Cowtown Logistics × First Article Systems · Fort Worth · October 2026
So don't apply AI.
Re-engineer the desk.
AI agents inside the systems Cowtown already runs: McLeod, Gmail, the customer portals, Chain. The agent does the clerical work. A broker still decides.
One load, tender to cash · each step sorted into one of four buckets
Delete Plain code Agent Human decision
Presented by
Patrick Barbour, First Article Systems
Prepared for
Cowtown Logistics leadership
This is already happening. Brokers are rebuilding the desk.
Brokerages that run on people are putting agents inside the TMS and rebuilding the work. The numbers that move: shipments per person, and how much of the day goes to selling.
1

A brokerage runs on people.

Reps read tenders, key orders, answer "is it still available," chase PODs and upload invoices. One operator running 50 to 70 loads faced 400+ emails a day and 20+ portals.

Augment case · vendor claim
2

Put agents inside the TMS.

C.H. Robinson runs 30+ agents: quoting, email order build, appointments, tracking. Its orders agent builds 5,500 truckload orders a day. McLeod opened an AI Order Creation interface for certified providers in Nov 2024.

C.H. Robinson · company claim · McLeod press release
3

The numbers move.

C.H. Robinson: 35%+, later 40%, more shipments per person per day since the end of 2022. Quote coverage from about 65% of requests to 100%, in about 30 seconds.

C.H. Robinson newsroom, UBS 2025 · company claims
4

Each rep is worth more.

More loads per rep at the same margin is more gross profit per head. The goal is not fewer brokers. It is each broker selling most of the day, not a third of it.

Our model, slide 11 · estimate
The big brokers did it with 30+ agents and in-house engineers. A 2,000-load desk needs the right three first.
Your desk is already complicated.
About 2,000 loads a month. Eight places the freight lives. Each one used a different way.
Loads
~2,000
a month, W-2 desk
Billing
~$1M
a week, direct
Authority
MC-900645
brokerage since 2014
Built for
Project freight
flatbed, open deck, OD
1System of record
McLeod TMSorders, loads, billing
2Carrier checks
Highwayidentity, fraud
Carrier411monitoring · likely
3Tracking, load boards
ChainAI track and trace
DATposting, covering
Truckstopposting, covering
2Outside the systems
Gmail, Drive, Chattenders, PODs, files
Customer portalsbilling uploads
The hard part
Demand is not the limit. Desk hours are.
"We cannot grow with our current customers until we add another Broker to help cover the freight we already have."
Colby Baskin, CEO · LinkedIn, Sep 10, 2026
if a tender email lands → read it, key it into McLeod
if a carrier calls → "still available? what's it pay?"
if a load delivers → chase the POD, upload to a portal
if a carrier looks off → Highway, Carrier411, a phone call
Open roles, Oct 2026: Track & Trace · Billing & AR · Carrier Sales · Broker
Sources: Colby Baskin, LinkedIn article Apr 2026 and post Sep 10, 2026 · FMCSA SAFER · cowtownlogistics.com · stack confirmed by Patrick Barbour, Oct 2026 (Carrier411 likely)
Understand a company's process.
Interviews get the tribal knowledge. Process-mining agents get the real data and the documents. This is how we would map Cowtown's desk.
1 · Interviews

The human side.

Carrier repsAccount repsFull-cycle brokersBilling & ARTrack & trace
Why a customer rule existsWho really prices ODWhich step is theatreThe workaroundThe unwritten margin floor
What they give you

The why behind every number. Shadow 2 brokers, 1 carrier rep, billing and tracking, at the desk.

2A · Mining the system of record

What McLeod knows.

McLeod TMSChain trackingDAT, Truckstop
Loads per stepTender to booked timeQuote response timeException rateDays delivery to invoiceRebills
What the agents give you

Every load, every timestamp, every hand-off. 30 days of McLeod data. All of the volume, none of the memory.

2B · Mining everything else

What lives outside it.

GmailGoogle DriveGoogle ChatCustomer portalsHighwayCarrier411
Tender and quote emailsRate-con PDFsThe sheet in DriveCustomer rulesExceptions settled in Google ChatUnbilled accessorials
What the agents give you

The rules nobody typed into McLeod, and where the documented process drifted. 30 days of email.

Interviews give the why. Mining gives the what. Together they give the map: every step, its volume, its wait, its owner.
Example 1. Tender to booked load.
A flatbed tender, from the inbox to a booked truck. The SOP says 7 steps. The inbox says 18.
Typical broker flow · to be validated in the assessment · not Cowtown data
SOP → typical
7→18
Loads
~2,000
a month, Cowtown
Modes
Flatbed
OD, project, van, reefer
System of record
McLeod
plus Gmail and portals
Open role
Broker
"to cover the freight we have"
What makes it hard
Project freight
Dims, tarps and permits must be exact. Tenders arrive as emails, PDFs and spreadsheets.
What the SOP says · 7 steps
1Receive tender
2Enter order
3Price, confirm
4Post load
5Book carrier
6Send rate con
7Dispatch
What the inbox shows · 18 steps · 3 loops back
In the SOPNot in the SOPA step done againA loop back
Sort every step. Delete, code, agent, human.
Eighteen typical steps, one bucket each. Four are deleted, not automated. Agents read, draft and chase. A broker prices, books, and decides on every flag.
Typical broker flow · to be validated in the assessment
Typical → after
18→14
Tender to booked load · 18 steps, in order
Delete
4 steps · rework and repeats
Plain code
5 steps · rules, no judgement
6 agents
Read, draft, chase, then stop
3 human decisions
Price, book, decide on flags
Delete Plain code Agent Human decision
Four steps gone. Six handed to agents. Three stay with a broker, the ones that make the money.
The agent works McLeod. A broker decides.
A tender arrives. The agent reads it, drafts the order in McLeod, checks the lane, and stops. A broker approves.
1Tender arrives
2Extract fields
3Draft order in McLeod
4Lane and margin check
5Ask a broker
6Broker approves
Live demo · illustrative, sample data
Gmail · tenders@
Gmail-style mock
Sample Shipper Co.Tender: FTW to Lake Charles, flatbed9:12
Carrier dispatchPOD for load SAMPLE-09988:47
Sample Mfg.Rate request: Dallas to OKC8:31
Dana K. <dispatch@sample-shipper.example>
Subject: Tender: FTW to Lake Charles, Thu pickup
Hi team, need a 48' flatbed. Pick up Thu 10/8, 7:00 AM at our yard, Fort Worth TX 76106. Deliver Fri 10/9 by 2:00 PM to the jobsite in Lake Charles LA 70601. 24 steel beams, 42,000 lbs, 8'6" wide, tarps required. $2,150 all-in. PO 55120.
Thanks, Dana
PDFPO-55120.pdf
Agent read · 9 fields · 0 missing
OriginFort Worth TXDestLake Charles LAPU10/8 07:00Del10/9 by 14:00Equip48' flatbedWeight42,000 lbRate$2,150TarpsYesPO55120
Order SAMPLE-1042 released in McLeod
TMS · Order Entry
McLeod-style mock
OrderStopsChargesCarrierDocs
OrderDraft · not releasedReleased · available
(draft)SAMPLE-1042
Stops
#TypeLocationAppointment
1Pickup
2Delivery
Lane and margin check · agentSample history
FTW → Lake Charles · last 6 flatbed loads, carrier pay
$1,650 to $1,820 · median $1,740
Margin floor (sample rule)15%
Max carrier buy$1,827
Margin at median buy$410 · 19%
Tarps requiredNo permit, 8'6"Rule: call 1 hr before delivery
Google Chat
Chat-style mock
Spaces
flatbed-desk
carrier-sales
billing
tracking
Desk Agent
flatbed-desk space
JP
Jordan P.9:05 AM
Flatbed board is tight this morning. Price with care.
A
Desk AgentAPP9:13 AM
@Sam R. draft order is ready in McLeod. It is yours to approve.
LaneFort Worth TX → Lake Charles LAPickupThu 10/8, 07:00Equipment48' flatbed, tarpsCustomer rate$2,150 all-inMax buy$1,827 at 15% floorLane median$1,740, 6 loads
ApproveEdit in McLeod
✓ Approved by Sam R., 9:14 AM. Order released in McLeod.
POD to cash. Billing decides the exceptions.
The POD arrives. The agent matches it, checks accessorials, builds the invoice packet, and stops on the exception.
1POD arrives
2Match to rate con
3Accessorial check
4Ask billing
5Billing approves
6Portal upload, AR queued
Live demo · illustrative, sample data
Gmail · billing@
Gmail-style mock
Sample Carrier Co.POD: SAMPLE-1042 delivered10:52
Bill of lading · POD
Pcs24
Delivered
10/09 10:38
Delivered this morning, signed POD attached. Please process.
Dispatch, Sample Carrier Co. (sample)
PDFPOD-1042.pdf
Customer AP portal
Sample portal
INV-SAMPLE-1042.pdf
✓
POD-1042.pdf
✓
RateCon-1042.pdf
✓
Detention-evidence.pdf
✓
Submitted 10/09, 11:02 AM
AR follow-up queued: Nov 9 if unpaid (net 30, sample terms)
Uploaded to customer portal
TMS · Billing · Load SAMPLE-1042
McLeod-style mock
OrderStopsBillingCarrier payDocs
Document match · agent6 of 6 match
FieldPODRate conOrder
Load #10421042SAMPLE-1042✓
ConsigneeJobsite, Lake CharlesLake Charles LALake Charles LA✓
Pieces242424✓
Delivered10/09 10:38by 10/09 14:00by 10/09 14:00✓
SignaturePresent✓
CarrierSample Carrier Co.Sample Carrier Co.Sample Carrier Co.✓
Accessorial check · agentException
Free time 2:00
+1:39
Arrived 07:0209:02Departed 10:41
Tracking timestamps · customer rule (sample): $75/hr after 2 hrs · proposed detention $123.75
Invoice packetReady to bill
Linehaul$2,150.00
Detention, approved$123.75
Total$2,273.75
Packet: invoice, POD, rate con, tracking timestamps
Google Chat
Chat-style mock
Spaces
flatbed-desk
carrier-sales
billing
tracking
billing space
MT
Morgan T.10:20 AM
Two rebills still open. On them.
A
Billing AgentAPP10:53 AM
SAMPLE-1042 POD matched. One exception before I bill.
ExceptionDetention at consigneeEvidenceIn 07:02, out 10:41 (tracking)Over free time1 hr 39 minRule$75/hr after 2 hrs (sample)Proposed$123.75
Approve and billEdit
✓ Approved by Morgan T., 10:58 AM. Packet built.
A
Billing AgentAPP11:02 AM
Uploaded to the customer portal. AR follow-up queued for Nov 9.
Vet at booking. A broker decides on flags.
A carrier is picked. The agent runs the checks, writes a dated record to the load, and never clears a flag on its own.
1Carrier picked
2Run checks
3Flag found, stop
4Ask the broker
5Broker decides
6Dated record on the load
Live demo · illustrative, sample data
TMS · Carrier assignment
McLeod-style mock
Load SAMPLE-1043FTW → Houston · 48' flatbed
Sample Carrier Co. (sample)
MC-SAMPLE · offered $1,150 · called in from (817) 555-0142
Rate confirmationHeld until checks clearReleased
Why it matters now

Montgomery v. Caribe Transport II (Supreme Court, May 2026) exposes brokers to state negligent-selection claims. Some broker liability renewals are up double and triple digits.

TIA: 1,600+ fraud reports, Sept 2024 to Feb 2025, up 65%.

Transport Topics, Aug 31, 2026 · DC Velocity on TIA report
Carrier checks · agent
Highway · Carrier411 · FMCSA
✓
Authority activeCarrier authority active, not out of service
FMCSA
✓
Insurance on fileAuto liability and cargo certificates current, monitored
Carrier411
✓
Identity verifiedCarrier identity check passed
Highway
✓
Equipment fitFlatbed on profile, tarps listed
Packet
✓
History with us3 prior loads, no claims
McLeod
✓
Email domainMatches the domain on the carrier packet
Packet
!
Contact mismatchCaller phone (817) 555-0142 is not on the carrier's verified contact list
Flag
1 flag. The agent stops. Rate con held.
Reasonable-care record · written to load SAMPLE-1043
Oct 6, 2026, 2:43 PM CT · 7 checks run · 1 flag (contact mismatch) · Decision: approved after callback to the verified number, by Sam R. · Evidence: FMCSA snapshot, insurance certificates, identity report, call note.
Record saved to the load in McLeod
Google Chat
Chat-style mock
Spaces
flatbed-desk
carrier-sales
billing
tracking
carrier-sales space
SR
Sam R.2:31 PM
Booking Sample Carrier on 1043 at $1,150 if they check out.
A
Carrier AgentAPP2:36 PM
@Sam R. one flag on SAMPLE-1043. Rate con is held. Your call.
CarrierSample Carrier Co. (sample)FlagCaller phone not on verified contactsPassed6 of 7 checksSuggestCall the verified number on file before release
Approve after callReject
✓ Approved by Sam R. after callback, 2:42 PM. Rate con released.
The agent never clears a flag on its own.
Eight workflows. Four buckets.
Eight agent workflows for Cowtown's desk, each with its bucket mix and the point where a person decides.
01Phase 1

Tender and order intake

Reads tender emails, PDFs and sheets. Drafts the order in McLeod with stops, weights, equipment and customer rules.

A person decidesBroker confirms the order before it goes live.
02Phase 2

Quote intake and price draft

Pulls lane history and market rate, drafts a quote inside margin rules, flags OD, permits and tarps.

A person decidesAny price outside the rules. OD and heavy haul stay human.
03Phase 3

Carrier comms and outreach

Answers "still available, what's it pay" on DAT and Truckstop loads, captures offers, sends targeted offers from history.

A person decidesBroker negotiates and books.
04Phase 3

Tracking exceptions

Extend Chain, the AI tracking Cowtown already runs. Do not rebuild it. Exceptions into McLeod, customer updates.

A person decidesOps handles the exceptions, only the exceptions.
05Phase 2

Carrier vetting evidence log

Highway, Carrier411 and FMCSA checks at booking. A dated reasonable-care record on every load.

A person decidesEvery flag. No auto-override, ever.
06Phase 1

POD-to-cash

POD matched to rate con and order, accessorials checked, invoice packet built, portal upload, AR follow-up.

A person decidesBilling approves discrepancies and rebills.
07Phase 1

Cowtown GPT

SOPs, customer rules, McLeod how-to and lane history, answered in Google Chat or Gmail. Powers the other agents' rules.

A person decidesContent owners approve every source.
08Phase 2 and 3

Detention, accessorials, scheduling

Detects dwell from tracking timestamps and drafts the bill. Books and moves appointments by email or dock portal.

A person decidesAccount manager approves the bill. Ops on conflicts.
Delete Plain code Agent Human decision Bucket mix is a planning view. The assessment sets the real one.
One broker. Five more loads a day.
Likely case. Agents take half to two thirds of the clerical work. 60% of the freed time goes to more loads. Margin per load held flat.
Estimates · not Cowtown data · the assessment measures the real baseline
Loads per rep per day
15→20.5
GP per rep per month
$94.5k→$128.9k
Share of day selling
~33%→~61%
Clerical hours freed a day
0→3.4
Scenarios · GP per rep per month
Baseline
$94.5k
Worst
$107.7k+14%
Likely
$128.9k+36%
Best
$167.0k+77%
Sanity checks

Baseline is real for this desk. $94.5k sits just under Cowtown's own posts that W-2 reps each do "$100K+ in GP a month."

Likely is not a vendor number. +36% sits at C.H. Robinson's disclosed 35 to 40% shipments-per-person gain, and well below vendor "2x" claims.

Freight is there. One Cowtown team "cannot grow with our current customers until we add another Broker."

Assumptions: 9-hour day · 15 loads per rep per day baseline · 21 working days · $300 GP per load (TIA $1,989 revenue x 14.5 to 17.8%) · clerical work 360 min a day · agents absorb 25 to 35% worst, 50 to 70% likely, 70 to 90% best · 60% of freed time goes to more loads · margin flat.
Realized loads = 15 + 0.6 x (freed minutes / new minutes per load). GP = loads x 21 x $300. Estimates, not a promise.
Same method. Three phases.
Nine to twelve months. Every phase runs the same loop: map the steps, sort them, build inside the tools you run, measure against the baseline.
M1M2M3M4M5M6M7M8M9M10M11M12
StartWeeks 0 to 3
Current State Assessment. $3,500, credited toward the build. Shadow the desk, pull 30 days of McLeod and email data, set the baseline.
Phase 1Months 1 to 4 · the wedge
Phase 1
Cowtown GPT SOPs, customer rules, lane historyInbox-to-McLeod tenders and quotes drafted in McLeodPOD-to-Cash match, invoice, portal, AR
Gate: confirm McLeod access path in the assessment, API or certified partner.
Phase 2Months 4 to 8
Phase 2
Quote pricing drafts inside margin rulesCarrier vetting log on top of HighwayDetention and accessorials
Phase 3Months 8 to 12
Phase 3
Carrier voice DAT, Truckstop loadsTracking exceptions extend ChainScheduling appointments
Shadow mode, then broker-approved, then automatic on proven rules. Weekly readout against the baseline.
Three people in one.
That is why this work is hard to hire for. All three workflows you just saw need all three skills.
1 · The operator

Knows how a desk runs.

Tender to cash, carrier sales, billing. Knows which step is theatre and who really decides. Brokerages have this person.

2 · The engineer

Ships into McLeod, Highway, portals.

Integrations, permissions, rollback, in the systems the desk already runs. Software firms have this person.

3 · The AI engineer

Agents, evals, guardrails.

Knows what a model can be trusted with, and where it has to stop and ask a broker. AI labs have this person.

1Knows how a desk runssales, carrier, billing, ops 2Ships into the TMSMcLeod, Highway, portals 3Builds the AI layeragents, evals, guardrails FDE First Article sits in the middle
Almost nobody has all three in one person. That is the seat First Article fills inside Cowtown.
Do this next week.
One thing a day, on Cowtown's own desk. Before any AI is involved.
Monday

List every system that holds freight data.

McLeod, Gmail and Drive, customer portals, Highway, Carrier411, Chain, DAT, Truckstop. Write which one wins when two disagree.

Tuesday

Pull 20 loads from last week.

Tender to paid. Count how many touched the inbox more than twice. Count the ones that went around a loop.

Wednesday

Walk one tender end to end.

From the email to the booked truck. Every click, every tab, every wait. Time the work against the waiting.

Thursday

Sort every step.

Delete it, give it to code, give it to an agent, or keep it with a broker. In that order. Most people start at the agent.

Friday

Pick the one step to remove first.

The smallest change that gives a broker time back this month. Then build that, inside McLeod and Gmail.

That is the assessment. We run it with your team in 2 to 3 weeks.
So don't apply AI. Re-engineer the desk.
Six steps. They work on a 2,000-load desk the same way they work on a public company.
01

Find the real process.

Interviews for the why. Mining McLeod and Gmail for the what.

02

Measure elapsed time against touch time.

The waiting is the process. The keying is minutes.

03

Define one process with one owner.

Someone who can sign it across sales, carrier and billing. Nothing about this step is AI.

04

Sort every step.

Delete, plain code, an agent, a broker. In that order.

05

Baseline before you build.

Loads per rep, quote response, days to invoice. Write the number on the wall.

06

Build inside the desk. Measure again.

McLeod, Gmail, Google Chat. No new dashboard. The baseline is the only proof it worked.

Next step
Book the Current State Assessment
$3,500 · 2 to 3 weeks · credited toward the build
With
Patrick Barbour
First Article Systems · firstarticlesystems.com
20-minute call
calendly.com/patrick-firstarticlesystems/discovery-call-20-min
Thank you.
Appendix. Assumptions and sources.
Every number traces to a public source, a labeled claim, or a stated estimate.
Output model · estimates
Workday9 hours (540 min)
Baseline loads per rep per day15 (low end of practitioner 15 to 20 range)
Working days per month21
GP per load$300 (TIA Q2 2026 $1,989 revenue x 14.5 to 17.8% margin)
Selling time12 min per load, about one third of the day
Clerical time360 min a day: carrier comms 150, tracking 90, order entry and quote prep 60, vetting and rate con 30, docs and billing 30
Share agents absorbWorst 25 to 35% · likely 50 to 70% · best 70 to 90%
Freed time used for loads60% (40% to prospecting and accounts)
ResultsWorst 17.1 loads, $107.7k, +14% · likely 20.5, $128.9k, +36% · best 26.5, $167.0k, +77%
Demo slides

Slides 7 to 9 use invented sample shippers, carriers, loads, people and rules. Mock screens are generic, not vendor software. McLeod requires AI order-creation providers to be McLeod Certified; the access path is confirmed in the assessment.

Sources · accessed Oct 6, 2026
  • Stack: McLeod, Highway, Chain, Google Workspace, DAT, Truckstop per Patrick Barbour, Oct 2026; Carrier411 likely, to confirm. Cowtown pages also show MyCarrierPackets and Power BI.
  • Cowtown: cowtownlogistics.com (home, agents, carriers, careers) · FMCSA SAFER, USDOT 2579786 · Colby Baskin, "A Random Day in Logistics," LinkedIn, Apr 17, 2026 · Colby Baskin broker hiring post, Sep 10, 2026
  • Benchmarks: TIA Q2 2026 market data via Logistics Management, Aug 31, 2026 · TIA Q3 2024 Market Report · practitioner LinkedIn thread on loads per rep (anecdote)
  • C.H. Robinson (company claims): newsroom, AI leadership metrics and 30+ agents · UBS 2025 conference transcript
  • McLeod: AI Order Creation interface press release, Nov 2024; certified partners
  • Risk: Transport Topics on Montgomery v. Caribe Transport II, Aug 31, 2026 · DC Velocity on TIA fraud report · Verisk CargoNet 2025 · FreightWaves capacity, Oct 6, 2026
  • Vendor claims: Augment (Armstrong), Parade, HappyRobot, Vooma, Chain. Not independent results.