I spent $7,500 on Claude last month. It returned 66x.

Last month my Claude API bill was $7,500.

Not subscriptions. Actual token usage. And most people see that number and assume I lost my mind.

I see the same number and call it the best money we've spent all year. Because that $7,500 found $500,000 in costs I cut three days later, sharpened our sales process, and started coaching my team for me.

That's a 66x return on the cost line alone. And I think the real number is higher.

Here's where it actually went.

Lead dossiers before every sales call

When a lead hits our inbox, the agent fleet builds a full dossier before our team ever opens the email.

Company size. Growth rate. Expansion potential. Decision makers. The angles we should pitch. For one multi-billion-dollar account that came in last week, the dossier flagged programmatic SEO, international content localization, and a Single Brain deployment to handle paid media across 40+ new locations.

Our reps walk into the call already knowing the play. After the call, the same agent listens to the recording, updates the CRM, and drafts the follow-up.

The work that used to take a BDR a full afternoon now takes zero seconds of human time. The reps spend that time on actual selling.

A coach that has no ego

I had Alfred, my chief-of-staff bot, review packaging for our YouTube editor. It pulled the YouTube API, looked at who was winning in the AI space (Nate Herk, Alex Finn, Matthew Berman), and gave us a real critique.

The verdict: stop competing on Claude code tutorials. Double down on the revenue and business angle. That's our only differentiated lane. Title scores were 4 out of 10, 3 out of 10, 7 out of 10. Specific. Useful. No emotional charge.

If I gave that feedback, it lands as "the boss is mad." A bot gives the same feedback and it lands as data. Same content, half the friction.

It also runs a Friday check-in across the team asking what each person automated that week. The high-leverage people get amplified. The people falling behind get a nudge, or a flag goes to their manager. The whole company gets sharper without me chasing anyone.

$500K cut in three days

The big number came from finance.

I pointed the agents at our spend and asked where we were leaking. They surfaced $500,000 in costs I could cut. I cut them in three days. That alone paid back the API bill 66 times over, and that's before counting the sales lift, the coaching, the product feedback loops, or the hours I got back.

This is what people miss when they squint at a $7,500 line item. You're not buying tokens. You're buying a leverage layer that touches sales, finance, content, ops, and people management at the same time.

Pay the toll on frontier models. Optimize later.

The pushback I get on this is always the same: "Eric, you can run all this on open source for free."

You can. I have the hardware. I've tried it. Open models are getting good. Gemma, Kimi, GLM, Qwen are all real options.

Here's the trade-off nobody wants to admit. If frontier models finish the job in a day and cost more, and open source finishes the same job in a week for free, the open source option is more expensive. Your time isn't free. Your team's time isn't free.

Right now I'm in learning mode, so I'm burning frontier tokens on purpose to figure out where the leverage actually is. Once I know, maybe 15-20% of spend stays on frontier and 80% drops to cheaper or open models. But that optimization comes after you understand the system, not before.

The month-by-month curve

Month one is a disaster. Broken automations, hallucinations, memory loss, gateway resets, agents forgetting context. Painful.

Month two things start to click. The agents remember. You learn what upgrades they actually need. The fleet stabilizes.

Month three is the flywheel. You're piping data sources into one shared brain, the agents query it, your team queries it, and the whole org starts moving 100x. You also see clearly which people are leaning in and which aren't, which is its own kind of information.

If you skip month one because it sucks, you never see month three.

The full breakdown of where the $7,500 went and what each piece returned is in the video: https://www.youtube.com/watch?v=y34wRangQ9M

To paying the toll early,

Eric Siu