Your team already sees the exceptions. That's not the bottleneck.
Your team already knows which shipments are broken. The expensive part is everything that happens after that.
Most tracking stacks are good at flagging. A carrier posts an exception scan, a dashboard turns it red, someone gets an alert. If you need the basics on what a delivery exception is, start there — this article picks up right after detection, with what happens once a shipment exception lands in your queue. Then a human opens the order, opens the carrier portal, reads the tracking history, decides whether this package warrants a customer email, and writes it.
That middle stretch is where the hours go.
An exception handling agent is an AI system that investigates flagged shipments, determines whether intervention is needed, and drafts the resolution for human approval.
Across 963 million shipments tracked between December 2025 and March 2026, roughly 6.2% hit some form of exception. That number is volume-weighted toward the largest shippers, so the typical merchant sees worse: a median rate of 7.7% per organization. At 50,000 orders a month, that is 3,800 to 13,000 cases needing a decision from someone.
Teams generally cover that ground one of two ways, and both hold up right until volume arrives:
Rule-based workflows fire a predetermined response whenever a shipment enters a given status. They are fast and consistent, and they treat a two-day weather delay that will clear on its own exactly like a package that slipped into a return flow three days ago. The rule sees a status. It cannot see severity, order value, customer history, or whether the problem is about to solve itself.
That last blind spot is the expensive one. Sorted by how they actually behave, exceptions fall into roughly three groups: about half are transient, meaning delays and customs holds that arrive late but arrive; around 15% genuinely require someone to act before the shipment moves again; and about 12% are terminal, where the package is gone and the only question left is claim, reship, or refund. A rule that fires on all three wastes your team's attention on the half that would have resolved itself, and gives the terminal cases the same urgency as a weather delay.
Manual investigation restores the judgment rules lack and charges you for it in labor. Teams report 10 to 15 minutes per case: pulling tracking updates, reviewing the order, sometimes calling the carrier. Quality shifts with whoever is on shift, and once the case closes, what your team learned goes nowhere.
Neither approach gets cheaper as you grow. Both cost more per order, not less.
An exception handling agent is built for that specific gap: the work between a shipment being flagged and a shipment being resolved.
What happens inside a single exception task
The unit of work is a task, not a shipment. That distinction changes the math on almost everything below.
The task, and why it exists
Exception tasks surface directly on your AfterShip Tracking home page, so the queue lives where your team already starts its day rather than in a separate tool.
Shipments that share a root cause are grouped into one task. Forty packages stuck behind the same regional hub congestion become a single item to work, not forty investigations that reach the same conclusion forty times.
Each task opens with a summary: what the issue is, which shipments it affects, and why it was created. Grouping organizes the work. It does not resolve anything on its own.
The reasoning behind the recommendation
Before recommending anything, the AI agent assembles context across the order, the shipment's tracking history, carrier performance baselines, and how similar cases resolved in the past. That last input is what a rules engine structurally cannot reach, and it runs on AfterShip Intelligence, the data layer and domain models behind the platform.
The reasoning is shown, not hidden. The agent explains which signals drove the recommendation, so your team can evaluate the logic instead of accepting a verdict.
If the summary is not enough, you can expand a grouped task into its underlying shipment list to verify, or ask the agent follow-up questions inside the task itself.
The drafted action, and the approval step
Recommended next steps arrive with the communication already drafted, generated in the customer's language and ready to review or edit. Your team stops writing from a blank page and starts editing a first draft.
Nothing sends until a person approves it. There are no confidence thresholds and no auto-execution: the agent prepares, you decide.
Once approved, execution runs step by step with visible progress, and the result is written to an action log recording what happened, when, who was contacted, and who signed off. In the current release, the executable action is sending email: customer notifications, carrier escalations, and internal alerts.
Approval is the point where your team's judgment enters, and it is the only place it has to.
Seven exception scenarios the agent handles
Seven scenarios account for most of what lands in a working queue. These map to the exception codes each carrier already uses — FedEx's "RTS" and "delivery exception" scans, UPS's "exception" and "returned to shipper" statuses, USPS's "undeliverable as addressed," and DHL's "clearance delay" — so the agent recognizes the same event no matter which carrier's label it arrives under.
| Scenario | What triggers it | What the agent does |
|---|---|---|
| Return-to-sender recovery | Carrier returns a package after failed delivery, refusal, or a bad address. | Catches the RTS scan early and drafts a customer email offering options, plus an alert so the warehouse holds the package. |
| Lost in transit | Carrier declares the shipment lost, or it goes inactive long enough to be consistent with loss. | Confirms the signal against tracking and order context, then drafts a proactive customer email before the complaint arrives. |
| Incorrect address and failed delivery | Wrong address, or repeated failed attempts leaving the package at a carrier facility. | Acts inside the carrier hold window and drafts an address-correction request, or a carrier escalation for repeat failures. |
| Stalled shipments | No first scan, no movement for days, or transit time drifting past expectation, with nothing flagged by the carrier. | Uses your own detection rules to surface at-risk shipments, then recommends notification or escalation by severity. |
| Proactive delay communication | Weather, hub congestion, or regional disruption delays a cohort of shipments. | Drafts one batch notification setting revised expectations, then keeps monitoring. |
| Customs holds | Missing importer details, unpaid duties, or slow clearance. | Identifies what clearance is waiting on and who must act, then drafts the request or escalation. |
| Noise triage | Short delays and re-attempts that clear without anyone touching them. | Recommends monitor-only, and re-analyses if the shipment's state degrades. |
Since deploying AfterShip Agent for exception handling, Dr. Squatch has improved exception resolution time by more than 58% and reduced its most frequent "where is my order" tickets by 25%, while driving a 42% lift in positive sentiment on agent-initiated conversations, as the agent handles more customer interactions.
"As AI continues to reshape how brands operate, we see a real opportunity to apply that intelligence to the moments that happen after a purchase," said Joan Park, Senior Customer Experience Analyst at Dr. Squatch. "AfterShip Intelligence gives our team the context and automation to resolve delivery issues more efficiently, while keeping our focus on delivering a great customer experience. It's an important step toward making post-purchase a more intelligent part of the customer journey."
Getting started with exception handling
Exception handling is available on Premium and Enterprise plans, and it's one piece of AfterShip's broader approach to shipping exception management.
If you are already on one of them, it’s in your account now. Open your Tracking home page and look at the exception tasks already waiting there. The most useful first move is to pick your single highest-volume exception type and work only that one for a week or two, approving and editing as you go. You will learn quickly where the recommendations match your team's instincts and where your policy differs, and the drafts get closer to your voice as you correct them.
If exception handling is not on your plan, upgrading is the way to reach it, and the Tracking plans page shows what that involves. There is no trial on lower tiers. Before you weigh that, run the arithmetic from the top of this article against your own numbers: your monthly shipment volume, your realistic exception rate, and the per-case investigation time it implies. That is what you already pay in labor to do this by hand, and it is the number an upgrade has to beat.
AfterShip Tracking: Upgrade now →
Exception handling: frequently asked questions
How is this different from the shipment alerts and notification flows I already have?
Alerts surface broad operational signals, like an exception rate spiking past your threshold, and leave the individual cases to your team. Notification flows are trigger-based templates, so every shipment entering a status gets the same message regardless of severity or how likely the delay is to clear on its own. Exception handling works the individual case: why this shipment is stuck, whether it warrants contact at all, and what the response should say.
Will the agent contact my customers on its own?
No. Every recommended action waits for a person to approve it, and no confidence threshold lets actions through automatically. You can edit any draft before it sends, and each approval is recorded in the action log with the name of whoever signed off.
How is this different from a customer service chatbot?
A chatbot answers inbound questions across your whole support surface. This AI agent does one operational job, using logistics context a general assistant does not have: tracking events, carrier behavior over time, and how comparable cases resolved. It complements your support platform rather than replacing it, so ticket and order records keep updating through your existing workflow, and the sender address can match your existing support email so replies land in the inbox your team already watches.
How accurate are the recommendations?
Because a person reviews everything before it executes, accuracy shows up as how much editing your team does rather than as a wrong email reaching a customer. Acceptance rates, reviewer edits, and final case outcomes all feed back in, so recommendations calibrate to your policies instead of a generic default.
What happens to customer data?
Shipment and order data is handled under AfterShip's standard data handling and security practices. Shopper personal information is not used to train models.
Which plans include exception handling?
Premium and Enterprise. If you are on either one, exception tasks are already surfacing on your Tracking home page, built on the platform data behind 11 billion shipments tracked across more than 1,400+ carriers.
