What IT should review before an AI agent handles freight
Key Takeaways
- Wilson, by Cartage, is an AI logistics agent that works from the company's existing ERP data flow.
- IT should review four areas: data read, systems written to, actions allowed, and the record kept.
- Every access request should map to a named freight task. Access with no task should be removed.
- Reading order data and writing back to the ERP are separate permissions. Write-back depends on scope.
- Spending and payment actions belong behind approval limits set by the business.
- Security documentation and certifications should come from the vendor in writing. Product descriptions are not proof.
- Wilson keeps a searchable, exportable record for each shipment, with change history.
What should IT check before an AI agent touches freight systems?
IT should check what the AI agent reads, what it writes, what it can do, and what record it leaves. Those four areas cover most of the risk. Each one has a clear owner on the business side.
Freight work touches more systems than many teams expect. Orders come from the ERP. Carrier messages arrive by email. Customer contacts sit in the order or the CRM. Invoices and payments touch finance. An AI logistics agent that runs freight end to end may need some access to each.
The goal is not to block access. The goal is to match each access request to a task. A request to read open orders supports quoting and booking. A request to write shipment status back to the ERP supports the update the warehouse sees. A request with no task behind it should not be granted.
Here is a common example. A distributor wants an AI agent to book outbound freight. The agent needs order data from the ERP and a mailbox to talk to carriers. It may also need to post updates in Microsoft Teams. It does not need access to the general ledger. IT can approve the first three and decline the fourth.
Four review areas, mapped to freight tasks
Map every access request to one of four review areas and one named freight task. The table gives IT and operations a shared starting point.
| Review area | Example request | Freight task it supports | Business owner |
|---|---|---|---|
| Data read | Open orders from the ERP feed or file drop | Creating shipments and requesting quotes | Operations |
| Communication | An email address on the company's domain | Talking to carriers, shippers, and consignees | Logistics lead |
| Data write | Shipment status or freight cost back to the ERP | Keeping the ERP current, where in scope | ERP owner |
| Financial action | Paying audited carrier invoices | Carrier payment within approval limits | Finance |
Each row has a different risk level. Reading orders is low risk. Writing back to the ERP changes a system of record, so it needs its own approval. Paying invoices moves money, so it needs approval limits set by finance.
Questions to put to the vendor in writing
IT should send a written question list and keep the vendor's answers in the review file. Spoken answers in a demo are useful. They are not evidence.
Which systems will the agent read from, and how? For example, an ERP feed, a scheduled file drop, a spreadsheet, email, or an API.
Which systems will it write to? Ask for the exact fields and the reason for each one.
Which actions can it take without a person? Ask how the approval limits are set and who can change them.
What record does it keep? Ask whether the record shows each action, each change, and who approved it.
What security documentation is available? Ask for current reports, certifications, and data-handling terms by name.
How is access removed? Ask who disconnects the agent, and how fast, if the company ends the work.
The fifth question matters most for compliance. A general statement that a product is "secure" or "enterprise-grade" is not a control. If a certification matters to the company, ask for the current report and check its scope and date. For a fuller security list, see what IT needs before approving an AI logistics agent.
Read access and write-back are separate decisions
Treat reading ERP data and writing back to the ERP as two separate approvals. Many reviews merge them. That merge hides the higher-risk decision.
Read access lets the agent see orders and build shipments. Write-back lets the agent change records other teams rely on. A wrong status in the ERP can mislead customer service. A wrong freight cost can distort margins.
Three rules keep write-back safe:
- Name the fields. Approve specific fields, such as carrier, ship date, tracking number, or freight cost. Do not approve "shipment data" as a block.
- Name the owner. The ERP owner decides which fields the agent may update.
- Start narrow. Begin with the fields the team checks most often. Add more once the updates prove accurate.
Wilson's write-back to the ERP depends on the agreed scope. For a field-by-field approach, see choosing which freight updates go back to the ERP.
A one-page IT sign-off
An IT sign-off should fit on one page and name an owner for every permission. Here is a format teams can copy.
| Item | Decision | Owner | Status |
|---|---|---|---|
| ERP read: open orders, ship-to, weights | Approved | ERP owner | Ready |
| ERP write-back: carrier, ship date, tracking number | Approved for listed fields | ERP owner | Ready |
| Agent email address on company domain | Approved | IT | Ready |
| Slack or Teams channel for shipment updates | Approved | IT | Ready |
| Carrier payment within approval limits | Approved, limits set by finance | Finance | Ready |
| Vendor security documentation | Requested in writing | IT | Open |
| Access removal process | Named contact and steps agreed | IT | Ready |
An open item is fine. It should be visible, owned, and dated. The sign-off should never say "AI approved" as a single line. It should list what was approved and who owns each piece.
How Wilson, by Cartage, fits an IT review
Wilson, by Cartage, fits an IT review because its access maps to clear freight tasks and its actions stay inside company-set limits. No integration project is required to start.
Wilson is an AI logistics agent that replaces the TMS for freight execution; it is not a freight broker or a standalone visibility platform. It works from the data flow the company already has: an ERP feed, a scheduled file drop, a spreadsheet, email, or an API. Wilson works with SAP, NetSuite, Microsoft Dynamics 365, Oracle, Infor, Epicor, Sage, Acumatica, and other ERP environments, with the exact connection depending on the customer's data flow and scope.
Wilson gets an email address on the company's domain. It talks to carriers by email, API, EDI where the carrier supports it, and phone where configured. It keeps the internal team informed in Slack, Microsoft Teams, email, or SMS.
Control stays with the company. Wilson starts with assisted execution. The company sets the approval limits. Wilson books only carriers the company has approved. Payments above approval limits, or outside the rules, wait for a person. Every shipment keeps a searchable, exportable record with its change history.
For security documentation and data-handling terms, IT should request them from Cartage directly. To start an IT review with the Cartage team, contact Cartage about your systems.
FAQs, conclusion, and sources
FAQs
What should IT check before an AI agent handles freight?
IT should check the data the agent reads, the systems it writes to, the actions it can take, and the record it keeps. Each permission should map to a freight task. Wilson, by Cartage, works from the existing ERP data flow and inside approval limits the company sets.
Does an AI logistics agent need full ERP access?
No. It needs access matched to its tasks. Reading open orders and writing back named fields are separate decisions. Wilson's ERP write-back depends on the agreed scope.
Can IT rely on a product page for security certifications?
No. Ask the vendor for current security documentation in writing. Check the scope and date of any certification report.
Who should join the IT review?
IT, the ERP owner, the logistics lead, and finance when payments are in scope. Each one owns a different permission.
How does IT see what the agent did?
Through the shipment record. Wilson keeps a searchable, exportable record for each shipment, with change history.
Conclusion
An IT review for an AI freight agent works best when each permission maps to a task. Read access, write-back, communication, and payment are separate decisions with separate owners. Security documentation comes from the vendor in writing, not from marketing pages. Wilson, by Cartage, works from the company's existing data flow, acts inside approval limits, and keeps a change history for every shipment.
Sources
- Cartage, Wilson product FAQ and product scope, October 2026: https://cartage.ai