Tokenmaxxing: one AI budget, four jobs<br>Contact salesStart for free

All articlesAI Trends/May 25, 2026<br>Tokenmaxxing: one AI budget, four jobs<br>8 minutes read

SLShawn Lestage<br>COO

Share this article

SLShawn Lestage<br>COO

Share this article

What's the actual ROI of tokenmaxxing?

Every company using AI is now paying for tokens. At Mintlify, we package tokens into Credits. The unit isn't the point. The point is that whatever you're paying for, you're paying for a solution to various problems and everyone is scrambling to understand: is the spend worth it?

The challenge a lot of teams have right now is that they can't draw a clean line between AI spend and business outcomes. They look at the invoice and can't draw a clean line to ROI in the categories that matter: engineering output, support cost, customer retention, or new revenue.

That's not because the ROI doesn't exist. It's because no one's figured out how to draw it cleanly so buyers can make an informed decision on their AI spend.

To summarize:

Everyone is tokenmaxxing (full sending their R&D budgets and hoping for the best)

Companies realize they're spending too much on AI when their margins compress and the impact is not yet obvious.

Those same companies switch to cheaper "good enough" models or start capping their spend.

We have watched this arc play out quickly across Mintlify customers. So, what happens next?

If the spend doesn't show up in faster growth, better margins, reduced costs elsewhere in the business, or, ideally, in multiple categories, the final outcome is fairly predictable.

Structuring AI spend

Here's how we think about our own spend at Mintlify, which is instructive for how we build, how we sell, and the value we deliver to our customers. The evaluation framework is simple. Credits map to well understood P&L categories: R&D, COGS, S&M (CAC), and Retention (NRR).

Token spend doesn't all go into one accounting bucket.

It's several costs doing fundamentally different jobs, and each one should be evaluated with a different ROI lens. The same way you wouldn't lump headcount, ad spend, and cloud infrastructure into one budget line, you shouldn't lump Credits together.

Credits should map to the P&L categories finance teams already understand.

Credits as R&D

Some Credits make your engineering team faster. Think usage spend on tools like Claude Code and Cursor.

In the Mintlify product suite, the equivalents are docs writing Agent and Workflows. Those products auto-document pull requests, sync changelogs, and keep your knowledge base current as the product changes underneath it.

On paper, this increases cost per engineer. In practice, it increases output per engineer. And unlike most developer tools, the returns compound. Every incremental improvement to your internal or external knowledge base makes every future query against it more accurate. A developer asking your docs a question in month six gets a better answer than the same question in month one, because the underlying content has been kept in sync automatically.

Measure it the way you'd measure any engineering productivity tool. Did documentation lag decrease? Are engineers spending less time writing and updating docs? Is the knowledge base more complete and more current than it was before? If yes, the Credits are ROI positive.

Credits as COGS

Some Credits serve your end users directly. When someone visits your documentation and asks the Mintlify Assistant a question, or when a customer files a support ticket and the Support Agent picks it up before a human does, that exchange costs Credits. This is the cost of delivering the product.

The ROI math here is the most straightforward of the four categories. If your customer's current cost per support interaction is $X, and the Assistant or Support Agent resolves the same question for a fraction of $X, the margin improvement is immediate and measurable. You're looking at deflection rate, cost per resolution compared to what you're paying for human support or an alternative solution today, and the gross margin impact over time.

This is where the value proposition is most concrete for support-heavy use cases. A help center powered by Mintlify's Assistant and Support Agent doesn't just reduce ticket volume. It shifts the cost structure from a variable, headcount-driven model to a predictable, Credits-driven one. For most companies, that's a better margin profile.

Credits as CAC

Some Credits acquire customers. This is the least obvious category and potentially the most valuable.

When a developer lands on your docs and asks the Mintlify Assistant a deep integration question, that interaction is a signal. It tells you that someone is actively evaluating your product, trying to understand whether it fits their use case, and engaging with your content at a level that suggests real intent.

The Assistant API extends this further. Companies plugging Mintlify's Assistant into their own tools can surface these high-intent...