The real asset isn't the TMS. It's what your team has learned.
A brokerage's system of record holds the load, the rate, and the status. It does not hold why a particular shipper's receiving dock refuses any driver who shows up outside a two-hour window, or why a dispatcher quietly stopped tendering a certain lane to a carrier that technically still passes every compliance check. That knowledge is real, it is accurate, and it took years to build. It just does not live anywhere a system can search.
Instead it lives in an email thread from eighteen months ago where an account rep explained a shipper's appointment rules to a coordinator who has since moved to a different desk. It lives in a contract PDF that nobody has reopened since it was signed, with a detention clause on page eleven that only gets read when a dispute is already underway. It lives in a customs broker's reply from the one time your team moved a specific commodity across a specific border, filed away in a folder nobody remembers the name of. And it lives, more than anywhere else, in the head of the dispatcher who has covered that lane for six years and just knows.
None of that is a documentation failure in the way people usually mean it. The knowledge exists. It is accurate. It simply was never written down in a place a new hire or a night-shift coordinator can find it without asking the one person who happens to be reachable. A knowledge system does not replace that person's judgment. It makes what they already know — and what your inboxes and contracts already contain — findable by whoever needs it, at the moment they need it, instead of only by the person who remembers where it is. It's one specific, deep version of the broader case for AI in logistics operations: finding what your brokerage already knows, faster than a person can.
What the system does, day to day
The questions it answers are narrow and specific, because that is how the work actually comes up.
- Customer-specific requirements. A coordinator asks what a shipper's receiving hours are, what their appointment scheduling rules require, or what they will and will not accept on a bill of lading — and gets the answer pulled from the account notes, the onboarding email, or the last three exception threads where the rule actually came up.
- Carrier history on a specific lane. Before tendering a load, a dispatcher asks which carriers have actually performed well on this lane — not just which ones are compliant on paper — and gets a history built from past load notes, on-time records, and the emails where a coordinator flagged a problem the compliance system never saw.
- Accessorial and detention terms. A billing question or a carrier dispute comes up, and instead of someone re-reading a signed contract from scratch, the system finds the specific clause — free time, the detention rate, what counts as a qualifying accessorial — and returns it with the contract and page it came from.
- Customs and documentation requirements. A cross-border load needs a specific set of papers for a specific commodity and country pair, and the system surfaces what was required the last time your team moved something similar, rather than a coordinator reconstructing it from memory under a deadline.
- Dispatcher SOPs, especially for the exceptions that barely ever come up. The everyday routine, most dispatchers already know cold. It's the load that only happens twice a year — an unusual accessorial, a rare commodity, a shipper with a one-off rule — where a written SOP or a past resolved case actually earns its keep.
Three retrievals that used to depend on one person being at their desk
Here is what that looks like on an actual shift.
A new hire, the night shift, and a shipper's receiving rules
It's 9pm and a load is booked for a shipper the new coordinator has never worked before. The account rep who set up that relationship left for the day hours ago. The coordinator asks the system what this shipper's receiving hours and appointment rules are. It searches the account's email history and finds the onboarding thread from a year earlier, plus two more recent emails where the shipper's dock manager corrected a driver who showed up without pre-scheduling. The answer comes back with the receiving window, the appointment requirement, and a note that this shipper rejects any BOL missing a PO number — each point linked to the specific email it came from. The coordinator confirms the appointment correctly on the first try, instead of guessing and finding out from an angry phone call the next morning.
Sourcing coverage at 2am, with no one around who remembers which carrier is actually good on this lane
A load needs coverage overnight and the dispatcher on duty has covered this lane maybe twice before. She asks the system which carriers have actually performed well on it. It pulls together the record — three carriers with recent loads on that lane, one with a pattern of late check-ins noted in past dispatcher emails despite a clean compliance file, another whose contract has a detention rate well below what this lane typically runs into given how that shipper's dock tends to run behind. The dispatcher tenders the load to the carrier with the better real history and the better detention terms, a decision the senior dispatcher who normally handles that lane would have made from memory — except tonight, nobody had to wait for her to be reachable.
A cross-border load with paperwork nobody has needed since last year
A load crosses into Canada carrying a commodity your brokerage rarely handles on that lane, and the coordinator working it has never filed the paperwork for this specific combination. She asks the system what documentation this move needs. It finds the customs broker's email from the one prior shipment of the same commodity across the same border, along with the internal note a dispatcher wrote afterward about a form that was nearly filed wrong. The system returns both, cited, and flags that it found exactly one prior instance — not a pattern confident enough to treat as a settled rule. The coordinator gets the paperwork right the first time on a move that comes up rarely enough that nobody keeps it memorized, and files it correctly instead of guessing and finding out at the border.
Where the value concentrates, and where this doesn't help
Be clear-eyed about who this is actually for. On a normal weekday shift, with the senior dispatcher at her desk and the account rep one Slack message away, most of this knowledge is already reachable — someone just asks. The value of a system like this shows up specifically on the night shift, on a new hire's first solo week, and on the exceptions that come up too rarely for anyone to keep fresh in memory. If your brokerage is small enough that everyone already knows everything everyone else knows, this will not change much. If your team has grown past the point where one person can hold it all, or if your best dispatcher is the only reason certain lanes run smoothly, that is exactly the gap this closes.
Every answer the system gives cites the specific email, contract, or past case it came from, so a coordinator can check it rather than take the system's word for it. Where the record is thin — one prior shipment, a note that was never confirmed — the system says so rather than presenting a single data point as a settled fact. And we do not invent numbers to make this sound more finished than it is: no fabricated statistic about how many hours this saves your team, because the honest answer depends entirely on how much of your operation's knowledge currently lives in one person's head versus how much is already written down.
Connecting to your inbox, contracts, and TMS
This runs on the same foundation as Calfy's knowledge systems work generally, pointed at the specific places a brokerage keeps what it knows.
Connection. We index the shared inbox, contract repository, carrier scorecards or spreadsheets, and whatever notes your TMS already holds against a load or account. Nothing needs to move into a new system first — the system searches what you already have.
Permissions. A dispatcher sees what a dispatcher could already see; a carrier's rate history stays visible to the people who already had access to it. The system mirrors your existing access rules rather than opening everything to everyone with a login.
Documents, not just email. Contracts, signed rate confirmations, and scanned customs paperwork are where a lot of this knowledge actually sits, and getting text out of a PDF or a scanned form reliably is its own discipline — covered in more depth on the AI document review page.
Retrieval, not invention. The underlying approach — search first, answer from what was actually found, cite the source — is the same retrieval-augmented generation pattern behind any properly grounded system, and the knowledge base search use case covers the general shape of it.
Where it sits next to an agent. This system answers questions. It does not tender a load, source a carrier, or send a quote. Where you also want something that reads an RFQ and acts on it inside your TMS, that is the AI agents for logistics work — and the two are built to run on the same indexed knowledge, so an agent's decisions and a dispatcher's manual lookups draw from the same source rather than two different truths.