There's a particular kind of bad morning where the first sign of a fulfillment problem is a customer email, already annoyed, asking where their order is. The order's been late for days. Nobody noticed, because the information needed to notice was split across a store platform, a logistics system, and a support inbox, with no single place showing all three at once.

This is the normal state for most fulfillment operations, not a failure of anyone's attention. The systems weren't built to talk to each other, so the only thing that reliably connects them is a person checking all three — and checking all three, for every order, every day, isn't something a person can actually sustain.

A flat deadline is the wrong rule

The instinct fix is a single SLA: flag anything late by more than X days. But a flat rule treats every carrier and every supplier the same, which means it either cries wolf on a supplier that's reliably a little slow, or misses a real problem with a supplier that's usually fast. The right comparison isn't a company-wide cutoff — it's each order against what that specific carrier and that specific supplier actually promised.

What real triage looks like

Someone who's good at this doesn't treat every delay the same. They flag against the carrier's own promised window, not one universal deadline. They check whether a delayed order already has an open support ticket, and if it does, escalate before a second complaint arrives, not after. They apply supplier-specific thresholds — a historically reliable supplier gets flagged sooner than a known-slower one, because the same lateness means something different for each. They track the on-time rate on a rolling weekly window, not monthly, because a monthly number catches a developing problem weeks too late. And they route returns by reason — a defect goes to a different queue than someone simply changing their mind, because one is a supplier problem worth tracking and the other isn't.

Agencize learns exactly that as a playbook — not a fixed SLA, but the actual triage judgment captured from watching how someone handles fulfillment issues across systems, including the exceptions they make ("don't flag Supplier B until day four, they're always a little behind"). Once learned, it stays connected to the store, the logistics data, and the support tool together, watching for the same patterns automatically. Delays get caught against the right threshold, escalations happen before the second complaint, and the only orders that reach a person are the ones that actually need a decision.

The customer email never gets written, because the problem got caught first.

See how a playbook gets learned, or see this exact use case running as an Instant App.