“Where is my order?” is the most expensive cheap question in a small business. It takes ninety seconds to answer and it arrives forty times a week. Each one pulls a person away from the conversations that actually change revenue: the quote, the upsell, the customer who is about to leave. Frontline staff end up working as a human FAQ.
What an order-status agent actually does
A useful agent does a narrow job well. It verifies who is asking, pulls the live order and tracking record from your order system, answers in your brand’s voice, and hands the conversation to a person, with the full chat attached, the moment the request goes outside its scope.
The design choice that matters is grounding. A generic chatbot that answers from a pile of FAQ text will deflect. An agent connected to real order and policy data will resolve. The value comes from settling the actual transaction, not from making the question go away.
The enterprise proof point, with a caveat
Salesforce reports that L’Oréal runs an order-status agent on its YSL Beauty and Kiehl’s sites, available around the clock through live chat, with handoff to a human advisor when needed. Salesforce cites a 64% reduction in case handling time for L’Oréal’s care advisors (Salesforce customer story). The wording of that metric differs between regional versions of the page, and the source is the vendor describing its own customer, so treat the number as an illustration of direction rather than a benchmark for your shop.
The principle and the tool
The principle: automate the answer, protect the judgment. The tool is a helpdesk connected to your order system, such as Gorgias (built for online stores) or Intercom. The mechanism is the same in both: a lookup against live order data, a short list of approved answers, and a handoff rule.
A version that fits a small business
You do not need an enterprise platform to start. You need three things:
- A connection to your order data, read-only to begin with.
- A short list of intents the agent may handle: order status, delivery window, return policy, store hours.
- An escalation rule: anything about money, a damaged item, or an upset customer goes to a person immediately.
Start by reading your last 100 support messages and tagging them. Most owners find that a large share are the same three questions. That share is your automation ceiling, and it is measured from your own inbox rather than from a vendor slide.
Boundaries
Automation owns identity checks, data lookups, routing, and the transcript. Gen AI owns phrasing and summarising. Humans own refunds, exceptions, and any decision that costs money. An agent that can issue refunds on its own is a liability, not a feature.
Every status question your team answers by hand is a conversation about something else that did not happen. What share of your support inbox is “where is my order”?