FreighAI
FreighAI · Playground article
Lesson 3 of 6 · 10 min read

What can go wrong with AI in a forwarding office, and what counters it?

Every new tool brings risk. AI brings its own kind. This page names seven risks in plain words. Ours are on the list too.

None of them is a reason to stop. Each one has a counter, and most counters are questions you ask before you sign.

See the seven risksThe twelve questions
The short answer

Seven things can go wrong. Wrong answers. Your data. Getting stuck with one supplier. The team not using it. A half-done start. Buying the wrong thing. Doing nothing. Each one has a counter, and most counters are questions you can ask out loud before you sign anything.

Lesson 3 of 6 · 10 min read

By the end of this lesson you can

  1. Name the seven risks and the counter for each.
  2. Say which risk is yours, by your role.
  3. Ask the five data questions before signing anything.
  4. Put a cost on doing nothing for a year.

What are the seven risks?

These seven cover almost everything that goes wrong.

Read the third column first. The counter is nearly always a habit or a question, not a feature you buy.

The seven risks of AI in a forwarding office, and the counter to each.
  1. Wrong answers
    What can go wrong
    The tool prices a lane wrong. Or it promises a date it cannot keep.
    What counters it
    A person approves every reply before it sends. Every quote version is kept. Ask any vendor to show you one it got wrong.
    Which programme helps
    The session, the workshop
  2. Your data
    What can go wrong
    Your rates, customers and margins end up somewhere they should not.
    What counters it
    Ask where the data sits. Ask who can see it. Ask whether it teaches anything used for another firm.
    Which programme helps
    The assessment
  3. Getting stuck
    What can go wrong
    You stop, and you cannot get your data out.
    What counters it
    Ask for the export before you sign. Keep your TMS as the record if you want to.
    Which programme helps
    The checklist, the session
  4. Adoption
    What can go wrong
    The team does not use it.
    What counters it
    One task screen instead of the inbox. People approve instead of write. The desk tries it on its own enquiry first.
    Which programme helps
    The workshop
  5. Half-done adoption
    What can go wrong
    Bought, set up, never really running.
    What counters it
    Lane by lane, with the team in the room, until it runs on your own data.
    Which programme helps
    The TAT reduction programme
  6. Buying wrong
    What can go wrong
    An old system with AI buttons, sold as AI.
    What counters it
    The five tells and the twelve questions. Ask for each one on a real enquiry of yours.
    Which programme helps
    The webinar, the checklist
  7. Doing nothing
    What can go wrong
    Another year of quotes at one to two hours each, while the first good quote keeps winning.
    What counters it
    Know your number. Then decide.
    Which programme helps
    The assessment

Six of the seven are about how you buy and how you start. Only one is about the software itself. That surprises most people.

The last row is the odd one out. Nobody sends you a bill for it, so nobody counts it.

Which risk is mine?

Each desk feels a different one first. Find yours here.

  • Owner. Buying wrong, and doing nothing. You carry both. Ask to see one real enquiry go all the way through before you sign.
  • COO. Half-done adoption. Bought and never really running is the usual ending. Pick one lane and finish it.
  • Pricing desk. Wrong answers. A bad price in front of a customer is the fear. The approve step is your counter.
  • Operations desk. Adoption. A new screen on top of a busy day is a hard sell. Try it on your own job, not a demo one.
  • Accounts. Your data. Bills, margins and customer terms all sit with you. Ask the five data questions below.

Your first risk is not the whole list. It is where you start reading.

What does “a person approves” actually protect against?

It protects against a wrong answer reaching your customer.

Your team approves every reply before it sends. That one rule is what makes the rest safe to use.

Your team approves. They do not compose, forward, file or chase. The reply goes out in their name.

A worked example: the wrong price, caught in time

An enquiry lands for four pallets, Jebel Ali to Hamburg.

The draft quote comes back in minutes. It reads well. It is still wrong.

The rate line came off a card that ran out last month. The pricing person spots it, swaps the line, and approves.

The customer never sees the wrong price. Nothing was saved, because nothing was lost. That is the gate doing its job.

Every version of a quote is kept, and the line that changed is shown. So next week you can see what the draft said, what went out, and who changed it.

Where the wrong price stops
    1. 01Enquiry landsFour pallets, Jebel Ali to Hamburg.
    2. 02Rates fetchedCards, carrier and past prices, read for you.
    3. 03Draft writtenA priced quote, ready to read.
    4. 04Old rate spottedThe card ran out last month.
    1. GATEA person approvesThe line is swapped, then it goes out.
    2. 06Sent in their nameThe customer sees one price, the right one.
    3. 07Version keptWhat changed, and who changed it.
The same enquiry, step by step. Six steps happen without you. The fifth is a person, and nothing passes it without a yes.

A gate only works if it sits where people already are. One example. Your people work in FreighAI, not in the inbox. The enquiry, what was found and the draft all sit on one screen.

Ask any vendor two things here. Whose name goes on the reply. And what the tool does when nobody approves.

What should I ask about my data before I sign anything?

Five questions. Ask them out loud, and write the answers down.

  • 01Where does my data sit? Name the country and whose machines.
  • 02Who inside your company can open my rates and my margins?
  • 03Does my data teach anything that another firm then uses?
  • 04Can I export all of it, any day, in a file I can read?
  • 05What happens to my data on the day I stop paying?

A good answer is short and plain. A long answer with no country in it is not an answer.

The same rule holds in the Playground. Your numbers are never shown to another firm.

Ask what it does not do, as well

Scope is where quiet surprises live. Ours, in the open. On its own, FreighAI runs your enquiries, quotes, jobs, tracking replies, carrier bills, payment follow-ups and your accounts. FreighAI does not file customs declarations. It treats customs as a service provider or agent, and your team adds and updates the filing details by hand.

A vendor who answers that question plainly is easier to trust than one who says yes to everything.

Recommendation

Get the export answer in writing before you sign, not after. A tool you can walk away from is a tool you can trust. If the answer is vague, that is your answer.

What does doing nothing cost?

A year. Here it is in round numbers.

Take a desk of four people. They handle 20 enquiries a day. Each quote takes one to two hours by hand.

That is 20 to 40 hours of quoting, every day.

Where the rate can be fetched without waiting on a person, the same quote goes out in 10 to 15 minutes instead of 1 to 2 hours.

Not every lane works that way. Where an agent has to price it by hand, the wait belongs to whoever is answering.

Sit on that gap for a year and the cost is a year of it. This page exists so that nobody sits on it out of ignorance.

Recommendation

Count your own quotes for one week before you talk to anybody. Enquiries in, hours spent, quotes sent, quotes won. Then decide with your own number in front of you, not ours.

Who is in control?

The worry we hear most: what if the AI misbehaves, and do I have any control? Here is the honest answer, part by part.

The control model, part by part. Ask any vendor to show you each one on a real thread.
  1. What always waits for a person?
    The answer
    Every message a customer sees waits for a person. The agent does the work, creates a task, and notifies the person to sign it off. Quotes, counter-offers, tracking answers, anything a customer reads.
  2. What can the person do at the sign-off?
    The answer
    At the sign-off, the person can edit the message, approve it, or reject it. It is an editable panel, and that is the point of it. Every version of a quote is kept, and the line that changed is shown.
  3. Whose name goes on it?
    The answer
    An approved message goes out in the person’s own name, with their own signature, from their own email. No customer ever sees that AI works behind it.
  4. What runs on its own, and under whose rules?
    The answer
    Autonomy is set per task type: supervised, where a person signs, or automatic. Every message a customer sees is locked to supervised. Chasers, reminders, tracking checks and quote expiry can be set to run on their own. The rules and timing for chasers and reminders are set by the head of operations or an admin. Quote expiry rules are set by the pricing executive. Chasers and reminders can be switched off for one customer or one lane.
  5. Who decides who may approve what?
    The answer
    Role-based access control: the admin decides who can see, edit and approve what, role by role. A junior desk can be allowed to draft and not to send.
  6. So what does 85% mean?
    The answer
    The 85% is effort taken out of the cycle: the reading, fetching, drafting and chasing. A person still signs every customer-facing message.
  7. What is not there yet?
    The answer
    There is no pause switch yet for one customer, one lane or the whole company. Today the person at the sign-off is that control: nothing a customer sees goes out without a person. Such cases are the exception. There is no automatic margin floor yet. Today it is the staff’s job at the sign-off to keep a quote above the margin. Automatic floors are planned.
In one line

The agent works, a person signs, the rules are yours, and the customer only ever sees your name.

Where do I go next?

Start with the risk that worries you most.

  • Afraid of buying wrong? Read the five tells that separate an add-on from a system built for this.
  • Close to signing? Take the twelve questions into the call, in writing.
  • Afraid the team will not use it? Sit in the workshop and watch it run on a real enquiry.
  • Bought something that never really ran? The TAT reduction programme is built for exactly that.

All of it is free to read. Nothing on this page needs a sales call first.

Questions

Common questions

01

Is AI too risky for a small forwarding office?

No, but going in blind is. Six of the seven risks are handled by questions you ask before you sign. They cost you nothing. The seventh is doing nothing, and nobody sends you a bill for that one.

02

Which of the seven is the biggest?

Half-done adoption. Buying is the easy part. Running it lane by lane, with the team in the room, is the part that gets dropped when the week gets busy. That is why the counter is a programme and not a feature.

03

Does a person approving every reply slow the desk down?

Reading a draft takes seconds. Writing one takes minutes. The approve step is where the time is saved, not where it is lost. It is also the only place a wrong price can be caught before a customer sees it.

04

Do these risks apply to your own product too?

Yes, all seven. Ask us the same questions you would ask anyone else. Ask to see one it got wrong, ask what happened next, and ask for the export terms in writing.

Sources
Previous
← Lesson 2 · Where AI pays
Next
Lesson 4 · Add-on or AI-native →
READY WHEN YOU ARE

Take the seven risks into your next call

Print the third column. Ask for each counter out loud, and write down what you hear the same day. A vendor who welcomes the questions is telling you something. So is one who does not.

The twelve questionsAll Playground explainers