There is a popular argument going around right now that companies should stop chasing immediate ROI from AI and instead treat it as a long experiment. Build the muscle first and let competence come before results.

That is a reasonable position for a company still finding its footing with agentic AI. It is not the position a shipper needs to hear from the partner moving their packages across borders today. When a package gets stuck at customs, or a support email sits unanswered for two days, that is a lost customer. We call it Shadow Loss.

Over the past several years, we have invested in AI automation in international shipping built to hold up under the volatility of today's climate. Some of it is AI. Some of it is automation and infrastructure we have built over years, running on its own logic rather than a model. We are not going to blur that line to make a bigger AI claim. What all of it shares is this: it is running in production today.

What we found when we looked at our own support queue

In early 2025, we sat down with our customer service data instead of guessing at it. Over a three-month span, we found agents were handling upward of 150 email inquiries a day, part of a broader pool of more than 14,000 messages, roughly 83% of which were shipping related, mostly tracking updates and missing package questions. Nearly all of it ran through a manual process. Our CSRs were spending most of their time triaging repetitive requests instead of the complex cases that actually needed human judgment, and it was not a model that could scale with volume.

Working with our AI partner Tenjumps, we built and deployed ePost Assist, an AI-backed chatbot trained specifically on our own inquiry patterns, not a generic support bot bolted onto a help desk. Within two months it became the first line of response for our customers. Generic bots often make up an answer when they get stuck. ePost Assist is strictly anchored to our live database. If the facts are not there, it does not guess, it loops in a human.

The result: it now instantly resolves more than 70% of our daily tickets, mostly tracking and status requests, and prompts customers for missing details upfront so cases that do need a human get resolved faster too. Our team went from drowning in routine tracking questions to focusing on the shipments that actually need a person's judgment. And because it never clocks out, our customers in 200+ countries and territories get a real answer at 2am local time, not a queue position.

Sort to light: cutting the manual step out of the physical process

The support queue is one side of the operation. The warehouse floor is the other, and it used to run on the same manual pattern: scan a parcel, then sort it by hand into a bag.

We replaced that with a system our team uses with ease. Now the process is simple: scan, then follow the light. Operators no longer need to memorize where a parcel goes, the system tells them, off a database connection updating in real time. That matters more than it sounds like it should: a gaylord coming off the line loaded with mixed packages can go straight to sort to light with no separate fine hand-sort step first, and the sortation scheme itself can be reconfigured any time we need to change how parcels get grouped.

The same real-time data that drives the sortation also drives operational decisions most providers are making on gut feel: which bins are filling up, how much container and airline space is left, and how to get the maximum weight into every sack before it leaves. We hold every bag to a 24-hour departure window to its destination country. During peak season, when we lean heavily on temporary staff, the same light-directed process means a first-day temp sorts as accurately as someone who has done it for a year, because the system is doing the routing decision, not their memory of a sortation chart.

We are one of the largest international parcel consolidators in the United States, and in September we are expanding into a larger Chicagoland facility, growing from 45,000 to 104,000 square feet, to house our increased investment in parcel and mail sorting automation. The expansion adds a dockside TSA cage, a raised ops platform, a 10,000 square foot mail area, and upgraded office space including a large boardroom and IT think room. The result of minimizing manual handling errors and maximizing bag and container space is not only internal efficiency, it is cost savings we pass along to customers.

Sort to light still puts a person in the loop to physically place each parcel. We have also invested in a belted automated sortation system that removes that step entirely: parcels move through on a belt and get sorted with no human intervention at all. Between the two systems, we have automation matched to the job, light-guided sortation where a person's hands are still the right tool, and fully automated sortation where they are not.

Automation that keeps a shipment moving, not just a customer informed

In addition to customer service and physical processing, we also focus on what happens when a shipment itself hits a wall mid-route.

Every international shipper eventually loses a carrier mid-stream: a strike, a capacity crunch, a service suspension. What separates a resilient network from a fragile one is what happens in the minutes after.

Our multi-carrier network is built to reroute automatically the moment a carrier fails, rather than waiting for someone to notice and manually rebook. When Canada Post went on strike, our system rerouted 47,000 shipments within 48 hours with zero SLA failures. This is infrastructure making that call automatically, shipment by shipment, without waiting on a human to notice.

What we're building next

None of this is where we stop. The difference between us and a provider still deciding whether to invest is that our next moves build on infrastructure that already works, not a blank slate.

This year, we are:

• Extending the same automation model into address validation, catching a bad or incomplete address before a shipment leaves rather than after it fails somewhere in a foreign postal system.
• Applying agentic AI to HS code classification, automating a step in customs declarations that is still manual and error-prone at most providers, and one that directly determines how fast a shipment clears customs.
• Building automated review of product descriptions at the point of shipment, flagging language vague or generic enough to trigger a customs hold and surfacing clearer, more compliant alternatives for the shipper to approve before it ships.

Financial integrity: catching billing anomalies before they hit the ledger

The operational side of shipping moves fast, but the financial tail can drag on for months. In international logistics, matching what you expected a shipment to cost against what a partner or vendor actually invoiced is a notoriously manual, retroactive headache. Most companies pay the invoice first and spend months afterward auditing line items to chase down overcharges and refunds, if they catch the discrepancies at all.

We built automation to close that gap. Before we pay a vendor invoice, our system checks it against our own internally calculated, real-time shipment costs and flags anything that does not match, using historical data to tell a reasoned rate change from an anomaly. This protects our margins and catches billing errors before they are locked in, rather than months later during a quarterly review.

That control shows up for you in a simpler way: we bill upfront based on your agreed rates. We do not add ancillary charges or recalculated duties and taxes after the fact, because we have already caught and resolved the discrepancy on our end before your invoice is even cut.

The same real-time visibility into vendor costs also means we spot carrier rate spikes the moment they happen, so we can reroute future shipments before a rate change ever reaches your invoice.

Why this matters more than it sounds like it should

International shipping fails quietly. A customer whose package gets stuck in customs, or who never gets a real answer about where their order is, usually does not complain, they just do not order again. That is Shadow Loss: the customers you lose without ever knowing you lost them.

Automation like this is the difference between catching a failure in minutes, whether that is a stuck package or a downed carrier, and finding out about it when a customer churns. Systems like these are part of how ePost protects the second sale.

The real question for shippers evaluating a partner right now

Any provider can say they are working on AI. The more useful question when evaluating an international shipping partner is simpler: is their automation running in production today, resolving real tickets and rerouting real shipments, or is it still in a lab?

At ePost, 70% of tickets are resolved before a human ever sees them, a sort floor where light and belts replaced guesswork, and 47,000 shipments kept moving when a major carrier stopped. Across the network, that adds up to more than 20 million shipments delivered on time and to their correct destination last year.

Frequently asked questions

How much of ePost's customer service is automated? ePost Assist, our AI-backed support chatbot, instantly resolves more than 70% of daily support tickets, mostly tracking and status requests, and loops in a human for anything it cannot verify against live shipment data.

What happens when a shipping carrier fails, like during a strike? ePost's multi-carrier network reroutes shipments automatically the moment a carrier fails. When Canada Post went on strike, the system rerouted 47,000 shipments within 48 hours with zero SLA failures.

What is Shadow Loss in international shipping? Shadow Loss is the customer churn that follows a shipping failure a customer never reports. A package stuck in customs or an unanswered tracking question often does not generate a complaint, it just costs a repeat customer.

Is ePost's automation AI, or traditional automation software? Both, applied to different problems. Customer service, HS code classification, and billing anomaly detection use AI. Sort to light, belted sortation, and multi-carrier rerouting are automation and infrastructure running on fixed logic rather than a model.

How does ePost verify vendor shipping invoices? Before paying a vendor invoice, ePost's system checks it against internally calculated, real-time shipment costs and flags anomalies using historical data, catching billing errors before payment rather than during a later audit.

How many shipments does ePost handle successfully? ePost delivered more than 20 million shipments on time and to their correct destination last year, across a network spanning 200+ countries and territories.