AI automation for freight forwarders: the naked sales desk

Walk into a small freight forwarder and you will find two different companies. The operations desk is systematised: a TMS moving files from booking to delivery, customs filings going out electronically, milestones tracked. The sales desk, three feet away, is a shared inbox. Rate requests arrive buried in forwarded threads, quotes get priced from a spreadsheet and someone’s memory, and follow-up happens if somebody remembers. Ops is tooled. Sales is naked. That gap is where AI automation actually pays off for a forwarder, and this note is about why the gap exists and what to do about it first.

The short version:

• Fewer than 40% of forwarders run a forwarding management system at all, per the Adelante SCM State of Digitization research (2025). The share running a real sales system is far smaller.
• When Freightos surveyed the top 20 forwarders in 2018, manual quoting averaged 57 hours from enquiry to quote, with a range of 23 to 141 hours.
• Contacting a lead within the hour made firms nearly seven times more likely to qualify it (Harvard Business Review, 2011, across 1.25 million leads). That is general B2B evidence, not a freight number, but a rate desk lives it daily.
• The fix is intake first, documents second, and a person signing off on anything that touches a licence.

Why is the sales desk still manual when the ops desk is not?

Because every system a forwarder buys is an operations system. The TMS exists to move a file and keep CBP happy, and it only meets a shipment after someone has read the email, priced the lane and typed the details in. The big platforms ship CRM modules, and almost nobody at the small end uses them; read the reviews and you will find complaints about support, cost and learning curve, and none about the CRM, because the CRM sits untouched. Meanwhile the rate desk runs on Outlook, a rate sheet that arrived as an Excel attachment, and whoever answered last.

That is not laziness. A five-to-fifty-person forwarder is short-staffed by structure: the same people quoting are chasing a chassis, explaining an exam fee that a random selection caused, and racing an ISF clock that runs against loading at origin, not arrival. The desk does not need another platform to log into. It needs the reading, extracting, drafting and remembering taken off its hands, inside the tools it already uses.

What does slow quoting actually cost?

More than the lost file. A spot quote that goes out in hours instead of days wins on responsiveness before rate even enters the conversation, and most SMB forwarder sites still promise no turnaround at all. The quiet cost is visibility: when quotes live in an inbox, nobody can say which lanes are carrying the year, which customers get quoted and never book, or how much open value is sitting unchased. Margins in this business are thin enough that one stale rate sheet eats the file. You do not need my number for this; take your own quotes per month, your average file value and your win rate, and do the arithmetic. There is a calculator on my freight page that does nothing more clever than that.

What should a small forwarder automate first?

Intake, in this order: rate requests captured out of email threads into one pipeline (lane, cargo, weight, dates, terms), quote drafts written in your voice for a person to price and send, and follow-up that chases every open quote until it books or dies honestly. A useful side effect: the written rate trail exists before the cargo loads, which your compliance side will appreciate.

The docs desk is the second act, not the first: pre-alerts and arrival notices read on arrival, commercial-invoice and customs fields pre-filled for a human check, carrier invoices reconciled against the quote. It pays, but intake pays faster and proves the approach.

How you buy it matters as much as what you buy:

Route	Best for	The main tradeoff
Enterprise AI platform	Top-50 forwarders with volume minimums to clear	Priced and sized for the big end; you rent it forever
TMS add-on module	Firms already deep in one vendor’s stack	The sales desk still has to come to the ops system
A build you own	Small desks with one specific leak	Scoped to you; you own it and run it, and no meter runs on your growth

What should you not automate?

Anything with your licence or your judgment on it. Customs sign-off, pricing decisions, claims, anything with legal exposure: a person does those, every time. Most AI failures in this industry come from pretending otherwise. The honest shape of the system is that it removes re-typing and forgetting, and leaves responsibility exactly where CBP expects to find it.

Who this is not for: if you are quoting five files a month, a shared-inbox rule and a same-day reply habit will beat any system, and if you want a platform with a helpdesk and a logo wall, I am the wrong shape of supplier. I am one builder, and I say no to most of what I am shown.

If the gap I described looks like your desk, start with what I would automate first, or just send me one real rate request exactly as it arrived and I will tell you honestly whether it is worth building anything at all.