After 30 days on a few lanes, should the AI logistics agent expand?
Key Takeaways
- Decide on expansion with numbers, not impressions. Compare the lanes the agent runs with how those same lanes ran before it started.
- 6 metrics carry the decision: manual touches per load, coordinator hours, time to book, missed pickups, time from exception to notification and POD collection time.
- There are 2 ways to expand, and they need different evidence. More lanes needs stable results on similar freight. More autonomy needs a record of approvals the team granted without changes.
- Wilson, by Cartage, is an AI logistics agent that starts with assisted execution on a few lanes, and the company expands lanes and the scope of execution step by step as the company gains confidence in the workflow.
- Holding is a valid result. If a metric got worse, fix the cause before adding scope.
Why is "is it working?" hard to answer after 30 days?
"Is it working?" is hard to answer after 30 days when nobody recorded how the same lanes ran before the agent started. Without a baseline, the team can only report that the inbox feels quieter. The fix is to rebuild the before numbers for those lanes and compare them metric by metric.
Thirty days is a common point for teams to review, not a fixed Cartage timeline. The decision affects whoever owns logistics headcount and service levels, usually a director of logistics or VP of operations at a manufacturer or distributor. It matters because coordinator capacity is hard to add: the BLS projects 18% growth in logistician jobs from 2025 to 2035, with about 26,600 openings a year.
One load, before and after
Take one outbound LTL load on a lane the agent now runs for a distributor of electrical components.
- Before, Monday 8:05 a.m. The order lands in Microsoft Dynamics 365. The coordinator emails three carriers for quotes, waits, books the cheapest, sends the rate confirmation (the document that locks the agreed rate and terms with the carrier), calls the dispatcher to confirm the pickup window and later chases the POD (proof of delivery, the signed receipt). For example, that's 11 touches across two days.
- After, Monday 8:05 a.m. The same order reaches the agent through the ERP feed. It requests quotes, books within the approval rules, confirms the pickup and requests the POD after delivery. The coordinator approves one booking above the cost threshold. For example, that's 2 touches.
The comparison only works if someone counted the 11. On most teams, nobody did. The order history and email threads still hold the evidence, so the baseline can be rebuilt.
Which metrics decide whether to expand?
6 metrics, measured on the same lanes before and after, decide most expansion calls. Each has a clear signal for expanding and a clear signal for holding.
| # | Metric | How to measure it | Signal to expand | Signal to hold |
|---|---|---|---|---|
| 1 | Manual touches per load | Emails, calls and edits a coordinator made per load | Clearly lower | Flat, or work moved to another person |
| 2 | Coordinator hours | Hours spent on these lanes per week | Lower | Flat or higher |
| 3 | Time to book | From shipment created to carrier confirmed | Same or faster | Slower, or bookings waiting on approvals |
| 4 | Missed pickups | Missed pickups per 100 loads | Same or fewer | More than before |
| 5 | Exception to notification | From a delay or missed pickup to the team and customer being told | Faster | Slower, or notifications missing |
| 6 | POD collection time | From delivery to POD filed | Faster | Slower, or PODs still chased by hand |
Metrics 1 and 2 show whether work actually left the team. If touches dropped but a different person now reviews everything, the work moved instead of disappearing.
Metric 3 often reveals approval friction. If bookings sit waiting for sign-off, time to book can get worse even when the agent is fast. That’s a signal to adjust rules, not to stop.
Metrics 4 to 6 protect service. A missed pickup is a pickup the carrier didn’t make in the agreed window. If those rise, expanding would spread the problem to more lanes.
How do you decide: more lanes, more autonomy, or hold?
Use the 6 metrics to pick one of 3 decisions, and change one thing at a time. Each decision needs different evidence and a different person to sign off.
| Decision | Evidence needed | What changes | Who approves |
|---|---|---|---|
| Hold | Any metric worse than before | Fix the cause: a carrier, a rule or missing order data | Logistics manager |
| Add lanes | Metrics improved; approvals are routine | Similar lanes with similar carriers and freight | Director of logistics |
| Add autonomy | Metrics improved; the team approves almost everything unchanged | A higher threshold, or one approval step removed | Whoever owns the approval rules |
Adding lanes works best when the new lanes share carriers and freight types with the current ones. A lane with new carriers or a new mode starts its own baseline.
Adding autonomy works best when the change history shows the team approving the same kind of action without edits, week after week. That record is the evidence; a feeling that “it’s always right” is not.
Common mistakes in the expansion decision
- Expanding without a baseline. Every result looks like a win, and the first bad week has no reference point.
- Changing lanes and rules at once. If a metric slips, you can’t tell which change caused it.
- Counting only the agent’s activity. Messages sent says nothing about touches the team still made.
- Removing approvals on judgment calls. A new carrier or a large rate change still needs a person.
See how to start an AI logistics agent on a few lanes.
How does Wilson, by Cartage, expand from a few lanes?
Wilson starts with assisted execution on a few lanes, and the company expands lanes and the scope of execution step by step, based on how those lanes run compared with before. Traditional logistics software gives teams information. Wilson executes the work required to move shipments forward.
| Step | Trigger | What Wilson does | When a person steps in |
|---|---|---|---|
| 1. Onboard | The company picks the first lanes | Gets its own email, system access and the playbook: carriers, rules, approval limits and contacts | Setting the rules |
| 2. Assisted execution | Loads arrive on those lanes | Runs quotes, bookings, tracking and follow-ups under the approval rules | Approving the actions the company chose |
| 3. Record | Every action | Records every message, confirmation and change in a change history | Reviewing the record |
| 4. Review | The company compares results | The lanes Wilson runs are compared with how they ran before | Deciding whether to expand |
| 5. Expand | Results support it | Takes on more lanes, or more actions proceed under the rules | Changing approval limits |
Controls
Approvals can sit on any action the company chooses: bookings or rebookings above a cost threshold, carrier selections or customer notifications. Actions under the rules proceed, and the rest wait for a person. Expanding autonomy means changing those rules, one at a time.
Wilson escalates when a carrier can’t confirm, when no approved carrier accepts a tender or when a decision exceeds the approval rules. It books only carriers the company has approved, and new carriers introduced by Cartage become available only after the company approves them.
Deployment time depends on scope. Standard freight with existing carriers is the fastest to set up, while ERP integrations and custom workflows are scoped up front. For how approvals work in detail, see when an AI logistics agent asks for approval.
What does expanding autonomy look like in practice?
Expanding autonomy means fewer actions wait for a person, one rule at a time, on lanes that have earned it. Each change should be measured before the next.
- Raise the cost threshold for bookings that need approval, for example from $1,500 to $2,500 on a lane where the team approved every booking unchanged.
- Let rebooking after a missed pickup run automatically on lanes with enough approved carriers.
- Let customer notifications go out without review once the team trusts the wording and timing.
- Keep approvals on decisions that still need judgment, like a new carrier or a large rate change.
What to ask a vendor to show in a demo
- “Show me the change history for one lane over a month, and how many approvals were granted unchanged.”
- “Change this approval threshold now, and show me which loads it affects.”
- “Show me the before and after numbers for manual touches per load on one lane.”
- “Add a lane with a new carrier and show me what the team has to approve.”
For the bigger picture, see what is an AI logistics agent.
FAQs
- Should an AI logistics agent expand after 30 days on a few lanes? Expand if the 6 metrics are the same or better than before on those lanes, and hold if any got worse. With Wilson, by Cartage, the company expands lanes and the scope of execution step by step.
- Is this a pilot? No. Cartage doesn't run pilots; most customers start Wilson on a few lanes and expand from there.
- How do teams measure Wilson's impact? By comparing the lanes Wilson runs to how those lanes ran before: manual touches per load, coordinator hours, time to book, missed pickups, time from exception to notification and POD collection time.
- Does Wilson act on its own from day one? No. Wilson starts with assisted execution. The company decides which actions need approval, and the scope of execution is expanded step by step.
- What can a company set approvals on? Any action it chooses, for example bookings or rebookings above a cost threshold, carrier selections or customer notifications.
- How long does it take to add lanes? It depends on scope. Standard freight with existing carriers is the fastest to set up; ERP integrations and custom workflows are scoped up front.
- How is Wilson priced? There's no subscription fee. Pricing is a percentage of the carrier payments that run through Wilson, and that one fee covers create, quote, tender, track, audit and pay. Pricing scales down as volume grows, and Enterprise pricing is custom.
Conclusion
After 30 days on a few lanes, the expansion call should come from six numbers on those lanes, each compared with how they ran before. Improve across the board and the next step is more lanes or more autonomy, one change at a time. Slip on any one and the next step is fixing the cause.
Wilson, by Cartage, starts with assisted execution, records every action in a change history and expands one rule at a time, so the evidence for each step is already on the shipments. Bring one lane's numbers from before and after, and walk through the expansion decision with Wilson's change history in a demo at cartage.ai.
Sources
- U.S. Bureau of Labor Statistics, Occupational Outlook Handbook: Logisticians, 2025–2035 projections: https://www.bls.gov/ooh/business-and-financial/logisticians.htm
- Cartage, Wilson product FAQ, October 2026: https://cartage.ai