Pilot the Follow-Up, Not the Chatbot

Key Takeaways

A first freight automation test should prove that one follow-up loop closes, not that an AI can hold a pleasant conversation.
The strongest loop to test is the unanswered carrier message: the second ask, the channel change, the escalation, and the confirmed response written to the shipment.
Define three things before the test starts: the trigger, the follow-up ladder, and the pass condition. Without them, any result looks acceptable.
Force the failure on purpose. A loop where the carrier answers the first email proves very little.
Wilson, by Cartage, follows up across configured channels, escalates under the company's rules, and records each attempt on the shipment. Most teams start it on a few lanes and expand from there.

What should a buyer's first test of freight automation prove?

It should prove that one real follow-up loop closes without a coordinator carrying it: the carrier doesn't answer, the system asks again, escalates when the rule says so, and ends with a confirmed response recorded on the shipment. Whatever the team calls its first test, scoping it to one loop is what makes the result readable.

The temptation is to judge AI for freight by how it sounds. Does it write a polite email? Does it understand a messy reply? Those questions matter, but they are the easy part. Plenty of tools can draft a good message. Far fewer can remember that the message went unanswered, decide when to ask again, switch to a phone call, and tell the right person when the carrier still hasn't confirmed.

That second set of behaviors is where coordinator time goes. A single pickup confirmation that needs three touches and a phone call is small; the same pattern repeated across every load each day is a large share of the week, as the worksheet in what carrier follow-up costs a manufacturer shows. A first test built around that loop measures the work the team actually wants to hand off.

Cartage doesn't offer a formal trial program. Most teams start Wilson on a few lanes and expand from there. The advice here is about how the buyer frames its own first loop, whatever the rollout looks like.

How to define the loop before the test starts

Write the loop down as a short spec with five parts. If any part is missing, the team will argue about the result afterward instead of reading it.

The trigger. The exact event that starts the loop. Example: a load is booked with an approved carrier and no pickup confirmation has arrived by a set time on the day before pickup.
The first ask. What goes to the carrier, on which channel, and what answer counts as a response. "We'll get back to you" is not a pickup time.
The follow-up ladder. What happens after each silence: a second email to the same contact, then a call to dispatch, then a message to the carrier's backup contact. The spacing between steps comes from the company's SOP, not from the software's default.
The escalation rule. The point at which a person takes over, and who that person is. Example: after the third unanswered attempt, or when the pickup is less than a set number of hours away.
The pass condition. A sentence anyone can check against a shipment. Example: "The pickup time is confirmed by the carrier and recorded on the shipment, or the load is escalated to the named owner under the rule, with every attempt attached."

The pass condition deliberately allows two good endings. A loop that escalates correctly has done its job, even if the carrier never answered. A loop that stalls silently has failed, even if the carrier eventually replied to someone's personal inbox.

What to watch while the loop runs

Watch the gaps between steps, because that is where manual follow-up usually breaks. Four points deserve attention on every load in the test.

Memory. When the carrier replies on the second attempt, does the reply land on the right shipment, or does someone have to match it by hand? Replies that quote the wrong reference number or come from a different person at the carrier are the real test.

Persistence without spam. The ladder should advance on schedule, and stop the moment a valid answer arrives. A system that keeps asking after the carrier confirmed creates its own problem with the carrier.

Channel change. If the ladder includes a phone call, check that the call happened, that the answer was captured, and that the shipment reflects it. A note that says "called" is not the same as a confirmed time.

The escalation package. When the loop hands off, look at what the person receives. A useful handoff names the load, the carrier, each attempt with its time and channel, any partial answers, and the decision that is now needed. If the owner has to open three inboxes to understand the situation, the escalation shifted work rather than removing it. For the question of who should own that handoff, see who escalates unanswered carrier messages.

Count manual touches on the loop before the test and during it: every email, call and status update a person made to get a confirmation. That single number usually tells the story more clearly than any survey of the team.

When the loop fails, find out which part failed

A failed loop is useful only if the team can name the cause. Most failures fall into one of five groups, and each points to a different fix:

Data. The load arrived without a carrier contact, a pickup number or a correct date. Fix the order feed or the intake rule before blaming the follow-up.
Carrier behavior. Some carriers only answer the phone, or only answer one dispatcher's email address. Add that contact or channel to the ladder.
Unclear rule. The SOP said "follow up as needed." The software can't act on that, and neither can a new coordinator. Write the spacing and the escalation point as numbers.
Approval boundary. The carrier answered, but the answer required a decision (a later pickup, an extra charge) that nobody had authorized. That is a correct escalation, not a failure, unless the boundary was set too tightly for routine cases.
Connection to the order data. The confirmation was captured but never reached the record the team uses. Check where the result is written and who can see it.

Decide in advance what result would justify extending to a second loop, such as pickup confirmations on more lanes or a stale-ETA loop, and what result would mean fixing the setup first. Expanding a loop that only works on happy paths spreads the same gaps across more loads.

A one-loop test card for unanswered carrier messages

Fill in the left columns before the test and the right column with what actually happened. The values in the middle column are illustrative examples, not recommended settings.

Stage	Example setting	What a pass looks like
Trigger	No pickup confirmation by noon the day before	Loop starts without a person noticing first
First ask	Email to the carrier's dispatch address	Message references the right load and asks for a pickup time
Second ask	Email again after a set interval	Sent on schedule; stops if the carrier already replied
Channel change	Call to dispatch	Call placed, answer captured on the shipment
Escalation	Named coordinator after the third silence	Handoff includes every attempt and the decision needed
Confirmed response	Carrier gives a firm pickup time	Time recorded on the shipment and the right people told
Manual touches	Counted before and during	Fewer touches per confirmed pickup

Run the card on loads where the carrier answers quickly and on loads where you ask a friendly carrier not to answer. Both columns of results belong in the same review.

How Wilson, by Cartage, runs the follow-up loop

Traditional logistics software gives teams information. Wilson executes the work required to move shipments forward, and the unanswered carrier message is a good place to see that.

Wilson follows up with carriers by email, by API or EDI where the carrier supports it, and by phone where calls are configured, at the frequency set in the company's SOP. Each message, confirmation and change is recorded on the shipment, and call transcripts are saved on the order. When the carrier confirms, Wilson updates the shipment and tells the team through Slack or Microsoft Teams, and the customer by email or SMS where configured. When the carrier can't confirm, or the answer needs a decision outside the approval limits, Wilson escalates to the team with the history attached. If a carrier can't cover the load, Wilson initiates rebooking from the company's approved carriers, automatically within the rules or after approval.

Wilson starts with assisted execution. The company sets which steps need approval, and the scope of execution can expand step by step as the team gains confidence in the workflow. Timing depends on scope: most teams start on a few lanes and expand from there. For the broader picture of what an agent does beyond follow-up, see what an AI logistics agent is.

Bring one follow-up loop your team chased by hand last week, and walk through how Wilson would run it, from the first unanswered message to the confirmed response.