AI isn’t all about automating core enterprise features at Fortune 500 corporations. Small and medium-sized companies also can use AI to optimize, economize, and in some instances compete extra successfully in opposition to a lot bigger rivals.
An Austin, Texas–primarily based vegan cheese-maker referred to as Insurgent Cheese used it to degree the enjoying discipline in opposition to a bigger provider. Particularly, the corporate developed a small system of AI instruments to assist it claw again overcharges from a significant transport service.
The corporate is maybe finest recognized for profitable a $750,000 funding from Mark Cuban, cash it used to develop Insurgent Cheese into what it says is now a $20 million enterprise. Cuban lately spoke in regards to the firm’s artful use of AI onstage on the Convergence AI occasion in Dallas.
The Downside
Rebel Cheese ships tens of hundreds of orders of perishable, handcrafted vegan cheese throughout the nation. The vacation season is by far its busiest interval. “This fall is all palms, heads down, get it out the door, be certain prospects are completely satisfied,” the corporate’s cofounder, Kirsten Maitland, wrote in a recent blog post. “There’s no time to cease and analyze something.”
After this previous vacation season, Maitland took a have a look at the corporate’s checking account, and one thing appeared off. Insurgent Cheese had simply had its finest vacation season ever, but the numbers didn’t replicate it. So she began digging to seek out out the place the earnings had been leaking away. She found that the corporate had paid $250,000 extra for transport than deliberate.
Hiring new staff to analysis and repair the issue wasn’t within the playing cards. So Maitland turned to Anthropic’s Claude. “I handed it a yr of invoices and a contract,” she tells Quick Firm in an electronic mail alternate, “and it discovered patterns I’d have wanted a forensic accountant to floor, which might have been time-consuming and costly.”
She says the service’s transport invoices run lots of of pages per week, with charges layered inside charges. “Most shippers don’t have the time or instruments to audit them,” Maitland says. For the service, the complexity was not a bug however a characteristic, and a worthwhile one.
Her evaluation turned up a number of causes, not only one. “Some had been our fault, like important weight overages on our packages, which we might repair,” she says. “The remainder had been on the service: That they had put a customized contract in place for us. Below that contract, any package deal bulging or weight overage triggered drastic value spikes.”
By the point Maitland sat down with representatives from the transport firm, she had analyzed a yr’s value of information and will present them precisely which contract clauses had been doing the harm. The largest overcharges mapped to a brand new weight restrict the service had applied, however not communicated, in early 2025. Their response was: “Nicely, you need to have caught it.” She vowed by no means to let that occur once more.
The construct
To construct the precise packages that learn the invoices and request refunds, Maitland used Manus, an AI orchestration layer that coordinates work amongst varied brokers and subagents, utilizing completely different fashions for various duties. Maitland says she additionally examined Bolt, Lovable, and Relay, however discovered that Manus dealt with the job extra simply and precisely.
After a whole lot of experimentation and discovery, she was in a position to architect a system that automated a data-heavy auditing course of that had beforehand required handbook evaluation of tens of hundreds of shipments. The construct unfolded in 4 distinct phases:
1. Standardizing the “Reality.” The method started with knowledge preparation. Maitland created two easy comma-separated values (CSV) templates. A “Zone Information File” contained Insurgent Cheese’s negotiated contract charges, whereas a “Transaction File” contained weekly bill knowledge. This gave the AI a transparent construction for evaluating “what we must always pay” in opposition to “what we had been truly billed.”
2. Designing the Blueprint. She uploaded instance invoices, Insurgent Cheese’s service contract, and a presentation detailing the outcomes of her Claude-assisted investigation into the overcharges to Manus. Reasonably than instantly asking the AI to “construct a instrument,” Maitland first used it to generate a complete “Necessities and Design Doc.” The doc served as a technical blueprint, laying out the enterprise logic for “fuzzy weight matching” and strategies for flagging discrepancies. The step ensured the AI understood edge instances like gas surcharges and weight brackets earlier than a single line of code was written.
3. Constructing by way of Orchestration. Maitland then requested Manus to construct a instrument primarily based on the blueprint doc. Her immediate started: “I would like you to construct a standalone, single-page net software that acts as a Provider Billing Discrepancy Detection Device (works for any service — UPS, FedEx, USPS, or your particular transport associate).” She stipulated that the instrument ought to flag each cargo the place the precise cost exceeded the contracted price by greater than ten cents. These overcharges would then be despatched to the service alongside a request for credit score for each discrepancy it couldn’t justify.
4. Steady Auditing and Strategic Insights. As soon as the instrument flags overcharges, Maitland feeds the information again into Claude, which analyzes the logs for higher-level patterns, resembling transport zones the place prices are spiking. That reworked the instrument from a easy bill checker right into a everlasting restoration system.
The Consequence
The system Maitland constructed audits all transport prices by evaluating service invoices in opposition to Insurgent Cheese’s contracted charges. It flags each discrepancy and generates a report that’s despatched on to the transport service, requesting credit score for each overcharge that may’t be justified. Maitland says the service has authorized and credited each declare the system has submitted thus far. She pays about $200 monthly for her Claude and Manus subscriptions, and says the corporate is now saving between $1,500 and $4,000 each week.
Now that Insurgent Cheese has gained expertise with, and belief in, AI automation, the corporate is already utilizing the expertise for different core enterprise features, Maitland says. She constructed an agent that displays the fundraising pipeline, researches VCs, and prepares her for investor conferences. She additionally constructed a website that handles inbound donations and partnership requests. One other instrument makes use of historic knowledge to draft advertisements.
“The larger shift was realizing that is what closes the hole for corporations our dimension,” she says. “We don’t have an engineering workforce. We don’t have an information analytics workforce. Just a few years in the past, I’d have needed to hire a marketing consultant … now I can do the work myself in a day.”
