I measure my AI spend against a 10x return.
A lot of people paying for AI tools know the monthly bill and can't say what came back from it. In a September 2026 Vibe Code to Profit live, I did that math on my own spend on camera. The bar I hold it to is at least 10x.
The business case
I think of AI as raw material. If you make semiconductors, sand is your raw material. For developers and entrepreneurs who vibe code, AI plays that role. The money you put into it only matters if it comes back on the business side.
The live was about whether the money going into AI credits and subscriptions turns into money in the bank. At $20, $200 or $2,000 a month, the number that matters is the return.
In the video I said the AI companies are moving the goalposts on usage and credits. I also said the industry is still working out how to measure this return. Prices and limits keep changing, so you need your own number. Without one, you end up burning money and tokens with no business model behind them.
Do the math first
These are my numbers from the live, as of September 2026. I was on the Codex plan at $200 a month and the Claude Max plan at $200 a month, so about $400 a month before any other subscriptions. Over nine months that's $3,600. In that same stretch, one customer paid me $60,000. That works out to about $16 back for every dollar, or roughly 1,500%.
I used one customer on purpose to keep the math narrow. I also said nobody has to hit 16x. I run a business, so I try to make sure my AI spend turns back into dollars and cents. The rule I gave the live was $10 back for every dollar in. I said it should be closer to $20, and 10 keeps it conservative.
Running the numbers on your own spend takes one prompt. Ask your favorite LLM to go through your email, find the AI subscriptions you've paid for, put them in a spreadsheet and total them. Then put what you've made from a client next to that total. If it's zero, put zero, and at least you know your ROI is negative for now. That's fine while you're still learning from customers, as long as you know it.
Watch the small charges
Subscriptions creep up on you. I'd just cancelled Manus, which had been costing me about $40 a month. I'd also been paying for v0, Lovable and Bolt while I tried different tools for different features. I called it $100 a month in quick math, and said it was more than that.
A $20 tool here and a $30 tool there turns into $150 a month, and at a 10x bar that $150 needs $1,500 coming back.
Jump to
- 0:00 Proving the ROI of AI
- 5:18 A business framework: AI as raw material
- 29:42 Codex credits and the 10x bar
- 39:46 Passing AI costs to customers
- 1:00:08 My AI agent setup
- 1:29:56 Portfolio review and buyer messaging
- 1:39:44 Writing code vs building a product
I walk through my numbers, the credit resets and my AI agent setup on screen in the full video on my channel: Prove the AI ROI live on YouTube.
Plan for cost changes
Your costs can move under you. My Codex usage had felt off for about a week before the live. On my 30-day chart, each gap was a day I'd run out of Codex credits and moved over to Claude for a while before coming back. In September I'd already had five credit resets on that $200 plan, which by my count made it about $1,000 of Codex use. During the live I was at 0% with four days until the next reset, deciding whether to spend more.
Next on the live I raised how you pass this cost on to your customers. I said it needs careful thought while the vendors keep changing the rules.
Productivity means output
Companies run into the same problem at a bigger scale. A team sets a budget, say $2,000 a month, and gives everybody a Claude subscription so they can build spreadsheets faster. The test is whether that turns into more customer conversations, more leads or more sales. A lot of developers I talk to are stuck in what I call token hell, building app after app on new ideas until 10 or 15 of them sit around with no customer. Productivity should show up as output, and more hours at the computer don't count.
Count the agents too
I run the same check on my AI staff. I use an AI agent setup to run parts of the company. I talk to a CEO bot, and it directs a VP of sales, a VP of operations, a CFO and a VP of marketing, each with workers under them. On the live, the setup counted 41 AI staff, plus me and three people. Call it 45, with about 90% of them digital.
Real employees are the easy part of the math, because customers have to cover their cost and then some. The AI side is harder to see because the bill is small. I put my AI agent setup on a $20 a month plan on purpose to see how far $20 goes, and it runs those 41 digital employees. They do research and spreadsheets, and they wrote the post announcing that live. I'm getting value from them. I'm not getting the value of 41 employees yet.
Put a number on tests
The same rule applies when I spend on growth. On the live I committed $500 to ads on Facebook for LeadScore AI, with the campaign brief from my AI agents and the ads built in Codex. For this test I held myself to the 16x I was already getting, so that $500 has to bring back at least $8,000. There's no point having AI do a lot of work if you can't prove it makes money back.
Money in the bank
The point of AI spend is money in the bank. Pick your bar, total what you pay, and put what came back next to it. If you already have customers and want the systems that tie your spend to revenue as you grow, that's what I cover in The Scaling Entrepreneur.
About Dwain
I'm Dwain Browne, Founder of LeadScore AI and SnapSuite. I'm a Toronto-based software architect with 25+ years across software, sales, operations and entrepreneurship. SnapSuite helps contractors and trades teams get paid faster, and LeadScore AI shows you which leads deserve your next call. I also run Vibe Code to Profit, a community helping people build software with AI.