Ecommerce

Answer “where is my order?” without a human reading it

“Where is my order” is the biggest ticket category most retailers have, and nearly all of it sits one lookup away from an answer. The part that needs judgement is whether this order is late enough to do something about, and that's a condition on an edge.

A tracking link only deflects the question when the tracking page has good news. The rest of the time the customer writes in anyway, and now someone on your team is reading a carrier feed to decide whether to reship.

40–60%

of inbound ecommerce support contacts ask where an order is. It's the biggest single category.

ShippyPro / Pango, 2026 benchmarks

The flow

What runs when the message arrives.

Every step below is a real node type on the canvas, and the log underneath is the execution it produces. A branch the run didn't take stays dark, same as it would in the console.

agents / order-status7 steps
message
classify
orders.get
carrier.get
estimate
reply
authorize
run_3c7b19order-statuscompleted · 0.74s
  1. 01agent.invokeorder-status · ses_2b71run-sync
  2. 02llm.call2 tool calls372ms
  3. 03tool.httpGET /v1/orders/9134200 · 96ms
  4. 04tool.httpGET /track/1Z9A82200 · 141ms
  5. 05transformeta within promise3ms
  6. 06run_jsformat reply2ms
  7. 07agent.return“arrives Thursday”0.74s

What it uses

What this leans on.

Two lookups at once

The model calls your order endpoint and the carrier's in the same turn. Use a `parallel` container when you want them explicitly concurrent.

The judgement is an expression

Whether an order counts as late is a CEL condition on an edge, reading the values the previous steps returned. Changing the threshold means editing one expression.

A person only for the expensive call

Answering is automatic. Authorising a reship goes to the Inbox, so nobody spends their day reading tracking numbers but somebody still signs off the cost.

Every answer is auditable

Thirty days of traces means “why did it tell the customer Thursday?” is a row you open rather than a theory you argue about.

What it meters

One execution per question. The two lookups are your own APIs and cost nothing here.