Your AI plan is budgeted to the wrong number
Get the free 6-page report that shows you which number to use instead
Three credible sources rate AI’s capability in computer and mathematical jobs at 94%, 33%, and 78.9%. All three are honest, and only one of them belongs in next year’s budget.
Free. The PDF arrives the moment you submit. No sales call, no demo request.
You already ran the pilots. Now you have to defend the next budget.
You have spent real money on AI: tokens, subscriptions, training, and outside help.
Your stakeholders are asking two questions at once. Looking backward, they want to know where the money went and what came back. Looking forward, they want to know what deserves more budget, what deserves less, and where the measurable results are. A vendor’s theoretical capability number answers neither question. Most of the public numbers you can reach for come from AI companies measuring their own chat logs, which only ever show what someone already thought to type into a chat window.
We built the report differently. We took real job postings from 28 companies and broke 3,234 distinct roles into their individual tasks. Then we scored every one of those tasks for AI capability, whether or not anyone has ever pointed an AI tool at that work.
WHAT THE PAPER IS ABOUT
- The three-number breakdown. Why 94%, 33%, and 78.9% describe the same jobs, what each one actually measured, and which one belongs in a budget model.
- How the 94% was made. The 2023 academic study scored 18,000 written task statements from O*NET using human and GPT-4 judges. Nobody handed a task to an AI and timed it. The report explains what that means for your planning.
- The roles that show zero AI usage. Anthropic’s data shows 30% of workers registering no measured AI usage at all. The report walks through cooks, lifeguards, and motorcycle mechanics to show why that reads as a measurement gap.
- The TRIPS scoring rubric, in full: Time, Repetitiveness, Importance, Pain, and Sufficient Data, plus a three-question version you can run on one task this afternoon.
- The augmentation number, disarmed. 93.2% of all 90,183 tasks route to augmentation rather than full automation. The report shows why that number is a policy artifact and should not anchor a plan.
- Where the ceiling is wrong in both directions. Our scores run higher than the academic ceiling in 5 of the 22 occupation categories we compared, and lower in the other 17.
What you can do with it
Walk into the budget conversation with a number you can defend
- Explain to a CFO why two published AI statistics differ by 61 points without either one being wrong.
- Anchor your forecast to present-day capability rather than a 2023 academic ceiling or a vendor’s usage log.
- Cite a source that has no reseller relationship with Anthropic or any other model vendor.
Plan against the 94% ceiling and you overbuild. Plan against the 33% usage figure and you underbuild while your competitors do not.
Find the opportunity in roles you already wrote off
- See why a job with zero recorded AI usage still has a narrow, real use case sitting just outside its core work.
- Get concrete examples across frontline and skilled-trade roles, including the computer-vision drowning-detection case for lifeguards.
- Understand why the theoretical ceiling oversells the office and undersells the shop floor.
The departments your team has already dismissed are the ones nobody has measured properly.
Score your own team’s tasks this week
- Run the three-question starter test on one weekly task: does it eat real time, does someone dread it, and do you have enough past examples to recognize a wrong answer?
- Use the full five-dimension TRIPS rubric to turn a to-do list into a rollout order.
- Weigh how much damage a wrong answer does before you hand any task to an AI.
Start with the work your team dreads. People adopt tools fastest when those tools remove work they never enjoyed.
Read this if
- You own an AI mandate at an operating company and need evidence before you commit to a bigger program.
- Someone has asked you what your AI strategy is, and a vendor demo is not an acceptable answer.
- You already have AI tools deployed and want a rigorous way to prioritize what comes next.
- You are heading into a budget-planning cycle and need the AI line item to survive scrutiny.
This report is not
- An AI 101 explainer. It assumes you have already run pilots and read the vendor decks.
- An analysis of your company. It uses 28 other companies’ public job postings, which means the task examples come from employers like Thermo Fisher Scientific and Turner Construction rather than from the roles on your own org chart.
- A tool recommendation. The report never tells you which AI product to buy.
MORAL OF THE STORY
Every number about AI capability is answering a specific question, and most of them are not answering yours. The 94% ceiling tells you what a panel of judges believed was theoretically possible three years ago. The 33% usage figure tells you what people have thought to type into a chat window so far. Neither one knows what your team does on a Tuesday. That answer comes from your own task list, scored honestly, one row at a time. Start with the work your team dreads, and the rest of the roadmap builds itself.
Get the report
Your AI plan is budgeted to the wrong number. Here is the right one.
Six pages. Three numbers explained. One scoring rubric you can run on your own team this week.
What Jobs Can AI Really Do? A Trust Insights Analysis
"*" indicates required fields
