← All use cases
Use case 13

Job builder

Guide a hiring manager from "I need a new role" to a levelled position description, a market match and a picture of the level. It runs without AI, so a licensed job catalogue never leaves the computer.

Vibe-coded tool No AI needed to run it For hiring managers and the remuneration team
Try itUse the real tool. Watch itFive steps, animated. Build itPrompts you can paste. Check the dataWhat is safe, and where. Take itDownload the finished file.

What it does and why it matters

A new role usually reaches the remuneration team as a job title and a number the manager has in mind. The position description comes later, written to fit the grade the manager hoped for. The team then spends its time arguing a role back down, not pricing it.

This tool turns that order around. The manager answers six questions about the work, two each for knowledge, complexity and accountability, and gets an indicative level before writing a word. The level is shown on your internal framework and on an external evaluation scale.

The manager then writes the position description against a fixed template. As they type, the tool flags wording that reads too senior or too junior for the level, such as "sets the strategy" in a Level 3 role. A flag is a prompt, not a block: the manager can reword it or keep it, and anything kept is listed for the remuneration team.

Next it compares the description with two catalogues, market jobs and your own internal roles, and shows the three closest in each with the shared words highlighted. This step uses word matching, not AI, which is the point: a survey provider's job catalogue is licensed, and here it stays on the computer.

The outputs are a Word document, plain text to paste into your HR system, a spider diagram of the role against the level above, and a summary for the remuneration team. Everything is marked as a draft for the team to confirm.

It was "vibe coded": you describe what you want to an AI assistant in plain English and let it write the code. Once built, it is ordinary code. It gives the same result every time and uses no AI when you run it.

1Try it

This is the real tool, running in your browser on an invented level framework, 127 invented market roles and 35 invented internal roles. Press "Load the example role" and step through. The example has a few phrases planted to set off the wording check.

The framework, the external level scale, both catalogues and the wording lists are invented. None of it comes from Aon or any other published method. If you load your own files, they are read by your own browser and nothing is uploaded.

2Watch it

Five steps from a new role to a draft the remuneration team can confirm. Press Play, or step through at your own pace.

Step 1 of 5Invented data
Complexity: what are the hardest problems the role solves regularly?
Problems are familiar. The role chooses between a few known options.
Problems vary. The role analyses them and adapts a known approach.
Problems are unusual and pull in different directions. The role develops new approaches within an established field.
Problems are new and cross several teams. The role must define the problem as well as solve it.

Answer six questions about the work. Two each for knowledge, complexity and accountability. The statements run from least to most demanding and carry no level numbers, so the manager describes the work and cannot simply pick a grade.

Remuneration Analyst: six answers
3, 3Knowledge
+
3, 3Complexity
+
3, 4Accountability
÷ 6 = 3.17
Level 3external evaluation level E14

Get an indicative level. The average of the six answers, with halves rounded down, kept within the levels open to the role's career stream. A conversion table turns it into an external evaluation level. The manager can move it one level either way, with a reason that goes to the remuneration team.

Position description, written for Level 3
Supports the annual remuneration review and keeps pay ranges current. Sets the remuneration strategy for the group. Presents pay outcomes to the Board. Data entry for the bonus file.
"Sets the remuneration strategy" reads too senior for Level 3. Usually seen from Level 7 up.
"the Board" reads too senior for Level 3. Usually seen from Level 6 up.
"Data entry" reads too junior for Level 3. Usually seen up to Level 2.

Write, and be challenged. The tool checks each section against lists of wording typical for each level. It catches telltale words, not meaning, so it is a prompt to think again and not a verdict. The manager can reword a phrase or keep it, and kept phrases are listed for the remuneration team.

Closest market matches, People family, within one level
Strong matchRemuneration Specialist (Proficient)E14, about Level 3. Shared words: remuneration pay salary survey ranges
Strong matchRemuneration Specialist (Senior)E16, about Level 4. Shared words: remuneration pay salary survey ranges
Weak matchHR Business Partner (Proficient)E14, about Level 3. Shared words: partner people change

Match without AI. The tool scores how many meaningful words the description shares with each catalogue role, giving rarer words more weight. It shows the top three in the market catalogue and, separately, in your internal roles. The manager picks one, or says there is no good match.

This role against a typical Level 4 role
To be a Level 4 role, this job would need:
  • Deep knowledge of one specialist field, so that others come to the role for answers.
  • Problems that are unusual and need new approaches.
  • Freedom to set its own priorities within agreed objectives.

See the level above. The solid shape is this role's six answers. The dashed shape is a typical role one level up. The list underneath says what would have to change in the work, which is a better conversation than "why can't it be a 4?".

Everything in this walkthrough is invented and matches the tool's built-in example role.

3Build it

You can build your own version with any capable AI assistant your employer allows, for example Microsoft Copilot, Claude or ChatGPT. You describe what you want, it writes the code, and you check the answers.

Build it with an invented framework and invented job descriptions. Never give the AI assistant a survey provider's job catalogue or evaluation method: load those into the finished tool yourself. Expect several rounds of "that's not quite right, change X". This is a bigger build than most on this site.

Step 1: Describe the tool

Say what the tool is, who uses it, and the order of the steps.

I lead a remuneration team in Australia and I am not a developer. Please build me a "job builder" for hiring managers as a single HTML file that I can save to my computer and open in a browser by double-clicking it. Requirements: - It must run offline with no AI. Use no external libraries, fonts, images or links, and send no data anywhere. Draw the chart yourself. - It is a six-step guide: (1) role type, (2) level, (3) write the position description, (4) match to job catalogues, (5) a picture of the level, (6) outputs. Let the user move back and forward freely. - Step 1 asks for job title, who the role reports to, job family and career stream. Use four streams: Support, Professional, Management, Executive. - Build in an invented internal framework of 8 levels. For each level write a name, a one-line summary, the streams it is open to, and one statement for each of six questions: knowledge depth, knowledge breadth, complexity of problems, guidance available, freedom to act, and reach of impact. Invent these yourself. Do not copy any published job evaluation method. - Build in an invented conversion table from each internal level to an "external evaluation level", and say on screen that it is invented. - Build in two invented job catalogues: about 100 market roles and about 30 internal roles. Each has a job code, title, job family, sub-family, level and a two-sentence description. Market roles use the external level scale; internal roles use the internal level. - Let me replace the framework, the conversion table and each catalogue by loading a CSV file, and download the current one as a template. - Mark every result "Indicative, for Reward to confirm". - Use plain English and Australian spelling. Keep the design clean. Before you write any code, tell me in a few sentences what you plan to build, and ask me about anything that is unclear.

Step 2: Spell out the level and the wording check

Do not let the AI guess the rules. Tell it exactly how the level is worked out and what gets flagged.

Now add the levelling and the position description. Follow these rules exactly. 1. Level questions: show the six questions grouped under Knowledge, Complexity and Accountability. For each, list the statements from every level, lowest first, as options. Do not show level numbers next to the options. The answer's score is the level its statement belongs to. 2. Indicative level = the average of the six scores, rounded to the nearest whole level, with exact halves rounded down. If that level is not open to the chosen career stream, use the nearest level that is, and say so. 3. Show the internal level, the external level from the conversion table, the average for each of the three factors, and a one-line explanation of the rule. 4. Also ask for direct reports, total team size, the budget the role decides, and the biggest decision the role makes alone. These go on the summary for the remuneration team, not on the position description. 5. Let the manager move the level one step up or down. If they do, ask for a written reason. 6. Position description template, in this order: Purpose, Objectives, Technical capabilities, Qualifications, Example tasks. Purpose is a short paragraph; the others are one item per line. Add a short header: job title, job family, career stream, level, reports to. 7. Wording check: build a list of about 60 words and phrases, each with the lowest and highest level where it is usual. Examples: "sets the strategy" from Level 7 up; "the board" from Level 6 up; "leads a team" from Level 4 up; "data entry" up to Level 2; "assists with" up to Level 3. As the manager types, flag any phrase that falls outside the role's level, say whether it reads too senior or too junior, and show wording typical for the level. Flag a number of years of experience that is unusual for the level in the same way. 8. A flag never blocks. Give each flag a "Keep anyway" button, and remember which flags were kept. 9. Beside the writing area, show the level's summary, wording typical for it, and a sentence pattern for a purpose, an objective and a task. Say on the page that the wording lists catch telltale words, not meaning.

Step 3: Add the matching, the picture and the outputs

Name the matching method, so you get something you can explain.

Now add the last three steps. Matching (no AI): - Compare the manager's job title and position description with every role in a catalogue using TF-IDF and cosine similarity: split the text into words, drop common words, reduce plurals and verb endings to a common form, weight rarer words more, and count title words double. - Show the three closest roles in the market catalogue and, separately, the three closest internal roles. For each show the title, code, family, level, the description with shared words highlighted, and a list of the shared words. - Rate each as a strong, moderate or weak match in words. Do not show a percentage. If nothing scores above the weak cut-off, say there is no match. - By default search only the same job family and roles within one level of the role. Add a "Search everything" tick box. - Convert market roles' external levels to internal levels with the conversion table before comparing levels. - If the closest role in the whole catalogue is two or more levels away and scores clearly higher than anything near the role's level, warn that the level or the wording may be off. If a clearly closer role sits in another family, say so. - The manager chooses one match in each list, or "No good match, refer to Reward". Show no pay figures. Level picture: - Draw a six-point spider diagram of the six scores. Overlay a dashed shape for a typical role one level up (all six points at that level), and offer a tick box to add the level below. - Under it, list what would have to change for the role to be the next level up: for each question scored below that level, show the next level's statement. - Add a side-by-side table and a button to download the picture. Outputs: - The position description as a Word document (.docx) and as plain text to copy, whole or section by section. - A plain-text summary for the remuneration team: the level and how it was reached, the six answers, any change the manager made and their reason, every wording flag and whether it was kept, and the matches with the manager's choice. - A progress file the manager can download and load back in later. It must hold the answers and text only, never the catalogues.

Step 4: Protect the data and make it prove itself

Lock down where data can go, then test the level rule with answers you can work out by hand.

Now set the rules for the data, then help me test. Data rules: - Do not save anything. Do not use cookies, local storage or any network request. Nothing may leave the computer. - Show a clear line near the top: "Everything runs in your own browser, with no AI. Nothing is uploaded or saved." - Show a second notice: "Job catalogues from survey providers are licensed. Check your licence before sharing this screen or its downloads." - Label the built-in framework and catalogues as invented wherever they are shown. - Treat every loaded file and every typed word as text to display, never as code. - If a file is missing a needed column, say which column in plain English and load nothing. - List every place in the code that could store or send data, and confirm there are none. Tests. Tell me what the tool should show for each before I try it: 1. Six answers of 3, 3, 3, 3, 3, 4 in the Professional stream. 2. Six answers of 3, 3, 4, 4, 4, 3 in the Professional stream. 3. Six answers of 4, 4, 4, 4, 3, 4 in the Professional stream. 4. Six answers of 1, 1, 1, 1, 1, 1 in the Professional stream, which starts at Level 2. 5. A Level 3 role whose purpose says "Sets the remuneration strategy" and whose tasks include "Data entry for the bonus file". If any result does not match a hand calculation, say so plainly and fix the rule. Do not adjust the test to make it fit.

The answers you should get

Invented example
Six answers, Professional streamAverageIndicative level
3, 3, 3, 3, 3, 43.17Level 3
3, 3, 4, 4, 4, 33.50Level 3
4, 4, 4, 4, 3, 43.83Level 4
1, 1, 1, 1, 1, 11.00Level 2
An exact half (3.50) rounds downLevel 3
"Sets the remuneration strategy" at Level 3Too senior
"Data entry" at Level 3Too junior

Two rows test the edges. The second lands exactly halfway and must round down, so a role has to be clearly in the higher level to get it. The fourth scores Level 1, which the Professional stream does not have, so the tool should give Level 2 and say why.

If the tool gives a different answer, it is wrong. Tell the AI what you expected and ask it to fix the rule.

Good habits

  • Describe the role, never the person. The questions are about the work when it is done properly, not about who you have in mind for it.
  • Treat every result as a draft. The tool gives managers a structured starting point; the remuneration team still decides the level and the match.
  • Test the wording lists on ten real position descriptions you have already graded, and adjust the lists where they flag the wrong things.
  • Match on what the job does, not its title. Titles vary far more than the work.
  • Keep the prompts you used so you can rebuild or audit the tool later.

4Check the data

What this touches

Position descriptions for roles, your level framework, and job catalogues. It is about roles, not people, so it holds no employee records. A market job catalogue and an evaluation method from a survey provider are licensed, and the licence sets where they can go.

What is safe

The finished tool runs in your browser, uses no AI and uploads nothing, so a licensed catalogue loaded into it stays on your computer. The progress file holds the manager's answers and text, not the catalogues. When you build your own, give the AI assistant invented material only.

Ask before you start

  • What does our survey licence say about loading the job catalogue into our own tools?
  • Who here is allowed to see the catalogue and the evaluation method?
  • May hiring managers see internal position descriptions for other teams?
  • Is the AI tool we would use for polishing approved for draft position descriptions?

The tool has a "copy a prompt" button for polishing the wording in an AI assistant. That prompt contains the manager's draft and the level description, not the catalogues, but it does leave the tool, so only paste it into an assistant your organisation has approved. A draft for a role that is not yet public, such as one tied to a restructure, can be sensitive in its own right. General information, not legal advice.

5Take it

Download the tool Open it in a new tab

It is a single file. Save it anywhere and double-click to open it. The framework, the conversion table and both catalogues can be downloaded from inside the tool as templates for your own.

What next: to size a role with an AI assistant and get a written reason for every call, see job evaluation. To turn a matched role into a pay figure, see the pay recommendation, or go back to all use cases.