Carrier Vetting Checklist: AI Checks With a Human Sign-Off

Let AI do the first pass: read the documents, compare names and dates, score fraud signals and draft a one-page brief. A person then verifies operating authority and insurance at the source, phones the insurer or agent, and signs the approval. AI shortens the checking; it never replaces the signature.

The reason is liability, not caution for its own sake. If a carrier you approved disappears with a load, the shipper asks who approved it and on what evidence. "The software said yes" is a poor answer; "I checked authority and insurance on this date and here is the record" is a good one.

Follow me on Instagram@sagnikteaches

What the checklist covers and who does each part

The split matters more than any single item, so start there. Machines are good at reading, comparing and counting. People are good at verifying against the outside world and deciding when something feels wrong.

Connect on LinkedInSagnik Bhattacharya

The documents themselves are covered in the carrier packet tutorial. This one starts when the packet is complete and someone has to decide.

Subscribe on YouTube@sagnikteaches

The checklist, grouped, with why each item is there

Authority and identity

• Authority is active and of the right type for your freight. A carrier with authority to haul property is not the same as one with authority limited to something else. Confirm both the status and the type on the public lookup, and note the date you checked. Why: freight moved by an unauthorised party can invalidate cover and cargo claims.
• Legal name, trading name and address match across the lookup record, the agreement and the certificate. Why: a mismatch is the earliest sign of a borrowed identity.
• Authority age. Note how long the authority has been active. A common approach is to send a new authority down a stricter route: a first load below a set value, a call to a reference, a shorter payment cycle. Your own threshold goes in your rules; the 90 days used below is an example, not a rule of law.
• Contact details are independent. Find the company's phone number from a source you did not receive from the carrier, such as the public record, and call it. Why: a fraudster controls the email, the phone number on the form and the website they point you to.

Insurance

• Limits at or above your floor. Auto liability and cargo, each against your written rule. A floor of $1,000,000 auto liability and $100,000 cargo is an illustrative house rule; the legal minimum for your freight type and your insurer's requirements may differ, so check both.
• Policy in force for the pickup date and the transit. Not just today. Why: a policy that ends on Wednesday does not cover a Thursday delivery.
• Certificate holder and insured name correct. The insured should be the carrier, not a third party.
• Confirmed with the agent or insurer by call or their verification channel. Why: a certificate is a copy of a claim about a policy, and copies can be altered. Record the name of the person you spoke to.
• Cancellation notices. Ask whether the agent will notify you if the policy lapses, and put a calendar reminder on the expiry date.

Safety record and operations

• Public safety figures reviewed. Paste what the public record shows and ask the AI to list anything that exceeds your thresholds, for example repeated out-of-service events. You decide what the thresholds are.
• Equipment fits the load. Trailer type, count and age of trucks, and whether the carrier has handled this freight before.
• Dispatch and driver contact. A named dispatcher and a number that answers outside office hours.

Payment

• Payee matches the carrier or a documented factoring company. Why: payment sent to a third party the carrier cannot explain is how funds disappear.
• Any change in payment details after approval triggers a call-back to the number already on file, never the number in the request.

Fraud signals the AI can score from the paperwork

A scoring list turns instinct into something repeatable. Give the model the signals and ask it to count them, then decide the outcome with a threshold you set. Each signal below is an indicator, not proof; honest carriers trigger some of them.

1. Authority active for fewer than 90 days (your threshold).
2. Contact email on a free webmail address or on a domain registered in the last few months.
3. Phone number is a virtual or forwarding number, or differs from the public record.
4. Certificate holder or insured name does not match the carrier name.
5. Insurance agent or insurer cannot be reached or the details look inconsistent.
6. Payee name differs from the carrier, with no factoring notice.
7. The carrier quotes well below the market rate on a high-value lane and agrees at once.
8. Reluctance to give driver name, truck number or tracking before pickup.
9. Pressure: "need the load now", "send the rate confirmation before the packet".
10. Rapid changes: a new dispatcher, a new email or a new bank account within weeks of approval.

A starting rule you can tune: zero or one signal, green; two or three, amber (a person calls and asks questions); four or more, red (no load until a manager has reviewed). The count only sorts the queue. The call still decides.

Double brokering: how it shows up and how to test for it

Double brokering usually appears after approval, not during it. A genuine carrier on paper accepts the load, then re-posts it to a load board, and a different company turns up at the dock. Vetting cannot stop that alone, so pair the checklist with checks around the load.

• Rate confirmation names the carrier, driver and truck. Ask for the driver's name and phone and the truck and trailer numbers before pickup.
• At pickup, compare what arrives with what you recorded. If the paperwork on the bill of lading shows a different company, stop and call.
• Tracking from the booked carrier. A tracking app that follows the driver, not just a dispatcher's promise.
• A clause in the agreement. A carrier agreement can prohibit re-brokering and state the consequence; ask your adviser for wording.

The AI can help on the load side by matching names: "Here is the rate confirmation and the pickup photo of the bill of lading. Do the carrier names match? List every difference." The match is quick, and a mismatch is something a person sees at once. For the email side of fraud, such as spoofed confirmations, the freight security tutorial goes further.

The pre-check prompt and a sample brief

Gather the packet extracts and the lookup text into one message. The model produces the brief the person reads before the calls.

You are preparing a carrier vetting brief for a freight broker.
Inputs below: extracted packet fields, text copied from the public
authority and safety record, and our rules (insurance floors, authority
age threshold, fraud signal list).

Output:
1. One-line summary and a recommended colour (green/amber/red) with the
   count of fraud signals triggered and which ones.
2. A table comparing name, address, phone, email and payee across every
   document. Mark each mismatch.
3. Insurance: limits against our floors, dates against pickup date, holder.
4. Questions I must ask by phone (maximum 6), each tied to a flag.
5. What you could NOT check from the inputs.

Do not state that anything is verified. Only I can verify.

An illustrative brief: "Amber. Two signals: authority active 71 days; payee name differs from carrier (no factoring notice attached). Names match across packet and lookup. Auto liability $1,000,000, cargo $100,000, both meet floors; expiry 14 Aug next year. Calls: (1) agent, confirm policy numbers and cargo limit; (2) carrier, explain payee; (3) carrier, name two shippers they have hauled for. Not checked: current authority status, agent's identity, driver details." Notice the last line, which tells the person exactly what remains theirs.

Three carriers through the checklist

A steady regional fleet, green. Authority active for years, names consistent, limits above the floors, agent confirms by phone in four minutes, payee matches. Total person time: 12 minutes. Approve for any lane.

A new authority with good paperwork, amber. Eight weeks old, owner-operator, clean certificate. The call to the agent confirms the policy; a call to the owner reveals two years driving for another firm. The broker approves for a lower-value first load, with tracking and a payment cycle agreed in writing, and notes the review date.

A carrier that rings alarm bells, red. Quoted rate 22% below the lane average, webmail address, payee in a different name and a request to send the rate confirmation straight away. The agent cannot be reached on the number printed on the certificate; the number found independently belongs to a different agency and has never heard of the policy. The broker declines and records the reason. That record is the reason to keep the checklist.

A filled-in sign-off block

The sign-off is the part nobody may delegate. Keep it short enough that it gets completed every time.

CARRIER APPROVAL RECORD
Carrier legal name: [name]        Authority number: [number]
Authority status checked at source on: [date] by: [name]  Result: [active / not]
Insurance confirmed with: [agent/insurer, person spoken to] on [date]
  Auto liability: [amount]  Cargo: [amount]  Expiry: [date]
Fraud signals triggered: [n] - [list]    Outcome: [green / amber / red]
Conditions (if amber): [e.g. first load under $X, tracking required]
Approved by: [name, role]   Date: [date]
Next review: [date]

If the approver changes an AI recommendation, write why. "Overrode amber because the carrier is a subsidiary of a firm we have used for six years" is a good record. A blank override is a problem in a dispute.

The phone call to the insurance agent, word for word

The call is the check that nothing else substitutes for, and people skip it because they do not know what to say. A short script removes the awkwardness. Find the agent's number from the insurer's own website or a business directory, not from the certificate. Then ask:

1. "I'm calling from [your company]. I hold a certificate for [carrier legal name], policy number [number]. Can you confirm that policy is in force today?"
2. "What are the effective and expiry dates, and the auto liability and cargo limits?"
3. "Is [our company] listed as certificate holder, and will you tell us if the policy is cancelled or not renewed?"
4. "Are there any exclusions or conditions for the freight we move, for example [commodity]?"
5. "Can you email the certificate from your own address so I have a copy from the source?"

Write down the person's name, the time and what they said. An agent who is evasive, says they cannot confirm anything, or does not recognise the policy is giving you the answer you called for. The AI can prepare this script for each carrier from the brief, filling in the numbers and adding questions about the specific flags, and it can turn your rough call notes into a tidy record afterwards. It must not do the call, and it must not decide that a note saying "seemed fine" is enough.

Testing the AI pre-check on carriers you already know

Before you rely on the pre-check, run it on ten carriers you have good reason to understand: six you trust, two you had trouble with, and two you rejected. Remove their names if you prefer and keep the documents. A sensible target is that none of your two bad carriers comes out green, and that fewer than three of your six good ones come out red.

Look at the misses closely. A good carrier flagged amber because its trading name differs from its legal name tells you to add a "doing business as" field to the packet. A bad carrier flagged green tells you which signal your list lacks. Adjust the list and the prompt and run the ten again. This usually takes an afternoon, and it gives you a result you can show an insurer: a tested routine rather than a hopeful one.

How AI pre-checks go wrong in vetting

• Claiming verification. A model may write "authority verified" because the packet contains an authority letter. Forbid the word in the prompt, as in the sample above, and make it say "stated in the packet, not checked".
• Treating the pasted lookup as current. The record you pasted yesterday may be out of date today. Put the date of the lookup in the prompt and re-run lookups on the day of approval.
• Smoothing over mismatches. Models are generous readers and will treat "Haulage" and "Transport" as the same business. Tell it that any difference is a mismatch, and let a person decide if it is harmless.
• Over-trusting its own colour. Green means "no signals found in what I was given", which is weaker than "safe". Rename the output "no signals found" if your team reads green as approval.
• Leaking data. Carrier files hold policy numbers and tax details. Use a plan that does not train on your content and keep originals in your own storage.

Making the sign-off stick in a small team

A checklist fails when the busiest person skips it on a Friday afternoon. Two habits help. First, tie the booking to the record: nobody can issue a rate confirmation to a carrier whose row in the tracker lacks an approval date, and the tracker is where dispatch looks first. Second, rotate who audits. Once a month, someone other than the approver pulls three approvals at random and checks that each has its lookups, call notes and signature. The audit takes about half an hour and finds more than any new tool.

In a team of three, the owner might approve everything for the first month, then delegate green carriers to a named dispatcher while keeping amber and red. Write the delegation in the same document, with the date. The AI can compile the monthly audit sample and the list of approvals that lack a field, which saves the dull part.

What a miss costs against what vetting costs

A quick sum puts the habit in proportion. Say a stolen load is worth $18,000 and your cargo cover has a deductible of $2,500 and a claim history that raises next year's premium. If the claim lands on you in full, one theft wipes out the margin on 60 average loads at $300 a load. Against that, 12 to 15 minutes of person time per new carrier, at an hourly cost of $35, is $7 to $8.75 a carrier. Even if you onboard 30 carriers a month, the checking costs at most about $262 a month, roughly $3,150 a year, which is still well under the value of one stolen load. These are illustrative figures; use your own margin, deductible and wage.

Liability and what to ask your adviser

The questions are practical. What does your contract with the shipper make you responsible for? What does your contingent cargo cover require you to check, and in what form? Does your insurer expect you to keep records of each approval? What does your carrier agreement say about re-brokering and indemnity? An AI can draft these questions and summarise your policy wording, but it cannot tell you what a court would decide, and the answers differ between places and contracts.

Whatever the answers, the practical rule is to keep the record: the brief, the dated lookups, the call notes and the signed block, stored with the carrier's file. If you want help setting up the whole routine with your own rules and tools, that is what my AI implementation consultation covers.

Keeping the checklist current

Review it every quarter against what went wrong. Add a signal that you saw in a real incident, adjust the age threshold if too many good carriers land in amber, and drop items nobody ever acts on. Re-vet carriers at renewal and whenever a load goes wrong; the load email and rate confirmation tutorial shows where the load-side checks fit. When a claim does arrive, the freight damage claim tutorial covers the evidence you will want to have collected.

Questions about vetting carriers with AI

What is double brokering?

Double brokering is when a carrier you booked passes the load to another carrier without your agreement, often through a load board. The broker and shipper then deal with an unknown party whose insurance and authority nobody checked, and payment disputes and cargo losses become far harder to resolve.

Can AI verify a carrier's insurance?

AI can read a certificate and spot gaps or odd details. It cannot confirm the policy is active. Only the insurer or the agent named on the certificate can, so a person contacts them using details found independently of the carrier's paperwork.

How long should vetting take per carrier?

With the packet already collected and AI pre-checks done, a person's part is often 10 to 15 minutes: lookups, one call to the agent, a read of the flags and a signature. A carrier with red flags takes longer, which is the point.

Who is liable if I approve a bad carrier?

That depends on your contracts, your insurance and the rules where you operate. Your shipper contract may make you answerable for the freight. Ask your adviser and insurer what your exposure is and what records they expect to see, and keep the sign-off record.

How often should approved carriers be re-vetted?

At least whenever a certificate renews, and on a schedule tied to use, such as monthly for weekly carriers and quarterly for occasional ones. Re-check straight away if a load goes wrong, a driver details mismatch appears or the carrier's contact details change.

Further reads

• AI for Freight Business Optimisation: Quotes, Loads and Follow-Ups — Quotes, loads and follow-ups once carriers are approved.
• Freight Invoice Audit: AI Checks Carrier Bills Against Quotes — Check carrier bills against the quote.
• Electronic Proof of Delivery: Settle Disputes With Photos and AI — Settle delivery disputes with photo evidence.
• Demurrage and Detention: AI Tracks Free Days and Drafts Disputes — Track waiting charges once loads are moving.
• Detention Pay: AI Tracks Wait Times and Invoices Brokers — See waiting-time claims from the carrier's side.
• Walmart OTIF Fines: Build a Dispute Pack From Delivery Records — Which Walmart OTIF fines are worth disputing, the five records to gather, and how AI assembles one evidence pack per PO.
• AI Agents for Logistics Companies: Which Tasks They Handle Today — Which logistics tasks an AI agent can take on today, which need a person to approve, and which to leave alone.
• AI for Owner-Operators: Eight Jobs You Can Hand to ChatGPT — Eight admin jobs for a one-truck owner-operator, each with a prompt, a worked figure and a check.
• Can AI Read Bills of Lading and Fill Your TMS? An Accuracy Test — How a small forwarder can test AI on 100 bills of lading and air waybills, and set up validation and an exceptions queue.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: general freight-brokerage practice. Legal minimums and liability differ by place and contract, so check yours with your adviser and insurer; worked figures are illustrative.