Do you have to replace your TMS to use AI?

Nobody says it in the meeting. It ends the meeting anyway.

The reasoning runs underneath, and it is not stupid. AI is software. New software is a system. A new system means moving the data. Moving the data means the last migration, which took most of a year and which the accounts team still brings up. So: not now. Revisit after the year end.

Your operations head is not being obstructive when she thinks this. She is remembering the weekend the job numbers restarted at one, the re-raised invoices, the three months when two systems ran side by side and neither was right. That memory is sound. It is being applied to the wrong purchase.

The Playground outline calls this the biggest silent reason forwarders never start, and silent is the operative word. It never becomes an objection you can answer, because it never gets said. It arrives disguised as a diary problem.

It also goes unexamined because buyers ask what the software does and rarely what it needs. So the vendor answers that for an hour, everyone nods, and the thing that would have decided it never comes up. Which is why B2 is worth five minutes. The assumption is usually wrong, and cheap to test.

Draw one enquiry end to end and the answer stops being a matter of opinion.

An enquiry lands in a shared mailbox at twenty to ten. Somebody reads it. The gross weight is missing, so somebody emails back. Two agents get asked for a rate on the leg you do not hold. Somebody looks up what that customer paid on the lane in March. A quote gets typed, the margin gets checked, it goes out. Two days later somebody chases it.

Now count how much of that touched your system of record. In most offices, none of it. Your system gets involved at the end, when the quote is won: the job is created, the file takes a number, the invoice is raised, the books close.

That is the real split, and it has nothing to do with AI. Your system of record holds what has to be right forever and gets looked at by an auditor. The hours before it are spent in an inbox, a spreadsheet, a WhatsApp thread and somebody's head.

Software of this kind is bought for those hours. That is the whole reason the answer to B2 is usually no. There is nothing to migrate, because the work it does was never in the system to begin with.

Two facts from outside the AI conversation point the same way. IATA describes ONE Record as “a standard for data sharing”, and its published principle is that “data stays at the source and the owner determines who may or may not access their data”. DCSA describes its container shipping standards as vendor neutral frameworks that promote interoperability. Neither is about your TMS. Both show which way the industry is going: data staying where it is and being shared. One system holding everything is the older idea.

It names the door, the direction and the moment your own system takes over, and it survives one follow-up.

The sheet gives the good answer as: no, and exactly how it sits alongside. Enquiry in, quote out. The second half does all the work. Anyone can say no.

Everything before the star is the hours you are buying back. Everything after it is shaded because it is yours and stays yours. If a vendor's version of that picture leaves your system out, they have answered B2 without meaning to.

The shape matters more than the supplier. Here is ours, in the wording we publish elsewhere on this site.

• Keep your TMS for the records. FreighAI runs the work around them.
• FreighAI is an AI-first freight operations platform that can run standalone or work alongside an existing TMS or ERP. It does not require a forced replacement.

Hold every supplier to a sentence of that kind, us included. A vendor who will not put their version in the proposal has not answered B2, whatever they said in the room.

Every vendor says yes to B2. These are the four ways yes can still cost you a year.

Trap one. Alongside that means a file at midnight

“We integrate with your TMS” can mean a nightly export somebody checks each morning and re-runs when it fails. That is not integration. It is a chore with a schedule. Ask what a person has to do, every day, for this to keep working. The honest answers run from nothing to twenty minutes a morning, and that gap matters more than the feature list.

Trap two. The write nobody scoped

Reading out of your system is easy to agree to. Writing into it is where your IT contact, your licence terms and eventually your auditor turn up. Ask which you are buying. If it is both, ask who owns each field when both sides can change it. That is the question that starts the argument in month four.

Trap three. The staged replacement

“We will start with quoting” is a good plan. “And then it makes sense to move the jobs across too” is a replacement on a slower clock. Nothing is wrong with wanting that. A lot is wrong with finding out in month seven. Ask it plainly. In a year, what is still in my system? Put the answer in the proposal.

Trap four. “It depends on your setup”

The sheet names this as the warning sign, and it is, but only in its bare form. “It depends” is a fine way to begin. A real answer says what it depends on: which mail system you run, whether your rate cards sit in files or in the system, whether the won job has to be created without a person typing it. A vendor who has done this lists the dependencies in a minute. One who has not keeps it general, because the vagueness is working for them.

Sometimes it is. An article that said never would be a sales argument, and this sheet is not one.

Three situations where replacement is the honest answer:

• You cannot get your own data out. If nobody can export the job history and the rate cards without raising a support ticket and waiting, no layer on top fixes that.
• Support has ended, or the person who maintained it has left. A system nobody can change is not a system of record. It is an archive with a login.
• The record is not in one place. Job numbers in one system, invoices in a second, accounts in a spreadsheet one person owns. Adding a layer over three sources of truth gives you four.

If you are in one of those, replacing is a real project and deserves its own clock, budget and sponsor. What it should not be is a decision made because an AI vendor happened to be in the room that month.

Leaving a supplier is a defined event now, not just a feeling. The EU Data Act, in force since 11 January 2024 and applicable since 12 September 2025, sets out a framework for customers to switch effectively between providers of data processing services. It is written for cloud services rather than freight software, and your contract still decides most of it. The reminder holds anyway: leaving has a defined cost, and walking into it by accident is the expensive way. This article explains the choice. It is not legal advice.

Write it the same day, one line per vendor, in their words. A fortnight later you will remember who was likeable and nothing else.

The last row is the dodge question from the sheet, and the one to keep whatever else you drop. It turns a comfortable answer into a list somebody can check.

Then put the same question to your own office. The changes are usually small, and the same whoever you buy from. One mailbox becomes properly shared. The rate cards go in one place instead of four. One person is named as the approver. That list is the real answer to B2. What you are agreeing to is not a migration. It is three decisions and a mailbox.

The next thing to ask is when. That is question B3, and it has an article of its own, because an honest answer has a date in it and a list of what would move it.