Articles
Why the token bill is not your actual problem
Over the past two years, most organizations with any appetite for technology received some version of the same directive from leadership: figure out AI. Not “build a strategy.” Not “identify three use cases with measurable ROI.” Just… figure it out.
So teams did what they do. They spun up pilots. They gave developers access to APIs. They connected a chatbot to a knowledge base. They built a proof of concept that demoed well in the conference room. There was energy, momentum, and genuine excitement.
And almost nobody stopped to ask: what does success actually look like here? What process is this replacing? What does the math need to be for this to pay for itself?
Then the invoice arrived.
For a lot of organizations, the first real moment of reckoning was not a failed project or a missed deadline. It was a line item. Token costs, API usage, model inference fees, whatever the vendor calls it. The number was visible. The return was not.
That is when token anxiety was born. Suddenly there were conversations about cost per query, context window efficiency, and whether the model being used was too large for the task. Teams started optimizing prompts not to get better outputs but to shorten them. Budget owners started asking whether AI was actually worth it.
Token anxiety is not a cost problem. It is a symptom of a planning problem that was there from day one.
The actual issue is not the token bill. Token costs are real, and at scale they deserve attention. But they are also predictable, measurable, and directly tied to usage. When you have a defined use case and a baseline to compare against, the math is straightforward. The problem is that most organizations never established that baseline, which means there is nothing to measure against. When ROI is undefined, any cost feels unjustified.
This is not a technology failure.
The organizations struggling right now are not struggling because the technology did not work. In most cases it worked fine. They are struggling because AI was introduced as an experiment rather than as a solution to a specific business problem.
Experiments are valuable. But experiments need a hypothesis. They need a definition of what a positive result looks like. When you skip that step, you end up with a collection of pilots that demonstrated capability but never translated into production value. And when someone eventually asks what all of it was worth, the honest answer is that no one knows.
That ambiguity is what creates the anxiety. Not the token bill.
What execution actually looks like.
Getting past this requires going back to basics before moving forward. That means identifying the specific workflows where AI creates measurable value, whether that is time savings, error reduction, faster cycle times, or revenue impact. It means putting a number on what the current process costs. It means defining what the AI-assisted version needs to deliver for the project to be worth doing.
It also means being honest about sequencing. Not everything should be automated at once. The organizations getting real return from AI right now are not the ones who launched the most pilots. They are the ones who identified the right two or three use cases, built them properly, measured them, and used those results to justify the next investment.
That is not a revolutionary idea. It is just how good technology projects work. The mandate to “figure out AI” created pressure to skip the fundamentals. Fixing token anxiety means going back and putting them in place.
About the AI Opportunity Roadmap
We built the AI Opportunity Roadmap specifically for this moment. It is a structured engagement that identifies your highest-value AI use cases, maps them to measurable outcomes, and gives you a sequenced plan you can actually execute against. If you are sitting on pilots that never went anywhere, or facing questions about ROI you cannot answer yet, this is the right starting point.