Solution

AI agents that answer the rate request before your competitor does

AI agents for logistics read the rate request that just landed, pull the lane, equipment, and weight out of an email nobody structured for you, and get a number back to the shipper while a coordinator is still on another call. For a freight brokerage or 3PL, that speed is most of the business — the broker who answers first usually wins the load, and everything after that is check calls, paperwork, and exceptions. Calfy builds agents that carry that work end to end, wired into the TMS and inbox you already run.

· Reviewed by Artur Horimoto, Founder & CEO

The job is speed, and speed is the bottleneck

A brokerage does not lose loads because its rates are wrong. It loses them because the reply went out twenty minutes after a competitor's, or never went out at all because the RFQ arrived at 4:50pm buried under sixty other emails. Inbound rate requests show up as free text — a shipper's own format, a broker network's template, a forwarded EDI 204 nobody re-keyed — and someone has to read each one, figure out the lane, the equipment type, the pickup window, and get a defensible number back before the shipper moves on to the next name on their list.

That same coordinator is also running check calls, chasing a bill of lading that never showed up, and explaining to a client why a load looks stuck on the map. None of those tasks is individually hard. At the volume a working brokerage runs — dozens of RFQs, dozens of loads in transit, a constant stream of carrier calls — they compete for the same hours, and the RFQ usually loses, because it looks less urgent than the phone ringing about a load that is already late.

An agent does not experience that competition. It reads every inbound message the moment it lands, in parallel, and moves each one forward without waiting for the last email to finish.

What the system does day to day

The agent sits across the surfaces where the work actually happens — the shared inbox, the TMS, the carrier network — rather than replacing any of them.

  • Reads inbound rate requests out of email, whatever the format, and turns the unstructured text into a structured quote request: origin, destination, equipment, weight, commodity, pickup date, any accessorials mentioned.
  • Sources carriers for a lane by reaching out across your network, tracking who has capacity and at what rate, and assembling the options rather than leaving a coordinator to work down a contact list one call at a time.
  • Runs check calls and status updates, pulling from a telematics or ELD feed where one exists and placing the call itself where it does not, then pushes the update to the shipper before they ask for it.
  • Handles documents — rate confirmations, bills of lading, proofs of delivery — matching each one to the right load number and flagging anything that does not add up.
  • Watches for exceptions: a load that has not moved in longer than the lane usually takes, a driver who missed a check-in window, a detention clock that just started at a shipper's dock.

Every one of those is a narrow, well-defined job. Stacked together, they are most of what keeps a brokerage's ops team busy, and almost none of it requires a human's judgment to execute — only to approve.

Walkthrough: an inbound RFQ becomes a quote

A shipper's logistics coordinator emails the brokerage's quote inbox: a paragraph of prose, a spot rate lane, a pickup date next Tuesday, a note that the freight is temperature-sensitive. No structured fields, no EDI wrapper.

The agent picks up the message within moments of it arriving. It parses the lane (origin ZIP, destination ZIP), the equipment requirement (reefer, implied by the temperature note), the weight and commodity where stated, and the pickup window. It checks that lane against your rate engine and recent carrier activity on similar freight, and drafts a rate.

If the number falls inside your standard pricing logic and margin floor, the agent sends the quote back in the shipper's own thread, in the tone and format your brokerage uses, inside the same window a strong coordinator would hit on their best day. If the lane is unusual — a rarely-quoted origin, a rate that would breach the margin floor, an accessorial that needs a human call — it drafts the quote anyway and routes it to a coordinator with the reasoning attached, so approval takes seconds instead of a cold start. Pricing the load and deciding what goes out the door always stays a person's call on anything outside the standard rules; the agent's job is to make sure that call gets made in minutes, not hours.

Walkthrough: sourcing a carrier before the pickup window closes

A load is booked and needs coverage. In most brokerages this means a coordinator working down a list of carrier contacts by phone or text, asking about capacity for a specific lane, noting who responds, and comparing rates by hand while the clock runs toward the pickup appointment.

The agent runs that outreach directly. It messages the carriers in your network who are compliant for the lane and equipment type, tracks who responds and at what rate, and checks each one against carrier history — on-time performance on that lane, any recent claims, insurance status. It surfaces the strongest two or three options with the reasoning next to each one, ready for a coordinator to confirm the pick with a single decision instead of building it from a blank page. Carriers who need a manual sign-off — new to your network, or flagged for a prior issue — are held out of the automated tender and routed to a person regardless of how strong their rate looks. Final carrier selection stays with the broker; the agent's job is to have the shortlist ready before the pickup window gets tight.

Walkthrough: a detention clock starts and nobody has to notice

A driver arrives at a shipper's dock for a scheduled pickup. Two hours later, the load still has not been loaded — the shipper's yard is backed up. In most operations, nobody notices until the driver calls dispatch to complain, hours after detention time started accruing.

The agent is watching the same load status the whole time. It knows the appointment time, the free time your contract allows, and the moment the driver checked in. When the clock passes the free-time threshold, it logs the detention start automatically, notifies the carrier and the shipper's contact with the timestamp, and begins building the record — arrival time, dock notes, any driver communication — that a detention claim needs later. If the load stays stuck long enough to put the delivery appointment at risk, the agent flags that as a separate exception and puts it in front of a coordinator, because a schedule call like that is a judgment call, not a status update. The agent's job here is noticing and documenting the moment it happens, not deciding how the dispute gets resolved.

What stays with a person

None of this replaces the coordinator or the account manager. Pricing decisions outside the standard rules, carrier selection where the shortlist genuinely does not have a clear winner, and any dispute involving money — a detention claim, a damaged-freight disagreement, a rate discrepancy a shipper pushes back on — go to a person, with the relevant history already assembled instead of a blank investigation. The agent's value is in how much of that investigation is already done by the time it reaches someone, not in making the call itself.

Integration notes: TMS, EDI, and the rest of the stack

An agent is only as useful as its reach into the systems a brokerage already runs, so this is where most of the engineering effort goes.

TMS and EDI. We connect to whatever TMS your operation runs, through its API where one is available, and through EDI transactions (204, 214, 210 and similar), a scheduled export, or an email-based workflow where a direct API is not. Freight operations commonly run a mix of a modern TMS alongside older systems or spreadsheets patched together over years, and that mix does not block the build — it just changes how the connection is made. We do not claim a formal partnership with any TMS vendor; we build the integration your specific stack needs.

Voice, where the work is voice. A large share of check calls and carrier capacity outreach in this industry still happens by phone. Where that is the bottleneck, the agent works through voice AI systems built to hold a real phone conversation rather than a chatbot that only understands text.

Documents. Rate confirmations, BOLs, and PODs typically arrive as PDFs, scans, or fax images across email. The agent reads them, matches each one to the correct load number by reference details rather than filename, and flags mismatches — a weight that does not match the BOL, a missing signature — before the document moves toward invoicing.

Permissions and logging. Agents get read access where reading is enough and scoped write access only where the job requires it — updating a load status, sending a quote reply — never a shared login with broad access. Every action the agent takes is logged, so a rate quoted three months ago can be traced back to exactly what it saw and why.

Not every workflow needs a full agent. Some of the document routing and status-notification work in a brokerage is better served by straightforward workflow automation than by a system making judgment calls. Part of scoping the engagement honestly is telling you which pieces need an AI agent and which need something simpler. If logistics is one of several parts of your business where this applies, the logistics industry page covers the wider picture; the AI agent glossary entry explains the underlying mechanics in plain terms if you want that first.

Frequently asked questions

How fast can an agent actually respond to a rate request?

As fast as it can parse the email and check the lane against your rate rules — typically within moments of the message arriving, for lanes inside your standard pricing logic. Anything outside those rules routes to a coordinator with the draft rate and reasoning already attached, so the approval is quick rather than a cold start.

Does the agent decide the final price or pick the carrier?

No. It prepares the quote and the carrier shortlist against rules your team sets, and anything inside the standard pricing and carrier-approval logic can go out automatically. Anything outside those rules — an unusual lane, a new carrier, a margin exception — goes to a person to decide. Pricing and final carrier selection stay with the broker.

Can this connect to our TMS and EDI setup?

Almost always. We connect through your TMS's API where one exists, and through EDI transactions, scheduled exports, or an email-based workflow where it does not. Bring the systems you run — TMS, load boards, telematics — to the first call and we will tell you what is realistic before any commitment.

How does it catch exceptions like a stalled load or a detention clock?

It watches load status against expected timing — appointment windows, transit norms for the lane, free-time terms in your contracts — and flags the moment something falls outside that pattern: a load that has not moved, a driver who missed a check-in, a detention clock that just started. It documents the event and notifies the right person immediately rather than waiting for someone to ask.

What does this cost and how long does it take to go live?

It depends on how many systems the agent touches and how much of the quoting and exception logic needs to be encoded. A single-workflow build, such as an RFQ-to-quote agent connected to one TMS, sits at the lower end; a system spanning quoting, carrier sourcing, and exception handling sits higher. We scope every engagement and agree a clear price before any build work starts, and most brokerages see a narrow version live in weeks.

Bring the RFQ inbox, the check-call routine, or the document backlog that costs your team the most hours, and a free 30-minute strategy call is enough for us to tell you whether an agent fits it and roughly what building one would take.

See what this looks like for your business

Thirty minutes, your actual workflow, and a straight answer on whether this is worth building.

Free 30 minutes. No pitch deck. You leave with a plan either way.