AI Agents Now Run This Transport Company's Bookings, Dispatch and Follow-Ups. Nobody Types Anymore.

A transport business I work with used to run on typing. Booking requests came in by email, web form and SMS. Someone read each one, typed it into the CRM, typed a confirmation back to the customer, typed the job onto a run sheet, and then, if anyone remembered, typed a follow-up a few days later. Multiply that across every job, every day, and you've got a capable person spending their whole week doing data entry with a phone wedged under their ear.

Today nobody types any of it. Bookings, dispatch and follow-ups run on AI agents wired together with n8n. Not a chatbot bolted onto the website — the actual operational spine of the business. I built it, it's in production, and this is the honest build log, including the parts that broke. I won't name the client. Everything else here is real and running right now.

The before picture

The manual version had three failure modes, and every small operator will recognise them.

Bookings got missed whenever the one person who handled them was sick, driving, or just flat out. Follow-ups were the first casualty of any busy week — quotes went out and died in someone's inbox because nobody chased them. And the CRM was fiction. Plenty of jobs never made it in, which meant the owner had no real picture of repeat customers, quiet periods, or which work actually paid.

None of that is a people problem. It's what happens when a business grows past what one inbox and one brain can hold.

The architecture, at founder altitude

You don't need to read a line of code to follow this build. There are three layers.

Triggers. n8n watches every channel a booking can arrive through: the inbox, the website form, SMS. The moment a message lands, a workflow fires. No polling by a human. No 'I'll check emails after this run'.

The agent. The message goes to an AI model — I use Claude and GPT-4 depending on the job — which turns a messy human request into structured data: pickup, drop-off, date, load details, contact. It's grounded with RAG over the company's own pricing and service rules, so when it quotes, it quotes from the business's actual rates, not from the model's imagination. That distinction is most of the game.

The systems it touches. The agent writes the booking into the CRM, slots the job into dispatch, notifies the driver, and sends the customer a confirmation. Every step is logged. Every record stays consistent, because no human is re-keying anything between systems.

Where the agent decides — and where it doesn't

This is the part most AI-automation content skips, and it's the part that separates a production system from a demo.

A standard, complete request gets handled end to end. Quote, confirm, book, dispatch. No human involved. A request with missing details doesn't get guessed at — the agent replies to the customer and asks. The back-and-forth that used to eat staff time now happens between the customer and the workflow, at whatever hour the customer happens to be awake.

And anything unusual — an odd load, an address that won't resolve, a price outside normal bounds — drops into a human escalation lane with the agent's draft attached. A person reviews it and clicks approve. Seconds, not minutes. The rule I build to is simple: the agent has to know what it doesn't know. An agent that guesses confidently is worse than the manual process it replaced.

Follow-ups that never forget

Here's the bit that turns an ops build into a revenue build. Follow-up is marketing. Most small businesses just never do it consistently, because it's nobody's actual job.

So the workflows do it instead. Confirmation the moment a booking lands. A reminder before pickup. After the job, the invoice goes out with a review request. Quotes that haven't been answered get chased on a schedule until they're won or dead. Customers who've gone quiet get a nudge.

A human forgets these things by Wednesday of a busy week. A workflow has never once forgotten. That reliability compounds. Reviews stack up, quotes stop leaking, old customers come back. And it costs nobody any extra effort.

What broke (and how it got fixed)

Now the honest part. Three things went wrong, and each one changed how I build every agent since.

Dates. Humans write 'next Friday' and 'end of the month', and the model would parse them confidently, and sometimes wrongly. A confidently wrong booking is a disaster. The fix: strict structured output, validation on every date and address, and a confidence threshold. Anything below it routes to a human instead of proceeding.

Silent failures. Early on, an API hiccup killed a workflow mid-run and it died quietly. A booking nearly vanished — the exact failure the whole system existed to prevent. Now every workflow has an error path that alerts a human the moment something breaks. Automation is allowed to fail. It's never allowed to fail silently.

Tone. I originally let the model write customer messages freely. It started fine, then drifted: a bit too chirpy, a bit too long. The fix was tight message templates the agent fills in, with freedom only where it's safe. Customers should never be able to tell a machine sent it. Now they can't.

What this actually means for a business like yours

The person who used to type bookings all day still works there. They now handle the escalation lane and the customers who genuinely need a human — the work software can't do. Everything else runs 24/7, weekends included, no sick days, no Monday backlog.

There's a line in my pitch deck: the output of a full marketing team at the cost of one person. I mean it literally. Booking capture, CRM hygiene, follow-up sequences, review generation, reactivation — that's several job descriptions' worth of output, running as software while you sleep.

I've been doing this a while. Over 10+ years I've helped 110+ businesses grow, and I built my own e-commerce brand from nothing to 7-figure annual revenue. Every dollar of ad spend was my own money — that teaches you to optimise harder than anyone working with someone else's budget. It's also why I build automations that protect revenue first and look clever second.

Where to start

Not by automating everything. You start with the one manual process that's quietly costing you the most, and in most businesses I look at, it's follow-up.

If you want to see what a version of this looks like for your business, send me your website URL and I'll put together a free, personalised 30-Day Growth Plan within 48 hours. Real recommendations for your specific business, not a template. And I work month-to-month with no lock-in, because the results should be the reason you stay.

Nobody at that transport company types bookings anymore. There's no reason your business should either.