ROMTI · Return on Model Token Investment
What is your AI actually returning?
A price per token is not a cost. The cost is what one usable result takes.
ROMTI is how we measure it. Every AI tool you pay for and every job you hand it, priced per result instead of per million tokens, with each number labelled by where it came from.
The rate moves. The result is what you keep.
The same capability is getting cheaper while the top tier costs more, and the models underneath your workflows retire on someone else’s calendar. A process built around a price breaks at every change. A process built around a result survives it.
What one result really costs
The ROMTI line
the token rate × runs to a usable output + minutes spent checking it + the cost of switching when the model retires
Only the first term is on a pricing page. The other three are where the money goes, and they are the part we measure.
What we look at
Every seat and subscription
What you pay for, what gets used, and where two tools do one job. Built from receipts, so it is real money, not an estimate.
Every job you give AI
Each task matched to the lightest model that does it right. Judgment and ambiguity earn a frontier model. Extraction and reformatting do not.
Cost per result
Your real workflows, priced by the runs they take and the minutes someone spends checking them, not by the tokens alone.
Retirement exposure
Every saved prompt, default and automation pinned to a model with a retirement date, so the next change is a scheduled task instead of a broken morning.
What you keep
- A stack ledger. Every AI tool you pay for, what it is used for, and what to cut.
- A model map your team can run. Which job goes to which model, written so it works without us in the room.
- Cost per result for your top workflows, before and after the changes.
- A retirement watchlist and a quarterly re-price routine, so the numbers stay true after we leave.
Every number says where it came from
Most consumer AI plans do not export token usage to anyone, including the person paying. So we never claim to read your bill, and every figure carries one of four labels.
Receipts
What you were actually charged. The strongest number in the room.
Measured
From usage logs your own tools keep, where they keep them.
Metered
From API and admin accounts that report usage directly.
Modelled
Estimated from published list prices and your settings. Always called an estimate.
Price one prompt first
The free ROMTI simulator runs in your browser. Paste a real prompt, pick a model, and see what one request, a full conversation and a month of the habit cost across Claude, ChatGPT, Gemini, Copilot and Grok. Nothing you type leaves the page.
Questions people ask
What does ROMTI stand for?
Return on Model Token Investment. It measures what your AI spend returns per usable result, not per million tokens, so a cheaper model that needs three tries and a more expensive one that needs one can be compared honestly.
Do you need access to my AI accounts?
No. We work from what you were charged, the settings you choose, and, where your tools keep them, your own usage logs. Most consumer AI plans do not export token usage to anyone, including the person paying, so we never claim to read your bill. Every number is labelled by where it came from.
Is the cheapest AI model always the best value?
No. The cost of a result is the token rate, times the runs it takes to get a usable output, plus the minutes someone spends checking it, plus the cost of switching when a model retires. A model that costs five times more per run is cheaper if it saves one round of review.
Start with a read
Thirty minutes, six questions, one written observation. Bring the AI tools you pay for. No pitch until the last two minutes.
Book a free strategy evaluation
Prices cited are OpenAI’s published list rates, checked September 29, 2026: GPT-6 Sol and Luna, GPT-6 Astra, model retirements. The July Sol price is from DataNorth. The worked example is in Return on Token Investment. Prickled takes no commission, referral fee or incentive from any vendor named here.