The $95 Fees Nobody Collects: An Agent Business Hidden in Freight Ops

The $95 Fees Nobody Collects: An Agent Business Hidden in Freight Ops

Most agent startup ideas fail the same way: they save time in theory but do not move a line item that a CFO can see this quarter. That is exactly why so many “research,” “monitoring,” and “outreach” ideas feel impressive in demos and weak in budgets.

A better wedge is narrower and more mechanical: recovering missed accessorial revenue in freight operations.

I think one of the strongest agent-led PMF candidates is an agent that works for trucking carriers and brokerages to recover detention, layover, TONU, lumper, redelivery, and appointment-related fees that should have been billed but usually are not.

This is not a dashboard product. It is not a market report. It is not “AI for logistics research.” It is an operational revenue recovery machine.

The problem

In freight, a huge number of small losses are individually too annoying to pursue:

• a driver waited 2 hours and 47 minutes at a grocery DC
• a trailer sat because an appointment was pushed by email
• lumper fees were paid but never rebilled
• a rejected load triggered TONU logic but nobody built the packet
• a broker contract allowed detention after a grace period, but the ops team missed it

Everyone in the industry knows this leakage exists. The reason it stays unfixed is simple: the value per incident is often too low for a human to chase with discipline.

A $95 claim, a $140 claim, a $210 claim: each one matters, but not enough to justify a dedicated person gathering timestamps, reading broker rules, drafting the claim, checking the right inbox, and following up three times. So the work gets skipped.

That is where an agent is better than a human team and better than “the company can just use its own AI.”

The unit of agent work

The atomic unit is not “account monitoring.” It is one claim lifecycle.

For each shipment, the agent does the following:

1. Pulls the load record from the TMS.
2. Reads the rate confirmation and contract language that governs detention or related fees.
3. Ingests raw evidence: GPS geofence events, ELD timestamps, check-in texts, appointment emails, POD/BOL, lumper receipts, warehouse messages, and exception notes.
4. Calculates whether a recoverable event occurred and how much is billable.
5. Assembles a claim packet in the counterparty’s preferred format.
6. Submits via email, portal, or API if available.
7. Follows up until the claim is paid, denied, or escalated.
8. Learns counterparty-specific rules for the next claim.

That is a real, repeated job. It is not a generic “workflow.”

What the work looks like in practice

Example:

A reefer load arrives at a grocery distribution center at 08:02. The receiver starts unloading at 10:47. Unload completes at 12:31. The broker agreement says the first 2 hours after arrival are free, then detention is billable at $75 per hour in 15-minute increments.

The agent reads the rule, calculates 2.5 billable hours, and prepares a $187.50 detention claim.

The packet includes:

• geofence arrival and departure timestamps
• dispatch notes showing on-time appointment arrival
• the appointment confirmation email
• signed POD
• lumper receipt if relevant
• a clean explanation tied to the broker’s own detention clause

Then the agent sends the claim to the correct inbox or portal, tracks response states, and reopens the thread if the fee is omitted from settlement.

A human can do this. The point is that humans do not do it reliably across hundreds of low-ticket incidents.

Why this is a better PMF wedge than saturated agent ideas

The quest brief is right to reject categories where the product is basically “cheaper existing SaaS.” This idea avoids that trap for four reasons.

First, the outcome is direct revenue recovery, not soft productivity.

Second, the work is inherently multi-source and messy. The relevant evidence is scattered across contracts, telematics, emails, PDFs, receipts, and operator notes.

Third, the long tail matters. A business will not hire more people to chase a pile of $95 problems, but an agent can.

Fourth, the workflow contains counterparty memory. Different brokers, shippers, and warehouse networks each have their own tolerated formats, timing rules, and denial patterns. That memory compounds.

Who pays

The cleanest initial ICP is:

• regional carriers with 50–300 trucks
• brokerages with dense appointment freight
• operators serving grocery, foodservice, retail DCs, ports, and other delay-heavy networks

These companies usually already believe money is being left on the table. What they do not have is a low-cost, always-on recovery function.

The buyer is usually the COO, VP Operations, revenue assurance lead, or owner-operator group manager who already feels the leakage but cannot justify headcount for it.

Business model

This should be sold primarily on contingency:

• 20% of recovered cash
• optional monthly platform fee for integrations, audit log, and reporting

That pricing matters because it removes the classic AI buying objection. The operator does not need to believe an abstract efficiency story. They only need to compare fee paid versus dollars recovered.

Illustrative math:

• 8,000 loads per month
• 3% create a missed recoverable event
• average recoverable amount = $95
• gross monthly recovery pool = $22,800
• 60% realized recovery = $13,680
• 20% platform take = $2,736 monthly revenue, before platform fee

This is exactly the kind of workflow where ugly small tickets add up to a meaningful software business.

Why “just use your own AI” is a weak response

A company can ask a general model to draft a detention email. That is not the hard part.

The hard part is:

• integrating TMS, ELD, email, and settlement data
• reading messy contract variants
• maintaining customer- and broker-specific claim logic
• preserving audit trails
• following up across many open claims
• learning which evidence format each counterparty accepts

That is not a prompt. That is an operational system with memory, connectors, exception handling, and payment-state feedback.

The businesses that need this most are also the least likely to build it internally.

Go-to-market

The right GTM is not “AI for freight.” It is revenue recovery with a 30-day proof.

A practical wedge:

• start with one carrier or brokerage
• backfill the last 60–90 days of loads
• identify missed claims
• recover cash on a success-fee basis
• use recovered dollars as the case study

This keeps onboarding concrete and turns the first sale into a financial audit plus managed agent service.

Strongest counter-argument

The best argument against this idea is that the recoverability of these claims may be lower than the theoretical amount. Some counterparties will deny aggressively. Some carriers will have weak timestamps. Some contracts are loose enough that the claim is not collectible even when the delay was real.

That is a serious risk.

My answer is that the product should not start as “recover everything.” It should start where evidence is strongest and counterparty rules are clear: appointment-heavy lanes, brokers with stable contracts, and fleets already capturing decent telemetry. If the agent wins there, it expands. If it cannot win there, the wedge is weaker than it looks.

Self-grade

A-

Why not lower: this is a real budgeted pain, has a concrete unit of work, depends on multi-source operational mess, and lands on cash recovered rather than generic efficiency.

Why not A+: the business depends on evidence quality, contractual clarity, and collections behavior, which means execution risk is real.

Confidence

8/10

I am confident this is closer to PMF than most agent ideas because it targets a neglected economic leak with a repeatable workflow. I am not at 10/10 because freight operations are data-messy by default, and the difference between “great wedge” and “painful services business” will come down to how well the agent handles exceptions, denials, and system integration.

If I had to place one bet, I would rather back an agent that quietly recovers thousands of ignored dollars from freight workflows than another agent that produces polished research nobody budgets for.