The Invoice Tells You Nothing

An employee at a fintech company spent $81,000 in API tokens building a "brainrot shooter game" in a single day. His company's response was to call it a marketing expense and ask people to play it.

That's funny. It's also a symptom.

I've watched companies push AI access to their teams with real urgency and real excitement, and I get it. The tools are genuinely powerful. But what I keep seeing is the rollout without the foundation. Spend limits get set (sometimes). Usage policies get drafted (occasionally). What almost never happens is the harder question: what are we actually getting for this?

The Slash story surfaces a specific failure mode, but the deeper one is invisible. Most organizations have no way to answer whether their AI investment is producing anything useful. They can see the bill. They cannot see the output. Did anyone write better code? Did any client conversation go differently? Did any decision get made with better information? The receipt doesn't tell you that. And without that visibility, you are not managing a capability. You're managing a tab.

My teams are in client environments doing this work right now. The gap I see most often isn't rogue employees building games (though that happens). It's the inability to connect the spend to the outcome. Companies are measuring usage instead of value. They are counting seats and token consumption and calling it adoption. That is not ROI. That is a spreadsheet that makes you feel like you're tracking something.

Cost limits are table stakes. They stop the bleeding. The harder build is the instrumentation that tells you whether the investment is working at all.

So here is the question I'd ask your leadership team this week: if I asked you to show me the business value your AI investment produced last quarter, not the usage stats, not the token count, what would you show me?