AI has quickly become one of the most powerful tools available to modern businesses. Companies are using AI to write emails, analyze information, generate code, summarize documents, create content, support customers, and automate countless other tasks.
But there is a hidden problem that many businesses may not be watching closely enough: the cost of how AI is being used.
Your company may not actually need less AI. It may simply need to use the right AI model for the right task.
That was the discovery Jeremy Barry and his co-founder made after encountering an unexpected problem during their internships. Heavy AI usage triggered warnings about their organizations' token consumption. What initially looked like an employee-usage problem became something much bigger: a business opportunity to help companies route AI requests more efficiently.
As the co-founder of Zumah, Jeremy is working on a system designed to analyze an AI request before it is processed and automatically route it to the most cost-effective model capable of completing the task.
The potential impact is significant. Jeremy says Zumah has seen potential AI cost reductions ranging from 60% to 85%, while an early client pilot reduced token costs by approximately 70% in just two weeks.
The bigger lesson for business owners, however, goes beyond AI.
Technology should create a return—not become an uncontrolled expense.
The Hidden Cost of AI Isn't Always the AI
When business owners think about AI expenses, they often focus on subscriptions.
A company might pay for ChatGPT, Claude, an API, a custom large language model, or another AI platform and assume that the subscription or software fee represents the cost of using AI.
At scale, however, the economics can become much more complicated.
Jeremy explains that large language models use tokens as a measurement for the text and context they process. Prompts, responses, and additional context all contribute to token consumption. For an individual using AI occasionally, the cost may seem insignificant. For an organization where dozens, hundreds, or thousands of employees are constantly interacting with AI systems, those costs can multiply rapidly.
This creates an important distinction:
The question isn't simply, "How much AI are we using?"
The better question is:
"Are we using the most appropriate AI model for each task?"
Why Businesses May Be Paying for More AI Power Than They Need
One of the central problems Jeremy identifies is that users often default to the most powerful model available.
That makes sense from the employee's perspective.
If a powerful model is available, why not use it?
The problem is that not every task requires maximum capability.
Consider something as simple as checking the grammar of a short email or summarizing a straightforward document. A highly sophisticated model may be capable of completing the task, but a less expensive model may be capable of producing an equally useful result.
If employees consistently send simple requests to premium models, the organization can end up paying premium prices for relatively simple work.
Jeremy describes Zumah's approach as essentially finding the most efficient route for each request. Instead of asking employees to constantly think about which model they should use, the system analyzes the request and routes it to a model capable of handling the task at a lower cost.
That concept changes the conversation from AI reduction to AI optimization.
The goal isn't necessarily to stop employees from using AI.
It is to make the AI infrastructure behind their work more efficient.
The Business Opportunity Hidden Inside an Internship Problem
One of the most interesting parts of Jeremy's story is how the idea for Zumah emerged.
Jeremy and his friend Sam were already interested in entrepreneurship, coding, and artificial intelligence. They had previously built an app called Crawler, which attracted more than 500 users but ultimately became more of a learning experience than a profitable business.
Then both entrepreneurs started internships in Washington, D.C.
Within roughly the first week, each received a warning related to their AI usage. Jeremy says his intern group had consumed a substantial portion of the company's token budget within its first couple of days. His friend experienced a similar issue at Capital One.
Instead of simply changing their own behavior, they started asking a bigger question:
If this is happening to us, how many other companies are experiencing the same problem?
They began speaking with other people, including alumni from William & Mary, and discovered that excessive AI costs appeared to be a broader issue.
That led them to build the technology.
It's a powerful entrepreneurial pattern: experience a problem personally, investigate whether the problem is widespread, and then build a solution around it.
AI Cost Optimization Doesn't Have to Change Employee Behavior
One of the strongest ideas behind Zumah's approach is that employees shouldn't necessarily have to become AI cost-management experts.
Imagine telling every employee:
"Before you enter this prompt, determine which AI model is cheapest while still being powerful enough to complete the task."
That creates friction.
It also creates another training requirement for the organization.
Instead, Zumah's system is designed to work in the background. Jeremy explains that it can integrate with platforms such as ChatGPT and Claude, while the company's SDK can also be integrated with custom AI systems.
The objective is simple: let employees continue working the way they already work while the technology handles the optimization behind the scenes.
Jeremy compares the concept to choosing the optimal highway to reach a destination.
The destination stays the same.
The route changes.
And the goal is to find the route that gets you there efficiently and economically.
Why AI Costs Become More Important at Enterprise Scale
For a single person, saving a few cents on an AI request may not seem meaningful.
For a company with a large workforce making thousands or millions of AI requests, the equation changes dramatically.
This is where scale becomes important.
A small business might not have enough AI usage for optimization to make a meaningful financial difference. But as AI becomes embedded into everyday workflows, the aggregate cost can become substantial.
Jeremy explains that businesses paying directly for API usage are particularly relevant candidates for this type of optimization. He also notes that organizations with more than roughly ten people may begin encountering usage patterns where API costs become meaningful, depending on how they use AI.
That suggests an important question for growing companies:
As AI adoption increases, are you measuring the cost of AI with the same discipline you apply to other operating expenses?
If the answer is no, there may be an opportunity to improve.
AI Cost Optimization Starts With Measurement
Before a company can optimize an expense, it has to understand the expense.
That's why Jeremy describes one of Zumah's first steps as a token audit.
The audit is designed to look at how the organization is using AI and identify opportunities for improvement. Jeremy says the initial audit is free and begins with a conversation about how the business currently uses AI.
This reflects a broader business principle that applies far beyond artificial intelligence:
You cannot optimize what you don't measure.
The same principle applies to payroll, advertising, inventory, software subscriptions, sales conversion, customer acquisition, and operational processes.
AI should be treated the same way.
If your company is spending money on AI but doesn't know where that money is going, it becomes difficult to determine whether the investment is producing an appropriate return.
A Real-World AI Cost Reduction Example
The concept becomes more compelling when it moves from theory into production.
Jeremy shared that Zumah had begun piloting its system with its first client. After approximately two weeks, the company had reduced that client's token costs by around 70%.
That result should not be interpreted as a guarantee that every organization will achieve the same reduction. AI usage patterns differ significantly between businesses, and savings depend on how an organization uses models and APIs.
But the pilot demonstrates why measuring AI usage can matter.
A percentage reduction that sounds modest in isolation can become financially significant when applied to a large and growing AI budget.
And as businesses continue incorporating AI into more workflows, those costs may become an increasingly important component of operating expenses.
Could AI Cost Management Eventually Become Invisible?
Jeremy's long-term vision is particularly interesting.
Today, because Zumah is still early in its development and working with a small number of clients, the company remains closely involved with customers to understand what works, identify bugs, and improve the product.
But the long-term objective is less friction.
Jeremy envisions a system that can eventually operate almost entirely in the background, allowing businesses to use AI without constantly thinking about the optimization layer underneath it.
That could become an important direction for enterprise AI.
As companies adopt more AI tools, they may need a new category of infrastructure focused not simply on making AI more capable, but on making AI more economical, measurable, and controllable.
The future of business AI may therefore involve two separate questions:
What can AI do?
And:
What is the most efficient way to make AI do it?
The Bigger Entrepreneurial Lesson: Find the Problem Behind the Problem
Jeremy's story also offers an important lesson for entrepreneurs.
He and his co-founder didn't start with a sophisticated business plan for AI cost optimization.
They encountered a problem.
They experienced the problem themselves.
Then they investigated whether other people had the same problem.
From there, they began building a solution.
That sequence—problem, investigation, validation, solution—is one of the most practical foundations of entrepreneurship.
Their earlier experience building Crawler also mattered. Even though that project wasn't ultimately positioned as a profitable business, it gave them experience building technology, attracting users, and learning what happens when an idea meets the real world.
Entrepreneurship rarely follows a perfectly straight line.
Sometimes the project that doesn't become the business teaches you how to build the business that does.
AI Should Be an Investment, Not an Uncontrolled Expense
Perhaps the most important takeaway from the conversation is simple:
AI should generate a return.
Jeremy argues that companies shouldn't think about AI as an inherently expensive technology that will inevitably consume their budgets. Instead, businesses should focus on whether the money they invest in AI produces a meaningful return.
That is a mindset worth applying to every technology investment.
Don't ask only:
"Can we use AI?"
Ask:
"Where does AI create value?"
"How much does that value cost?"
"Are we using the right tools for the job?"
"Are we measuring our usage?"
"And are we getting the return we expected?"
For business owners and leaders, these questions can turn AI from another technology expense into a measurable business asset.
What Business Leaders Should Do Next
If your organization is using AI extensively, the first step may not be buying another AI tool.
It may be taking a closer look at what you already have.
Understand which AI systems your employees are using. Determine whether you're paying subscription fees, API costs, or both. Examine how frequently employees and applications interact with AI models. Look for repetitive, simple tasks that may not require the most expensive models.
Most importantly, start measuring.
AI adoption is accelerating, and the companies that benefit most may not simply be the companies using the most AI. They may be the companies that understand how to integrate AI into their systems while maintaining control over cost, productivity, and return on investment.
For more conversations about business growth, systems, communication, personal development, and the strategies entrepreneurs can use to build stronger businesses, visit Glovisor.
The Future of AI Is About More Than Capability
AI has already changed what businesses can accomplish.
The next stage may be about learning how to accomplish those things more intelligently.
Jeremy Barry and Zumah are tackling one piece of that challenge: making sure businesses aren't automatically paying premium prices for tasks that don't require premium processing.
The lesson isn't that businesses should use less AI.
It is that businesses should use AI more strategically.
As AI becomes increasingly embedded in everyday operations, cost optimization may become just as important as adoption.
Because the real goal isn't simply to have artificial intelligence working inside your company.
It's to make sure that intelligence creates a return.