Not the vendor pitch. A load-by-load account of what these tools do on a Canadian dispatch board, and the four places a dispatcher still decides.
Key takeaways
A dispatcher’s day is not one job. It is roughly six, interleaved: reading inbound freight offers, matching them to trucks and clocks, confirming appointments, chasing check calls, handling the exception that blows up at 14:20, and writing everything down so billing and safety can use it later. Software that helps with all six is rare. Software that helps with three or four is common, and that is where the return is.
Treat “AI dispatch” as a set of separate capabilities rather than a product category. Each one either takes a keystroke off your desk or it does not.
Freight offers arrive as email bodies, PDF rate confirmations, spreadsheet tenders and portal screenshots. The first genuinely useful thing a model does is extract the fields — origin, destination, pickup and delivery windows, weight, commodity, equipment, reference numbers, accessorial terms — and drop them into your TMS as a draft order rather than a re-keyed one.
This matters more than it sounds. Re-keying is where the wrong appointment date and the wrong trailer type enter your system, and both cost a truck. Extraction is also the lowest-risk place to start because a person is still going to look at the order before it is dispatched.
Matching is where AI is oversold. A model can rank candidate trucks by proximity, empty time, trailer type, customer preference and historical on-time performance. What it cannot do is invent hours.
The Commercial Vehicle Drivers Hours of Service Regulations cap a driver at 13 hours of driving time and 14 hours of on-duty time in a day (s.12), require at least 8 consecutive hours off-duty before driving again (s.13(1)), and impose a 16-hour elapsed-time window between those off-duty periods (s.13(3)). Cycle limits sit on top: 70 on-duty hours in 7 days under cycle 1 (s.26), or 120 in 14 days under cycle 2 (s.27), resettable only by 36 or 72 consecutive hours off (s.28). A suggestion that ignores any of these is not a suggestion, it is a violation waiting for a scale.
The tools worth buying read live duty status from the certified electronic logging device and filter the candidate list before a human sees it — and only a device on Transport Canada’s certification registry counts, since the underlying CCMTA ELD Technical Standard was updated to version 1.3 on September 29, 2025, replacing the version 1.2 that had stood since October 27, 2020. That is the actual value: a shorter list, not a decision.
Most customer complaints in trucking are not about late freight. They are about late information. Position data already exists in your telematics platform — Samsara, Geotab and Motive all expose it — and the gap is that nobody has time to turn it into a message.
Automated ETA updates, exception alerts when a truck stops moving inside a delivery window, and drafted customer emails that a dispatcher approves are the highest-volume, lowest-stakes wins on the desk. Note that monitoring drivers electronically carries its own obligations in some provinces — in Ontario, s.41.1.1(1) of the Employment Standards Act, 2000 requires any employer with 25 or more employees on January 1 of a year to have a written electronic monitoring policy in place by March 1 of that year, and s.41.1.1(3) requires a copy to reach every employee within 30 days, which Treadstone Law sets out here.
Section 87(1) of the hours-of-service regulations requires a motor carrier to monitor the compliance of each driver, and s.87(2) requires immediate remedial action plus a record of the dates of non-compliance and the action taken. Most small fleets do this in someone’s memory.
A model that reads the day’s duty-status records, flags the exceptions and drafts the remedial-action note for a safety person to approve turns an obligation you are technically failing into a file you can show an auditor. The same extraction feeds billing: Trimble TruckMate and McLeod LoadMaster both integrate order, dispatch and settlement in one record, so a clean order is a clean invoice. None of it works on a record that never arrives: s.85(1) of the regulations gives a driver 20 days after completing a record of duty status to forward it, with its supporting documents, to the home terminal, and puts the burden on the carrier, not just the driver, to make sure that happens — a dispatcher who never checks whether the paperwork actually came back is the one left explaining the gap to an inspector.
Worked example: one reload, start to finish
A rate confirmation arrives at 16:40 for a Brampton to Dorval pickup tomorrow 08:00, deliver 14:00. Extraction reads the PDF and creates a draft order with both windows and the reference number. Matching filters your available tractors: two are within range, but one driver started at 05:10 today, so his 16-hour window (s.13(3)) closes tonight and an 08:00 pickup 400 km away does not fit without an 8-hour break he has not started. The list returns one truck.
The dispatcher, not the model, makes the actual call — that driver has run Dorval twice and knows the gate. The system drafts the dispatch message and the customer confirmation; the dispatcher edits one line about the gate code and sends. Total desk time: under two minutes, against roughly fifteen. Nothing about the decision changed. The typing did.
It does not decide which customer gets the last truck on a Friday. It does not judge whether a driver who has already had a bad week should be sent into Montreal in freezing rain. It does not negotiate. And it does not replace the phone call to the broker who owes you a favour, which is still how a lot of Canadian freight moves.
Anything a vendor describes as autonomous dispatch should be read carefully. The regulatory record — the duty-status record, the remedial action note, the supporting documents — is signed by people, and a carrier that cannot say who decided has a problem no software fixes. Section 86(2) separately bans entering inaccurate information or altering, deleting or destroying a record of duty status, and s.86(3) bans disabling, deactivating, disengaging, jamming, or otherwise blocking or degrading a signal, or reprogramming an ELD so that it stops recording accurately — rules aimed squarely at any tool that promises to make the compliance clock disappear rather than simply read it back faster.
Extraction, ETA drafting and exception alerting do, because they are per-load savings and do not need scale. Optimisation across a network needs enough loads for a network to exist; below roughly 30 trucks a good dispatcher usually beats it.
The same matching logic applies, but be careful how you direct them. How much control you exercise over scheduling is one of the factors that separates a contractor from an employee, and the consequences of getting it wrong are set out here.
A 30-minute call is enough to tell you whether AI pays for itself on your dispatch desk.
