Give retail support agents a reliable answer when an order arrives in several packages, including partial deliveries and missing tracking.
A customer orders shoes, a coat and a belt. The shoes arrive on Monday. The other items leave a different warehouse on Tuesday. The CRM shows one order and one delivery date, so the service agent tells the customer the order was delivered. The customer opens a missing parcel claim for the entire purchase. Two teams now have to untangle a problem that began with the way the order was shown to support.
If you are assessing retail AI that works with a CRM, run a split shipment through the demo. Ask what the support agent sees for each item and what it can tell the customer while the second package is still moving. A clean answer requires more than a tracking link attached to the order.
Give each quantity a delivery state
Shopify’s Fulfillment reference exposes fulfillment line items, tracking information and event history. A connector can use those relationships to represent fulfillment separately from the order summary. Start by mapping the fields your staff can explain. Keep the order line identifier, shipped quantity, fulfillment identifier, tracking details if supplied and the last observed event time. Avoid grouping by product name. A customer can order two identical items and receive them in different boxes.
Imagine that three units from one line were ordered and two were fulfilled. If one of those two units was in a delivered package, the support reply needs to account for all three units. Show the agent one delivered, one in transit and one awaiting shipment. The CRM can keep its familiar order label at the top while the detail view explains what that label leaves out. Keep the source times visible so staff can distinguish an old carrier event from a fresh read of the commerce system.
Decide which orders the agent can explain
The following routing table works across a manual support screen and an AI-assisted reply flow. Review it with the warehouse team, which will know whether a label scan usually precedes pickup by hours or days.
| What the order view shows | Reply the agent can support | Route when the customer needs more |
|---|---|---|
| Some items delivered and the remainder in transit | Name the items in each package and share the observed tracking state | Open an item-level case if a shipped item is missing |
| Label exists with no carrier scan | Confirm that tracking has been created and pickup is unconfirmed | Ask fulfillment to check the handoff after its normal cutoff |
| Replacement and original share a product name | Explain the replacement only after linking it to its own shipment | Ask a person to reconcile inconsistent shipment IDs |
| Portal says delivered but claim names one missing unit | Acknowledge the item and report the package event as recorded | Pass the unit count and package ID to claims |
| One package number is linked to multiple lines | Present a single package update with its item list | Consolidate tracking checks for those lines |
Treat a promised date differently from an estimated arrival or an observed event. A replacement promise can trigger a credit or return decision, depending on your own policies. The connector should supply what it observed and the answer builder should copy only facts it can trace. If an upstream source returns a blank date, use wording that lets the customer know when the next update will come.
Make case creation item-aware
After the first reply works, test the follow-up. Suppose the customer received the shoes but the coat never appeared. An employee should be able to open a case for the coat without selecting the belt or retyping the delivery address. Give the support system a place to store the selected order line, quantity and shipment. Those identifiers let the warehouse answer the claim even if the product name or display status changes later.
Also test the case where a customer returns one unit from the first box and asks about the second box in the same conversation. Keep the return and shipment states visible together while making each action explicit. A return on one unit should not close the remaining package inquiry. If the customer asks for a refund, bring the policy and approval step into the agent’s view before any action is available.
Put an awkward order in the acceptance test
Build a test purchase with two units of one SKU and a different item. Split the first SKU across two deliveries, delay one tracking feed and give the other package a delivered scan. Ask the support tool for three answers: what arrived, what is still coming and what is known about the delayed package. Save the exact replies. Then let the delayed feed catch up and ask again. The reply should change only where the underlying shipment evidence changed.
Before turning on customer-facing automation, have staff try the same case directly from the CRM without opening a second tab. Record how they find the package and whether they can explain the distinction aloud. Let two different staff members read the reply before approving the template. If they give different interpretations of which item was delivered, the screen still needs work even when every API call succeeds.
Run the pilot across more than one carrier if the store uses several. One carrier may publish an acceptance scan immediately while another records the event on a later batch. Give agents a separate display for the time the last event happened and the time the store retrieved it. At the end of each shift, sample one case that required a supervisor. Trace the decision back to the two timestamps. That will tell you whether to fix a stale integration or align the warehouse and support teams on what the scan means.
Emerging Stacks builds item-level workflows through ecommerce software development for ecommerce teams. Contact Emerging Stacks with a representative order and the service screens your team uses to answer shipment questions.
Ready to work with us?
Get in Touch


