sessions a week across two countries, inside the contact centre Toters already runs, without scaling headcount proportionally with order volume. Twilio Flex stayed in place; nothing was migrated.
For the customer watching a delayed order, the difference is a specific arrival time in the first reply instead of a queue position — and for the agents, a floor that is no longer mostly people asking the same question.
