Doing more loads is one lever. Collecting what you already billed is the other.
A back-office agent is a program that reads the documents arriving by email, records the facts nobody wrote down, and checks each invoice line against the document that proves it. Ask how a twenty-person brokerage should use AI and every answer is about volume: automate the inbox, the quoting, the carrier outreach, get more loads out of the same people. That is sound and we would tell you the same. It is also half the question, because the money you invoiced and wrote off last quarter needs no extra loads and no extra people to recover either.
Two numbers decide which half to fix first, and most operations have only ever counted one of them.
How can a small freight broker use AI to do more loads without hiring?
By removing the highest-volume manual work first: quoting from email, load building, and carrier outreach. At twenty people those three consume more hours than anything else, the tooling for them is mature, and several vendors sell to operations your size. This part of the consensus is right and we are not going to argue with it.
What it is worth is easy to estimate honestly: count the documents one person re-keys in a day, multiply by the minutes each takes, and multiply by your loaded hourly cost. That is the ceiling of what extraction can give back. It is usually a real number and it is usually smaller than people expect, because the hours are spread thin across a team rather than concentrated in one role you could remove.
Why does nobody count the money already invoiced?
Because it was written off one invoice at a time, so it never appears as a number. A detention charge gets disputed, arguing costs more than the charge is worth, and it goes. Repeat that weekly for a year and it is a headcount, but no report anywhere shows it as one.
The scale is not in doubt. The American Transportation Research Institute (ATRI), in its September 2024 report Costs and Consequences of Truck Driver Detention, found that 94.5% of fleets bill for detention and are paid on fewer than half of those invoices. That is a count of invoices, not of dollars. What is in doubt, always, is your own share of it — which is exactly why it is worth an afternoon to count before anyone buys anything.
| More loads per person | Collect what you billed | |
|---|---|---|
| What it fixes | Hours spent re-keying and chasing. | Revenue already earned and written off. |
| How to size it | Documents re-keyed per day, times minutes, times hourly cost. | Accessorials invoiced last quarter, minus accessorials paid. |
| Who sells it to you | Most of the market. Inbox agents, voice agents, extraction. | Few vendors, because it is a change to what gets recorded rather than a feature. |
| What it needs from you | An integration and a few weeks of adoption. | An agent that records the facts no document carries, as the load moves. |
| Where it fails | When the document it reads was missing the field anyway. | When the facts are reconstructed after the load instead of captured during it. |
How do I calculate what manual data entry is really costing my company?
Count the documents your people re-key in a day, multiply by the minutes each takes, then multiply by your loaded hourly cost. That product is the ceiling of what manual data entry costs in wages, and it is a formula rather than a measurement, so count your own. Run it over a typical week rather than the best one, and treat the result as your number: nobody else can measure it for you, and any industry average you borrow will be wrong for your lanes.
Besides wages, what hidden costs come with manual data entry?
The wage line is the smallest part of what manual data entry costs, because the rest never shows up as a line at all. A wrong keyed field means rework. A disputed charge can cost more to argue than it is worth, so it is written off. Invoicing is delayed while paperwork is chased. Repetitive roles turn over, and every hour someone spends keying is an hour not spent covering loads.
How do I build a business case for reducing manual entry?
Put two measured numbers side by side: what re-keying costs in hours, and the gap between accessorials invoiced and accessorials paid. Use last quarter’s own records for both, and add the hidden costs as named items rather than guessed amounts. For scale, ATRI reports detention cost the industry $15.1 billion in 2023: $3.6 billion direct, $11.5 billion lost productivity.
What AI tools do 3PL operations teams actually need?
One of three, depending on which job is costing you most: intake and extraction, execution and visibility, or billing and dispute. Roundups cover the first two thoroughly because that is where the vendors are. The third is where a multi-client 3PL quietly loses margin, and it behaves differently: the loss is per-event, small, and absorbed rather than escalated.
It is also the one that gets worse with scale rather than better. More clients means more accessorial schedules, each with its own conditions, and a dispute answered from memory instead of from a dated record is a dispute you concede.
What should a small brokerage automate first?
Measure both numbers for one month before buying anything. At twenty people you cannot absorb a nine-month pilot that turns out to have solved the cheaper problem, and that is the common outcome rather than the rare one. catuh builds agents for three jobs, and the order you start them in is the decision.
- 01 / 03
Read the documents that arrive by email
Rate confirmations, bills of lading, proofs of delivery and carrier packets come in as attachments. The agent reads each one and writes its fields into the system you already use, with anything it cannot read escalated to a person.
What you get: Fewer hands re-keying the same paperwork.
- 02 / 03
Capture the facts no document records
Arrival and departure times, the exception at the pickup and the carrier verification are not on any form because nobody wrote them down. The agent records each one as the load moves, with its source and the time it was captured.
What you get: A dated record for the facts that decide a dispute.
- 03 / 03
Match every invoice line to the document that proves it
Before an invoice goes out, the agent checks each line against the document that supports it. A line with no proof behind it is held back for a person instead of being sent and disputed later.
What you get: Invoices that carry their own evidence.
The measuring costs nothing and gets skipped. You can start it this week without us: run one month of accessorials you invoiced against accessorials you were paid, and put the gap next to the re-keying estimate. Whichever number is bigger is your project, and it is not always the one being demoed to you.
Frequently asked questions
How can a small freight broker use AI to do more loads without hiring?
By removing the highest-volume manual work first, which for most twenty-person brokerages means quoting from email, load building, and carrier outreach. That is the standard answer and it is correct. It is also only one of the two levers. The other is collecting what you already billed: the American Transportation Research Institute's September 2024 report found that 94.5% of fleets bill for detention and are paid on fewer than half of those invoices. Recovering that needs no extra loads and no extra people, and most automation plans never look at it.
What AI tools do 3PL operations teams actually need?
It depends which of three jobs is costing you most: intake and document extraction, execution and visibility, or billing and dispute. Most roundups cover the first two because that is where the vendors are. The third is where small operations quietly lose margin, because a disputed accessorial is written off rather than fought, and each write-off is smaller than the cost of arguing about it.
Is a twenty-person brokerage too small for AI automation?
No, but it is small enough that the order matters more than the tooling. At that size you cannot run a nine-month pilot or absorb a failed integration. Start by measuring two numbers for one month: how many documents a person re-keys, and how much you billed and did not collect. The first is an extraction problem with mature solutions. The second is a capture problem, and no extraction tool fixes it.
What should a small brokerage automate first?
Whichever of the two costs more, measured rather than assumed. Re-keying is easy to count: documents per day times minutes each. Uncollected accessorials are easy to count too and almost nobody does it, because the invoices were written off one at a time. Put both numbers on one page and the order usually becomes obvious, and it is not always the one the demo is selling.
References
Every figure and every quotation on this page traces to one of these. Each was retrieved and read rather than cited second hand.
- ATRI, Costs and Consequences of Truck Driver Detention, September 2024 Source for the share of fleets that bill detention and the share of those invoices paid.
- DOT OIG, Report ST2018019, Commercial Driver Detention, January 2018 Federal audit of how driver detention is recorded and compensated.
Count the second number first.
Tell us what you invoiced in accessorials last quarter and what came in. If the two match, the collecting lever is not yours and we will say so, and you should spend the money on extraction instead. If they do not match, the gap is the size of the project.
A third number is worth having beside those two: how long it takes a new hire to bill an accessorial correctly without asking anyone. That is what standardising shortens.
The brief asks which process costs you money, and a person answers it in writing.