Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

Your new software bill is a 1980s arcade scam

AI may become cheaper to buy, but far more expensive to use.

[Photo: Getty Images]

Profile photo of Cliff Jurkiewicz

VP, Global Strategy and GM, Customer Advisory Council, Executive Evangelist · Phenom

Software budgeting used to be simple. You bought a seat license, you paid the bill, and you moved on.

That era is over.

The new enterprise model is a hybrid creature: you pay a fixed cost for the infrastructure, but your ultimate operating expense is driven by variable consumption. In this new era, total cost is determined entirely by the efficiency of the systems, that is a generalized system versus specialized. Less efficient systems mean spiraling costs; highly specialized systems mean massive savings.

A general contractor, for example, would be an appropriate choice to handle basic plumbing and electrical issues, but not to build a highly technical data center.

To understand where this is going, look at the evolution of the 1980s video arcade.

When a dollar isn’t a dollar

In the early days of the arcade, the math was transparent. Put a dollar in a change machine and get four quarters back. Every game in the room cost exactly 25 cents to play. The economic relationship between dollar and playtime was direct and predictable.

Then, arcade owners realized they could optimize revenue by changing the rules. They swapped quarters for custom tokens.

Suddenly, a dollar only bought three tokens, and the hot new game cost two tokens to play. By decoupling the real dollar from the actual activity, they made it incredibly difficult to figure out how much you were spending to get a high score.

But the real game wasn’t just playing; it was winning. You were playing to collect tickets to cash in at the prize booth. Every parent knows this math: you will easily spend $100 in tokens to walk away with a $5 plastic toy.

This exact setup is happening in enterprise software today. Major platforms are moving toward flexible credit systems that mirror the arcade ecosystem.

AI doesn’t process words or letters. It breaks text into tokens—small chunks, often pieces of words—and reads, writes, and is billed in them. A seven-word sentence, for example, would cost 10 tokens, because punctuation and contractions each cost their own token.

To help your leadership team understand what they are paying for, you must break the bill down into three distinct, interconnected costs.

1. The infrastructure cost    

In video arcade terms, think of  the base SaaS subscription as admission to the arcade. You pay a flat fee just to enter the room, look at the screens, and access the data. It keeps the lights on, but the infrastructure itself doesn’t solve business problems anymore; it’s just the floor you stand on.

2. The token cost and usage

Like playing an arcade game, you need to buy tokens to run the system. But in this case, the token is the chunk of text processed by the model. Companies buy credits and costs fluctuate based on data center supply and demand, much like electricity or gasoline. Because the base commodity price of a token will eventually equalize across ecosystems, trying to negotiate the lowest rate per token is a losing strategy. Machine efficiency is the real differentiator.

3. The integration cost      

Just as arcade cabinets rarely talked to each other—forcing you to cash out your tickets at one machine before feeding new tokens into the next—enterprise tools are often disconnected systems that require costly integration work just to pass a single result from one to another.

This is the hidden tax that breaks enterprise budgets. To get the ultimate business outcome, data usually  must travel across multiple tools. It’s the underlying data architecture of all the machines that will determine how well the data is analyzed and consumed from one machine to the next.

Legacy enterprise platforms have spent years buying up point solutions and trying to stitch them together. Because these systems were never designed to work as a single network, they require heavy external integrations and Application Programming Interface (API) calls just to move data between their own acquired modules. APIs allow different software applications to share data with each other.

You’re burning tokens on one machine just to translate data for the next machine, and vendors are increasingly charging separate line items for the API calls required to make the handoff.

The compounding input tax

To understand why specialized systems win the efficiency battle, look under the hood of how AI bills you. Every single interaction, whether typing a prompt or an automated agent triggers a background function is a two-part transaction. You pay for the input— data and context fed into the machine— and the output— the response the machine generates.

To make a conversation feel continuous, the software has to take the original prompt, the model’s first response and the second prompt, and send them all back into the machine as a massive, combined input block for the next turn. Every round of clarification multiplies the input bill.

This structural reality explains exactly why corporate chatbots run out of memory, start hallucinating, or force you to hit “New Chat.” The system could be reaching its physical limits if it’s not employing persistent caching.

When a law firm uses a generic model like ChatGPT to do contract research, it starts at a disadvantage  because the model knows a little bit about everything but nothing specific about case law. Conversely, a specialized tool like Harvey, built for the legal profession, already speaks the language of the courts. The context is baked into the platform from step one, keeping the initial input small and the costs low.

Integration vs. orchestration

This is where the mass-market players expose their biggest weakness. Legacy incumbents like the major Human Capital Management (HCM) systems and Applicant Tracking Systems (ATS) are rushing to bolt on acquisitions and point solutions, proudly showing off their “integrated networks.”

When challenged, their defenders ask: Aren’t these companies the industry experts?    

The answer is no. These platforms were built as systems of record and process. They were designed to store data, host static compliance fields, and track administrative workflows.

Highly specialized, purpose-built platforms, on the other hand, orchestrate the experience.

An orchestration layer meets an employee exactly within a specific corporate role, utilizing the technology they choose to use, to deliver the precise output required to run the business. True orchestration eliminates background data loops and maximizes token efficiency.

This is why specialized, applied AI platforms are how businesses win.

Ask about tokens, not dollars    

Multiply that operational inefficiency across the entire workforce. If your company runs an automated workflow 10,000 times a day, the difference between a precise system and an inefficient one is the difference between a predictable operating expense and a runaway utility bill that rivals payroll.

When a software vendor sits across from you to pitch new AI capabilities, the conversation will steer toward token discounts. They want you focused on the front-end exchange rate at the arcade counter. Don’t take the bait.

Stop asking how much things cost. Start asking: How many tokens does it use to get the output I need?    

Cliff Jurkiewicz is vice president of global strategy and general manager of the Customer Advisory Council at Phenom.