Maria Tkacheva

B2B SaaS · AI · Vehicle Transportation

AI-Assisted Load Discovery for Car Haulers

Building a 0→1 product that helps independent car haulers identify, evaluate, and select transportation opportunities that fit their trucks and planned routes

Driver Load Network

Load requests

Truck 12 · Chicago, IL → Indianapolis, IN via I-65

Active request

2 potential matches

Truck 12 · Open carrier

2 of 9 available

Chicago, IL → Indianapolis, IN

Leaving today 11:00 AM

SouthEast
I-65MerrillvilleLafayetteChampaignChicagoIndianapolisHeading south, east ↘

Offers for this request

RoutePay
Merrillville → IndianapolisPotential match$480
Lafayette → CarmelPotential match$310
Champaign → IndianapolisNeeds review$560

1 more offer not matched: needs 3 spaces

TL;DR
01. The Problem

Finding loads was possible. Finding the right ones was the hard part

Independent car haulers often spend hours comparing listings, routes, vehicle requirements, and broker information to fill available trailer space. A load may appear profitable but require more capacity than the driver has, involve an expensive detour, or conflict with pickup schedules. This makes finding suitable opportunities time-consuming and can leave truck capacity unused.

Truck 12 and three of its four offers

Truck 12

Open carrier

2 of 9 spaces available

Chicago → Indianapolis

Offer A

2 sedans

Pickup in Merrillville

Space2 of 2
RouteOn the route

Potential match

Offer B

3 sedans

Pickup in Gary

SpaceNeeds 3
RouteOn the route

Not matched

Offer C

2 sedans

Pickup in Champaign

Space2 of 2
Route58 mi off

Needs review

Challenge 01

Operational fit

Can this driver transport the load efficiently?

Challenge 02

Commercial access

Will the broker assign the load to this driver?

The opportunity was to reduce the time drivers spent evaluating loads before deciding which brokers to contact

02. The Insight That Changed My Direction

What Driver Research Revealed About Load Selection

My initial assumption was that independent car haulers needed a more convenient way to discover available transportation loads. But conversations with drivers and transportation experts revealed that their main challenge was deciding which opportunities fit the trips they were already operating.

  1. Finding 01

    Availability changes throughout the trip

    Truck specifications stay stable, but open spaces change as vehicles are loaded and unloaded.

    Design implication

    Matching needs an active request with remaining spaces, not just a saved truck profile

  2. Finding 02

    A good rate can hide an inefficient trip

    A higher-paying offer may mean extra driving, an awkward pickup, or a route against the planned direction.

    Design implication

    Matching has to weigh geography and preferences alongside vehicle compatibility

  3. Finding 03

    A match isn’t a booking

    Drivers could find a suitable load and still lose it to a carrier the broker preferred.

    Design implication

    Recommendations identify promising opportunities and never imply a load is secured

Initial direction

Make more available loads easier to discover

A searchable listing experience

Revised direction

Identify which loads are worth pursuing for a specific truck and trip

A decision-support workflow

03. Product Strategy

Defining Where AI Could Create Value

AI offered an opportunity to reduce manual load evaluation, but I first needed to understand where it could add meaningful value. The goal wasn’t to automate the entire transportation process. It was to determine whether AI could help drivers identify more suitable opportunities using their current truck capacity, availability, and planned route.

Product hypothesis

If AI could evaluate a driver’s available capacity, current location, and planned route against transportation offers, drivers could identify suitable loads faster, fill more available trailer space, and improve their earning opportunities.

Evaluating three product directions

Direction A

Searchable load board

What I evaluated

Could improving search and filtering reduce the work required to find relevant loads?

Tradeoff

Drivers would still evaluate route fit and current truck availability themselves.

Direction B

Full transportation marketplace

What I evaluated

Would connecting the entire transportation process create greater value than improving discovery alone?

Tradeoff

A much broader product requiring more operational processes and engineering effort.

Direction C

Selected direction

AI-assisted discovery and evaluation

What I evaluated

Could AI reduce manual screening by considering the driver’s truck, available space, route, and offer requirements together?

Why I selected it

It directly addressed the research findings while allowing the team to validate a focused product before expanding into other transportation workflows.

To test this direction, I focused the first version on the minimum workflow needed to connect a driver’s availability with relevant offers.

What we built first

  • Structured availability

    Open spaces, location and route for the current trip

  • AI-assisted matching

    Offers checked against the truck and the trip

  • Offer evaluation

    Rate, route and fit compared before contacting a broker

Deliberately deferred

  • Automated booking

    Brokers still decide who gets the load

  • Payment processing

    Not needed to test the hypothesis

  • Shipment tracking

    Only matters once a load is booked

  • Broker negotiations

    Stay between drivers and brokers

04. Three Decisions That Shaped the Experience

Decision 01

Capture the truck’s actual availability

A saved truck profile couldn’t represent changing availability. I separated permanent equipment information from current-trip details, allowing drivers to reuse their trailer profile while updating available spaces, location, and route preferences.

New load request

Draft

Trailer (saved)

Truck 12 · Open carrier · 9 spaces

Change

Annotation: Saved trailer information. Picked once and reused, since equipment rarely changes

Available spaces

−2+
of 9

Annotation: Remaining capacity. Two open spaces, not the nine-car total

Current location

Chicago, IL

Available from

Today 11:00 AM

Heading to

Indianapolis, IN

Preferred directions

NorthSouthEastWest

Additional constraints (optional)

Anything dispatchers should know

Annotation: Route preferences. Directions let matching include pickups along the way

Review request
Driver Load Network

New load request

Tell dispatchers what your truck can take, once

Trailer
Truck 12 · Open carrier9 spaces · savedTruck 7 · Enclosed2 spaces · saved
Available spaces
−2+2 of 9 available
Current locationChicago, IL
Available fromToday 11:00 AM
Heading toIndianapolis, IN
Preferred directions
NorthSouthEastWest

Dispatchers will see

  • Truck 12 · Open carrier
  • 2 of 9 spaces available
  • Chicago, IL · Today 11:00 AM
  • Heading to Indianapolis, IN · South, East
Review request

Design tradeoff

More fields would sharpen matching but slow every request. I kept what an initial evaluation needs and made other constraints optional.

Decision 02

Make recommendations understandable

A numerical match score wouldn’t explain whether an offer was compatible or simply required further evaluation. I designed three clear outcomes: Potential match, Needs review, and Not matched, each supported by the relevant transportation constraints.

Driver Load Network
REQ-218 · Truck 122 of 9 availableChicago → IndianapolisSouth, East4 offers

Design tradeoff

A score is easier to sort and display, but harder to act on. Three states give up fine-grained ranking for meaning.

Decision 03

Keep drivers in control

A higher-paying offer isn’t always the better choice if it requires a substantial detour. I brought payment, route deviation, vehicle requirements, and pickup timing into one comparison, allowing drivers to decide which opportunity to select without automatically booking transportation.

Driver Load Network
REQ-218 · Truck 122 of 9 availableChicago → IndianapolisSouth, East4 offers
Available spaces
2

Design tradeoff

Auto-selecting the top-ranked offer would remove a step, along with the context and control a consequential decision needs. AI stays a decision-support layer, not an autonomous dispatcher.

05. AI-Assisted Matching

Finding loads that fit the truck, space, and route

Transportation offers often include requirements buried in broker notes or inconsistent listing fields. AI helped turn that information into structured requirements that could be evaluated against a driver’s truck, availability, and planned route.

Broker note

“Vehicle doesn’t run. Winch required. Pickup before 3 PM.”

Extracted into the schema

Vehicle condition
InoperableSource: “Vehicle doesn’t run”
Required equipment
WinchSource: “Winch required”
Pickup deadline
3:00 PMSource: “Pickup before 3 PM”
Vehicle dimensions
UnknownNot stated, so left unknown

Designing the matching system

  1. 01

    Understand the offer

    An LLM extracts requirements from broker notes into structured fields, preserving the original source and leaving unknown information unresolved.

  2. 02

    Evaluate compatibility

    Rules check capacity, equipment, and pickup constraints before an offer can qualify as a potential match.

  3. 03

    Evaluate route relevance

    Routing and ranking assess how each eligible offer fits the driver’s location, schedule, and planned direction.

Each offer is evaluated on its own against the driver’s request

Driver context + offer information

Truck

Open carrier · 2 of 9 available

Route

I-65MerrillvilleLafayetteChampaignChicagoIndianapolisHeading south, east ↘

Offer

DL-3107

Merrillville, IN → Indianapolis, IN

2 sedans · Today 1:00–4:00 PM · $480

Broker note“Both sedans run and drive. Pickup 1–4 PM at the dealer lot.”

Matching evaluation

AI extraction

Language model

  • Vehicle condition Operable
  • Required equipment None stated
  • Pickup window 1:00–4:00 PM

Constraint checks

Rules engine

  • Spaces 2 needed · 2 available
  • Equipment Open carrier, no winch needed
  • Pickup Within availability from 11:00 AM

Route relevance

Routing and ranking

  • Route Merrillville is on I-65
  • Direction Heading south, east
  • Delivery Ends in Indianapolis

One outcome for this offer

Potential match

DL-3107 · this offer

Fits the open spaces and equipment, with pickup on the planned route

Other offers, evaluated the same way

  • Not matched DL-3115
    A hard constraint: needs 3 spaces, 2 available

  • Needs review DL-3123
    Fits the trailer, but pickup adds 58 mi

Designing for uncertainty

Missing information

The result names what the driver should confirm instead of assuming compatibility

Conflicting information

The offer moves to Needs review instead of receiving an overconfident recommendation

Outdated information

A match is reevaluated when availability or offer details change

Unavailable AI processing

Drivers can still inspect structured offer details and the compatibility checks

Evaluating matching quality

I kept AI quality separate from product success. Representative transportation scenarios test critical incompatibilities before release; driver behavior then shows whether valid recommendations are useful.

Compatibility correctness
Surfaced offers that violate a hard constraint
Extraction accuracy
Fields the model gets wrong or misses in broker notes
Explanation grounding
Reasons shown to drivers that aren’t supported by request or offer data
Recommendation relevance
Surfaced offers drivers consider worth reviewing
Workflow efficiency
Time to identify a promising opportunity
06. Outcomes

From manual screening to AI-assisted load evaluation

~56%

Reported reduction in workflow-completion time during beta testing

The experience brought together information drivers previously had to compare across multiple steps.

What Was Delivered

01

A connected driver workflow

Availability, matching, and evaluation in one experience.

02

AI-assisted recommendations

Relevant offers surfaced using structured transportation requirements and route context.

03

Driver-controlled decisions

Drivers evaluate tradeoffs and select opportunities without automatic booking.

What I Would Measure Next

Time to a viable opportunity
How quickly drivers identify a load they want to pursue
Recommendation acceptance
How often drivers consider suggested opportunities relevant
Contact-to-booking conversion
Whether better discovery contributes to more secured loads