- A Mark Cuban-backed vegan cheese company built an AI agent to audit weekly shipping invoices against carrier contracts... and recovered roughly $400,000 in one year.
- The win wasn't magic. It was document review at scale: zone, weight, dimensions, and contract terms compared line by line.
- They kept humans in the loop for uploads and final verification before disputing charges with carriers.
- Computer vision on outbound boxes catches bulges as small as 1/8 inch that trigger dimensional surcharges your eye might miss.
- If you have repetitive yes/no review work (invoices, vendor bills, job photos), that's where AI pays for itself fast.
Rebel Cheese makes luxury plant-based cheese wheels for charcuterie boards, not shredded bag stuff. After Shark Tank season 15, Mark Cuban and Lori Greiner backed the Austin company. Recently, they found a use case for AI that has nothing to do with marketing copy or chatbots... and everything to do with money already leaking out the door.
The problem: shipping invoices nobody has time to read
After a strong holiday season, CEO Kirsten Maitland noticed profits weren't matching sales. Digging into shipping costs, she estimated the company had overpaid carriers by about $250,000 (Fast Company).
That's not unusual. Carrier invoices can run hundreds of pages per week, with fees nested inside fees. Most small teams don't have a forensic accountant on staff to audit every line against every contract clause. For the carrier, complexity isn't always a bug. It's often profitable.
Rebel Cheese suspected overcharges across multiple shipping partners. Not necessarily fraud... just mismatches between what the contract said and what showed up on the bill.
What they built: an agent that reads contracts like a tireless clerk
The company built an AI agent that:
- Ingests weekly shipping invoices alongside each carrier's contract terms.
- Compares rates by zone, weight, package dimensions, and other specs in the agreement.
- Flags discrepancies with proof and data the team can take back to the carrier for credits.
Maitland told Fast Company she handed a year of invoices and a contract to Claude and it surfaced patterns that would have taken a human auditor weeks to find... including a new weight limit the carrier implemented in early 2025 without telling them.
They later used Manus to turn that one-time analysis into a repeatable Carrier Billing Discrepancy Detection Tool that flags shipments charged more than ten cents over the contracted rate and auto-generates refund requests.
The reported savings climbed to about $400,000 in one year (Yahoo Finance). Weekly recoveries reportedly run $1,500 to $4,000 against roughly $200/month in AI subscriptions... a concrete ROI story, not a pilot that never ships.
If you're still figuring out which tools fit problems like this, the AI tools checklist is a practical place to start curating what you'll actually use.
Layer two: computer vision on the box before it leaves the building
Invoice auditing catches money after the fact. Rebel Cheese went further.
They set up a camera connected to AI that scans packages before shipment, looking for bulging boxes... deformations as small as 1/8 of an inch. You might not notice that with your eyes. Carriers will, and dimensional weight surcharges can stack fast.
Catch the problem at the packing station and you avoid the charge entirely. That's the difference between recovery and prevention.
Human in the loop (and why 80% is a win)
Rebel Cheese didn't fire the humans and let the agent dispute charges blindly.
Humans upload invoices, review flagged outputs, and verify before contacting carriers. That's the pattern that actually works for small businesses:
- AI does the grunt work: read hundreds of pages, cross-reference contracts, narrow ten invoices down to three candidates.
- Humans do the judgment work: which disputes are real, which relationships matter, which credits to push for.
Out of ten flagged invoices, maybe only one is a valid overcharge. That's still a massive win. You didn't manually read ten contracts. You reviewed one decision.
We talk about this constantly on Infacto Daily: human in the loop and 80% is success. If AI clears 80% of the repetitive review and a person handles the 20% that actually moves money or risk, you got there faster than hiring a full-time auditor and training them for months.
That frees entry-level capacity for higher-leverage work... or, as Jackson put it on the episode, it saves the company a salary worth of manual invoice grinding.
Where this pattern shows up outside shipping
The Rebel Cheese story is shipping. The shape is universal:
Repetitive document or image review → yes/no or flag decision → human verifies → action.
Examples from the episode:
- HVAC / field service: Tech finishes a job, takes photos, AI checks against a checklist (cap on a valve, coil seated correctly) before the journeyman gets pulled in.
- Vendor bills: Same agent logic as shipping... contract rate vs. line item, every week.
- Any process where a junior person reads a pile of papers and hands three suspicious ones to a senior person.
The variables change (contracts update, new surcharge rules). Feed those updates into the system. The core job... compare this document to this rule set... doesn't evolve much. That's exactly what agents handle well.
Video review is still harder than photos today, but the direction is clear: record the work, let AI do first-pass QA, escalate only what needs expensive labor.
What to steal this week
You don't need a Shark Tank backer or a custom agency build to run the thought experiment:
- Pick one recurring bill or document stream you suspect is wrong but never audit (shipping, SaaS vendors, materials suppliers).
- Gather the contract + one month of invoices in one place.
- Ask an AI model to find line items that don't match contracted rates... zone, weight, unit price, whatever your agreement specifies.
- Have a human verify the top flags before you send any dispute.
- Log what you recover. If it beats the cost of the tool, automate weekly.
For a broader menu of free starting points, Infacto's tools hub collects practical utilities without stacking subscriptions you won't open.
Conclusion
Rebel Cheese didn't use AI to write Instagram captions. They used it to claw back $400,000 in shipping overcharges and catch box bulges before carriers did.
That's the pragmatic version of "how do I actually use this?" Document review. Contract comparison. Human verification. Repeat weekly.
If your business has a manual process where someone reads a stack of papers and answers "is this right?"... you probably have a Rebel Cheese-sized opportunity sitting in a folder nobody opens.
Want prompt starters for everyday business tasks like this? Browse the AI Prompt Library.