Blog

How to Tell if a Freight AI Platform Does the Work or Just Suggests It

By Connor Miller, Founder and CEO, FreightScout. Last updated 19 October 2026.

Here's the test: after the AI touches a load, did something change that a rep would otherwise have had to do by hand? Is the load built in the TMS, the carrier booked and written back, the check call made and logged, the invoice out the door? If yes, it does the work. If what changed is a new alert, a score, a "recommended next step" or a draft sitting in a queue nobody owns, it suggests the work, and your rep still does it.

Both kinds of software can be useful. They're just not the same buy, and a lot of demos make them look like the same buy.

Why does this question matter more now?

Because every vendor at every freight show now says "AI," and from the back of the room most of it looks alike. A nice screen. A chat box. A list of loads with colored flags.

I spent twelve years in freight and scaled a brokerage to nine figures. On a real floor the expensive part was never knowing that something needed doing. Everybody knew. The tender was sitting in the inbox. The truck had gone quiet. The POD hadn't come in. The problem was that a person had to stop what they were doing, open three screens, and do it.

Software that tells that person, again, that the thing needs doing hasn't taken anything off their plate. It's added a tab.

What's the difference between AI that suggests and AI that does?

Suggesting AI reads your data and tells you something. Doing AI reads your data, does the job, and writes the result back where your team already looks.

Put it in floor terms:

The job	AI that suggests	AI that does
A tender hits the inbox	Flags it, maybe pulls a few fields into a preview	Parses the email, PDF or EDI and builds the load in your TMS: stops, refs, rates
A load needs a truck	Ranks carriers and shows you a list	Vets the carrier, negotiates up to your ceiling, books it, handles the rate con, writes it back
A truck goes quiet	Turns the load red on a dashboard	Dials the carrier, logs the status to the load, escalates to a person when something's off
A POD comes in	Notes that it arrived	Sends the invoice and chases the payment

The left column still needs a rep to pick up the phone and key the TMS. The right column is the rep getting their afternoon back.

What should you ask on the demo?

Don't ask what it can do. Ask what it did, on a real load, and where the proof lives. These are the questions I'd ask if I were sitting on your side of the table.

1. "Show me the TMS row after it ran." Not the vendor's screen. Your TMS. If the load, the booked carrier or the check call note isn't sitting on the load in McLeod, Aljex, Tai or whatever you run, somebody on your floor is still re-keying it.

2. "Who dialed the phone?" Plenty of tools call a ping a check call. A ping tells you where the truck is. It doesn't tell you the receiver pushed the appointment or the driver's about to run out of hours. Ask whether the software places the call itself and logs what the driver said.

3. "What happens when the carrier asks for more money?" A tool that suggests will show you a recommended rate. A tool that does will negotiate, and it should have a hard ceiling you set. FreightScout negotiates up to your buy ceiling, and going above it needs a human yes.

4. "What happens when it hits something it can't handle?" Every real floor has the bad ref, the split stop, the rate that doesn't match the tender. Software that does the work should stop on those and hand them to a person, not guess and not quietly skip them.

5. "Where's the log?" If it acted, there should be a record of what it did, what it wrote back and who said yes. If nobody can show you that, you can't trust it with live freight, and you shouldn't.

6. "Is this running on live freight today, or is it on the roadmap?" Ask which parts are live, which run with a person approving each action, and which are still slides. Then ask to see the live part on a real load.

Isn't "the AI drafts, a person approves" just suggesting with extra steps?

No, and this is where a lot of buyers get confused.

There's a big difference between a tool that says "you should probably book a truck on this load" and a tool that has already vetted the carrier, negotiated the rate and staged the booking, so the rep's whole job is one yes. The first one leaves the work with your rep. The second one did the work and is waiting for permission.

That permission is a setting, not a limit. At FreightScout, autonomy is a dial you control, per team and per action. On Supervised, every action waits for a person. On Assisted, the check calls run on their own. On Autopilot, the invoice, the outreach and the booking go too. Booking the carrier and shipper outreach run supervised by default, and you can switch either to autopilot whenever you want. Every action is logged and reversible, and you can turn any of it off without breaking the rest.

Most teams start with everything on approve. As the wins stack up, they flip the low-risk, high-volume work to autopilot and keep judgment where it belongs.

So the question isn't "does a person approve it?" It's "when the person approves it, is the work already done?"

How do you spot a dashboard dressed up as an agent?

Listen for these on the call:

"Insights," "visibility," "a smarter dashboard." Ask how the result lands in the TMS today. If the answer is that your rep acts on it, it's a dashboard.
"We flag the exceptions." Good. Who works them? If every flag still becomes a phone call your coordinator has to make, you bought a louder inbox.
"Our AI recommends the best carrier." Recommends to whom? Who calls, negotiates and books?
A demo that never leaves the vendor's screen. If they won't show the result in a TMS, ask why.
Numbers with no source. If the AI hands you a rate, a margin or a ranking, it should be able to show the load or the lane behind it. FreightScout cites the load, the lane and the margin behind every claim, and when it doesn't have a figure it says so instead of making one up.

None of that means the tool is bad. It means you should price it as reporting, not as hours back on the floor.

What about the TMS? Does doing the work mean replacing it?

It shouldn't. The TMS your team already logs into stays the system of record. Software that does the work sits on top of it and writes the result back, so nobody has to learn a new system to see what happened on a load.

How FreightScout connects to it: By API where the TMS offers one, and by reading it in place where it doesn't, set up during implementation.

If a vendor's answer to "will it write to my TMS?" is a migration plan, that's a different project than the one you were shopping for.

How do you check it after you sign?

The demo is one load. The first few weeks are hundreds. Here's what I'd watch:

Re-keying. Walk the floor and ask your reps what they're still typing. If loads, booked carriers or check call notes are getting typed twice, something isn't writing back.

Who's making the calls. If your coordinators are still dialing healthy trucks, the check calls aren't running.

What's waiting on a yes. On approve, the queue should be full of finished work, not reminders. If your reps are opening items and then doing the job themselves, it's suggesting, not doing.

The exceptions. The messy loads should land on a person fast, with the context attached. That's the work that should stay human.

What this is not

This isn't an argument for taking people off the floor. The point of software that does the work is that your best reps stop spending their day keying loads and dialing healthy trucks, and spend it on the customer, the carrier relationship and the load that's actually going sideways. Automate the friction, not the relationship.

FAQ

How do I know if a freight AI platform actually does the work?

Look at what changed after it ran. If the load is built in your TMS, the carrier is booked and written back, the check call is made and logged, or the invoice is out, it did the work. If the only change is a new alert, score or recommendation, it suggested the work and a rep still has to do it.

What's the difference between an AI agent and a freight dashboard?

A dashboard reports and alerts, and a person still dials, keys and books. An agent performs the brokerage task and changes the state of the work: it builds the load, books the carrier, makes the call, logs it and escalates the exceptions.

Is AI that needs a human approval really doing the work?

It can be. If the carrier is already vetted, the rate negotiated and the booking staged so the rep only has to say yes, the work is done and waiting for permission. At FreightScout that approval is a dial you set per team and per action, from approve-every-move to autopilot.

What should I ask a freight AI vendor on a demo?

Ask to see the TMS row after it ran, who placed the call, what happens when the carrier asks for more than your ceiling, what it does with a load it can't handle, where the action log lives, and which parts run on live freight today.

Does a freight AI platform have to replace my TMS to do the work?

No. FreightScout sits on top of the TMS you already run, whether that's McLeod, Aljex, Tai or something else, and the TMS stays the system of record. By API where the TMS offers one, and by reading it in place where it doesn't, set up during implementation.

What parts of FreightScout run on live freight today?

The sales work runs in production today. Load building, carrier booking and check calls run supervised on live freight at a working brokerage. Billing is live and supervised, including invoicing and collections.

Can I turn the AI off if it does something wrong?

Yes. Every agent action at FreightScout is logged and reversible, and you can turn any of it off instantly without breaking the rest.

Book a demo: watch FreightScout build, cover, track and bill a load on the TMS you already run, and see the row it writes back. Book a demo

All posts: the blog index.