AI Dispatch Software: The Buyer Evaluation Checklist

Most AI dispatch demos look identical. A clean dashboard, a confident voice, a load that books itself in fifteen seconds. The problem is that the demo lane is hand-picked and the answers you actually need never come up. You find them three weeks into a trial, when a broker the system trusted turns out to be double-brokering and nobody can tell you why the load got ranked first.

This is a checklist of questions to ask before you sign. They are organized by category and written to be answerable. Vague products give vague answers. If a vendor cannot tell you exactly which boards they pull from, where the human approves a booking, or what happens to your broker emails, that is the answer. Read each section, then ask the questions out loud on the call and watch how fast the reply comes. The good ones land in a sentence. The bad ones turn into a paragraph about the future roadmap.

Source coverage and integrations

AI dispatch is only as good as what it can see. Freight is scattered across load boards, broker portals, email, SMS, and a TMS that may or may not talk to anything else. If the system misses your highest-value board or cannot read the inbox where half your loads arrive, the automation is decorative and your dispatcher is still doing the real work in a second window. Pin down exactly what it connects to before anything else.

1. Which load boards do you pull from directly, and are those official integrations or scraped?
2. Do you read broker email inboxes, and which providers (Gmail, Outlook, Microsoft 365)?
3. Can you ingest loads sent by SMS, EDI, or broker portals, or only board postings?
4. How fresh is the data? When a load is posted or pulled, how long until it appears for my dispatcher?
5. Does the system connect to my TMS, and is that two-way or read-only?
6. When you add a new source, do I pay more or wait for a build, or is it already covered?
7. What happens to a load that appears on two boards at different rates? Do you dedupe and show the best one?
8. Can I keep using my existing board logins, or do I have to route everything through you?

Ranking and rules

The whole pitch of AI dispatch is that it sorts good loads from bad ones for your operation. So make it show its work. A ranking you cannot configure is just someone else's opinion of your business, and your operation does not run on the average carrier's assumptions. With operating costs around $2.26 per mile (ATRI, 2024) and broker margins averaging near 13.5 percent (DAT, 2023), the difference between a load that clears and one that loses money is often a few cents per mile. A ranking that cannot see your real floor is not ranking, it is guessing.

1. What exactly drives the ranking? Ask for the full list of inputs, not "our AI."
2. Can I set my own thresholds for rate per mile, minimum margin, and maximum deadhead?
3. Does ranking account for driver hours of service, home time, and current location?
4. Can I encode lane preferences, equipment fit, and customer-specific requirements?
5. Can I block specific brokers or shippers and have that stick across the whole team?
6. When a load is ranked first, can the dispatcher see why in plain language?
7. Does the system factor in market context (lane rates, load-to-truck balance), or only the posted number?
8. How do I correct a bad recommendation, and does the system actually learn from the correction?
9. Are rules set per dispatcher, per team, or only company-wide?

Deadhead alone is worth this scrutiny. Empty miles commonly run 15 to 30 percent of total miles, and a ranking that ignores the empty leg back is optimizing the wrong number.

Negotiation and human approval

This is where most products are either too timid to be useful or too aggressive to be safe. The right answer is a system that drafts and proposes while a human owns every commitment. Find the exact line where control passes from software to person, and make sure it is where you want it.

1. Does the system negotiate rates, and if so, does it send offers on its own or draft them for approval?
2. Where exactly does the workflow stop and wait for a human before a load is booked?
3. Can I set a rate floor below which the system can never commit, regardless of what it negotiates?
4. Can a dispatcher edit a drafted broker message before it sends, or is it send-only?
5. Who is on record as the contact with the broker, my dispatcher or your bot?
6. If the AI books a bad load inside my rules, who is responsible?
7. Can I run it in draft-only mode indefinitely, or does the contract push me toward full automation?
8. Does it handle the follow-ups (rate con, check calls), or stop at the handshake?
9. Can I see a full transcript of every negotiation the system ran on my behalf?
10. Does the system know when to stop negotiating and escalate to a human, or does it haggle a relationship into the ground?

A system that books without a clear human gate is a liability, not a feature. The right default is review-everything until your team trusts it, then loosen deliberately, lane by lane. The point of automation here is to give your dispatcher more shots on goal, not to take the dispatcher out of the loop. Median dispatcher pay sits around $46,860 a year (BLS, 2023); the goal is to make that person handle more loads well, not to replace their judgment with a model that has never met your brokers.

Fraud and broker safety

Cargo theft hit roughly $725 million in reported losses in 2025 (CargoNet), and double-brokering keeps climbing. An AI that books fast without vetting just helps you get scammed faster. Ask how the system protects you, not just how quickly it moves.

1. Do you verify broker authority and MC status before surfacing or booking a load?
2. How do you detect double-brokered loads, and what happens when you find one?
3. Do you screen against known-fraud lists or flag brokers with a history of disputes?
4. If a broker looks suspicious, does the system warn the dispatcher or quietly proceed?
5. Do you check that the booking contact matches the company on the authority?
6. What stops the AI from being socially engineered by a spoofed broker email?
7. Do you keep a record I can hand to my insurer or to law enforcement after a theft?

Security and data

You are about to give a vendor your broker relationships, your rates, and your customers. That is the commercial core of your business. Treat the data questions as seriously as the dispatch ones.

1. Where is my data stored, and who at your company can see my loads and rates?
2. Do you use my data to train models that other carriers benefit from?
3. Can I export everything and delete my account cleanly when I leave?
4. What is your security posture? Ask for SOC 2 status or an equivalent, not a reassurance.
5. If the AI sends emails, are they from my domain, and do I keep those records?
6. What happens to my data if you get acquired or shut down?

Pricing and contract

Per-seat pricing on a tool meant to reduce headcount is worth a hard look. So is any plan that charges per booked load, which quietly taxes your growth. Get the real number, including the parts that scale with you.

1. Is pricing per seat, per truck, per booked load, or flat, and what scales as I grow?
2. What is the real monthly cost for my fleet size, including any usage or overage fees?
3. Is there a contract lock-in, or can I leave month to month?
4. What does onboarding cost, and is it included or billed separately?
5. Is there a genuine trial where I can run it on live loads before paying? A 14-day trial on real freight tells you more than any demo.

Onboarding and support

The gap between a signed contract and a system your dispatchers actually use is where most of these tools die. Ask who carries you across it.

1. How long until my team is productive, and what does week one actually look like?
2. Can I pilot one lane or one dispatcher before rolling out to everyone?
3. When something breaks during a live booking, how do I reach a human, and how fast?
4. Who configures my rules at the start, your team or mine, and what does that cost?

The takeaway

A strong product answers these quickly and specifically. It tells you which boards it reads, shows you why a load ranked first, names the exact point where a human approves, and explains how it keeps a double-brokered load off your screen. A weak product reaches for "our AI handles that." The questions above are designed to force that distinction into the open while it is still a sales call and not a stuck driver.

If you want a reference for what the answers should sound like, Numeo's AI Hub is built around the model these questions assume: the AI finds, ranks, adds market context, and drafts negotiations, but the dispatcher approves every commitment. Bring this list to that conversation, and to every other one. The vendors worth buying from will be glad you asked.