Order status is among the most automatable support questions, and easy to get subtly wrong. The agent reads the order system and the carrier's tracking, tells the customer where each item is and what the carrier estimates, and never promises a delivery date. Exceptions, such as lost, damaged or disputed deliveries, go to a person.
What is the order status process, and who owns it?
"Where is my order?" arrives in every channel, at every hour, and spikes around sales and holidays. The answer usually exists: the order system knows what was ordered, picked and dispatched, and the carrier knows where the parcel is. The work is joining those two records accurately and saying what they mean in plain language.
Customer support owns the conversation. Fulfilment owns the order record. The carrier owns tracking data, and the business owns the relationship with the carrier. Refunds and replacements belong to a team with authority to give them. The agent answers status at depth 3, a case set out in can an AI agent answer order status questions?, and stays inside the limits our guide to the AI support agent that knows when to say no describes.
What are the steps, and at what depth?
- Agent, depth 1: identifies the order from the customer's account, order number or email, with enough verification to avoid revealing another person's address or purchase.
- Agent, depth 1: reads the order record: items, payment status, picking and dispatch status, any split shipments.
- Agent, depth 1: reads carrier tracking for each shipment: latest scan, location, carrier estimate.
- Agent, depth 3: reports status per item. "Two items were dispatched yesterday with the carrier, who estimates delivery on Thursday. The third is still being picked."
- Agent, depth 3: makes permitted changes within your rules, such as cancelling an item not yet picked, where certified for that capability.
- Agent, depth 1: spots exceptions. No tracking movement beyond your threshold, a "delivered" scan the customer disputes, a returned-to-sender scan, a damaged parcel.
- Agent, depth 2: drafts the exception case with the order, tracking history and the customer's account, for a fulfilment adviser.
- Person, fulfilment adviser: decides on replacement, refund, carrier investigation or a goodwill gesture.
- Agent, depth 3: sets an update for the customer when the next tracking event arrives, if they ask for one.
| Step | The agent does | Depth | Stays with a person |
|---|---|---|---|
| Identify | Finds and verifies the order | 1 | Nothing |
| Order | Reads order record | 1 | Nothing |
| Tracking | Reads carrier data | 1 | Nothing |
| Report | States status per item | 3 | Nothing |
| Change | Cancels unpicked items within rules | 3 | Anything after dispatch |
| Exceptions | Spots them | 1 | Nothing |
| Draft | Prepares exception case | 2 | Replacement or refund decision |
| Update | Sets a notification | 3 | Nothing |
Which controls matter, and what evidence does each leave?
Estimates are labelled. "The carrier estimates Thursday" is accurate; "it will arrive Thursday" is a promise the business cannot keep. Evidence: the wording in sampled replies and scope-gate results.
Order data before carrier data. If the order system says not dispatched, a stray tracking number is not evidence of dispatch. Evidence: the sources cited in each status answer.
Verification before detail. Order number alone should not reveal a delivery address. Evidence: verification events against each disclosure.
Changes stay within rules. The agent may cancel only items in states your rules allow. Anything later is refused or held. Evidence: verdicts naming the order state that decided.
Exceptions carry the full history. The adviser should not have to ask the customer again. Evidence: the case record.
What should you measure?
- Status questions resolved without a person, by channel.
- Repeat contacts on the same order. Several "where is it?" messages mean the answer was unclear or the order is stuck.
- Exceptions spotted by the agent, against those raised later by an angry customer.
- Accuracy of status answers, sampled against what happened.
- Contacts avoided by updates, measured as fewer inbound questions on orders where an update was set.
Where does it usually go wrong?
Carrier data is late. Tracking scans lag reality. The agent should say when the last scan was, not only where.
Split shipments confuse customers. "Your order has been delivered" when only one parcel arrived starts a complaint. Report per shipment.
"Delivered" is treated as final. A delivered scan the customer disputes needs investigation, not an argument. Draft the case.
Payment and order status are mixed. "Why was I charged if it hasn't shipped?" is half billing; answering billing questions with an AI agent covers that half.
Recurring orders are forgotten. Subscription and repeat orders have their own schedule; customers asking about "this month's box" need the schedule as well as the shipment.
Peak periods change the answer. During a sale or a carrier backlog, normal estimates stop holding. If fulfilment has published a revised timescale, the agent should use it and say why. If it has not, the agent should report the carrier's estimate with a note that delays are known, rather than repeat a figure everyone knows is optimistic.
Where Pryme fits
In Pryme, the order system is a sealed source and carrier tracking arrives signed by the carrier, so the agent can show which fact came from where when they disagree. Cancelling an unpicked item is a separately certified capability at depth 3; anything after dispatch is outside the agent's role scope and is refused or held, with the verdict filed.
Questions people ask
Can the agent change the delivery address after dispatch?
Usually not directly, because the parcel is with the carrier. It can tell the customer what the carrier allows and draft a request. Before dispatch, an address change can be routine if the customer is properly verified.
Should the agent offer compensation for late orders?
No. It reports the delay and routes the request for a goodwill gesture to a person with authority to give one.
Is claim status the same problem?
The mechanics are similar, but a claim is a decision about money owed under a contract. Answering claim status questions from policyholders with an AI agent explains why the boundary there is tighter.