Methodology
How RoleCraft measures AI's impact on your job.
What we look at, where the data comes from, and how it becomes your Career Audit.
We start with what you do, not what you're called.
Two people with the same job title can spend their weeks very differently.
So RoleCraft doesn't stop at your title. It looks at the tasks that make up your work, one by one.
That is where AI makes its difference. And that is where your audit starts.
From your job title to your audit.
- We match your job. You tell us your job title. We match it to one of more than 3,000 occupations in ESCO, the European classification of occupations and skills. You see the three tasks that define the match. If it doesn't feel like your job, you choose another.
- You make the task list yours. We show the tasks that usually belong to your occupation. Keep what you do. Remove what you don't. Add what's missing, in your own words.
- We place each task. Each task goes into one of three lanes: AI-led (Automate), Human + AI (Augment) or Human-led (Anchor). The placement follows how people actually use AI at work today. The placement is measured, not guessed: real AI use from the Anthropic Economic Index, task ratings from Eloundou et al., presence and people work from O*NET's work activities, and ability exposure from Felten, Raj and Seamans.
- We score your runway. Your runway score, from 0 to 100, is how long before AI reaches most of your core tasks. It is set against the 2,441 occupations with a score; the typical occupation scores 63. The score combines observed AI use (Anthropic Economic Index), Microsoft's occupational AI data and Eloundou's task ratings into one number per occupation.
- You get six steps. Number your runway. Identify your depreciating assets. Reinvest in appreciating assets. Make your value visible & portable. Reprice early, and claim your stake. Future-proof your expertise.
Your hours come from how often each task occurs in your occupation, scaled to the week you give.
Under step 3, “What stays human in this occupation” lists the people and judgement activities from RoleCraft’s fixed set of seventeen that O*NET rates at an importance of 3 or more for your occupation, most important first. It is O*NET’s rating of how much the occupation relies on each activity, not a measurement of AI. It carries no hours and is not counted in any figure. Where O*NET does not rate your occupation, a closely related one’s ratings are used.
“On par” means a runway score within 4 points of the median of the 2,441 scored occupations (63). Of those occupations, 21% sit in that band (59 to 67), 39% score below it and 40% above.
The value of expertise is the premium a professional earns over entry pay, times years of experience. On EU-average pay it is €14,589 a year (EU professionals aged 30–49 earn €50,579, those under 30 €35,990) × 15 years = €220k; on the highest national pay levels, such as the Netherlands, it reaches up to €500k. Source: Eurostat Structure of Earnings Survey 2022 (earn_ses22_28), ISCO-08 group 2, enterprises with 10+ employees.
A few occupations are RoleCraft bespoke rather than ESCO: when no ESCO occupation fits a role, its task statements are written from current job profiles, laned by the same method, and marked as bespoke in the report.
How a task gets its lane.
Every task sits somewhere on two dimensions.
- C — Codifiability: how far the task can be specified and done by AI.
- T — Trust/judgement dependence: how much the result relies on human trust, taste or judgement.
| Codifiability | ||
|---|---|---|
| Trust/ | Low | High |
| High | AnchorHuman-led | AugmentHuman + AI |
| Low | Delegateneeds a human, but not you: VA, contractor or partnerInside Human-led today | AutomateAI-led |
We don't score tasks by hand. We measure both dimensions from the data. Codifiability comes from how people use AI for the task today, and what research says AI could do (Anthropic Economic Index; Eloundou et al.). Trust and judgement come from whether a task needs your presence, and how exposed its abilities are to AI (O*NET; Felten, Raj and Seamans). In your report, Automate is AI-led, Augment is Human + AI and Anchor is Human-led. A task no one has yet been seen handing to AI is also AI-led when research finds that AI alone halves the time it takes and nothing physical or personal stands in the way (Eloundou et al.). A small number of lane placements are editorial decisions by the author; we list them on request. Delegate is shown inside Human-led today.
Business models.
If you work for yourself, where your value sits depends on your business model.
| Model | Core value (highest weight) | Note |
|---|---|---|
| Freelance | Delivery quality, client relationships | Most exposed: delivery is often automatable |
| Productised service | Offer design, pricing | The packaging is the value |
| Product | Problem insight, distribution | Building itself is now cheap |
| Community | Presence, convening, trust | Almost nothing in the core should be automated |
| Platform | Matching quality, trust between sides | Protect trust mechanisms |
If you work for yourself, your audit asks what your business model is. Your answer marks the tasks at the core of your business: the work your model is valued for. In each lane, those tasks are listed first. In AI-led and Human + AI they carry the label “Core of your business”. When half or more of the hours at your core are AI-led, your audit adds one line: move from delivering the task to owning the outcome. Your answer never moves a task to another lane, and never changes your hours or your runway score.
Four questions. Four trusted sources.
- What is your job made of? Official descriptions of occupations, the tasks inside them and the skills they need. ESCO, European Commission · O*NET, US Department of Labor
- Where are people using AI today? Anonymised conversations with AI, mapped to work tasks. They show which tasks people hand to AI and which they do together with it. We check this against a second, independent study of AI use at work. Anthropic Economic Index · Microsoft Research, “Working with AI”
- Where could AI go next? Research into which tasks today's AI could technically support. It shows where more change may come, so you can prepare early. Eloundou et al., “GPTs are GPTs”, Science
- Where is demand for new skills growing? Job postings that show where employers are asking for new AI skills, and in which kinds of roles. Lightcast job posting data
Designed to work for you.
- You stay in charge of what matters most. If a mistake in a task could do real harm, tell us. That task is never marked AI-led. AI may help, but you stay in charge.
- Numbers from the data. Advice from people. The figures in your audit come from the data. The six steps and the recommendations are written by people.
- Your audit is yours. It is private to you. Your employer never sees it.
Get the most from your audit.
The more your task list reflects your real week, the more useful your audit becomes.
- Check the match. If the three tasks we show don't sound like your job, choose another occupation.
- Make the tasks yours. Official descriptions are a starting point. Remove what you don't do. Add what you do.
- Look at the date. AI use changes quickly. Your audit shows which data release it is built on.
Sources
ESCO v1.2.1 (European Commission) · O*NET 30.3 (US Department of Labor) · Anthropic Economic Index, release of 26 June 2026 (May 2026 data), compared with the August 2025 release · Working with AI (Tomlinson et al., Microsoft Research) · GPTs are GPTs (Eloundou, Manning, Mishkin and Rock, Science) · AI Occupational Exposure (Felten, Raj and Seamans) · Lightcast job posting data (US postings, 366 AI skills)
See what this means for your work.
Stay relevant and valuable. Free, private, no account.