← All use cases
Use case 12

Market movement tracker

Load several years of market data and see how pay has moved for each job family, level and job code. Find the biggest movers up and down, and decide how much to trust the small samples.

Vibe-coded tool No AI needed to run it For 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

Every year a new market survey arrives, and the first question is always the same: what moved? Answering it means lining up this year's file against last year's, job code by job code, and most teams do that by hand in a spreadsheet.

This tool does the lining up for you. Load one file per survey year and it shows how the median has moved for the whole market, for each job family and level, and for every job code. It tracks total fixed remuneration (TFR): base salary plus cashable benefits plus superannuation.

You can look at the latest year, the change since the first year, or the average per year. A chart shows every family on the same scale, a heat map shows which levels moved in which year, and two lists show the biggest movers up and down. Click anything to drill down to the job codes behind it.

It also deals with the problem that catches people out: small samples. A job with three incumbents can jump 30% in a year because one person changed, not because the market moved. You set the minimum number of incumbents and organisations, then choose to show, flag or leave out anything below it.

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 runs the same calculation every time and uses no AI when you run it.

1Try it

This is the real tool, running in your browser on five years of invented market data for 239 job codes. Switch "Small samples" between Show all and Exclude and watch the biggest movers change. Click a line, a row or a cell to drill down.

Every job code and figure is invented. If you load your own files, they are read by your own browser and nothing is uploaded.

2Watch it

Five steps from a stack of survey files to knowing what moved. Press Play, or step through at your own pace.

Step 1 of 5Invented data
market-data-2026.csv (one of five yearly files)
job_codejob_titlejob_familyleveltfr_medianincumbentsorganisations
TEC-SWE-4Software EngineerTechnologyLevel 4$162,40027040
DAT-DSC-5Data ScientistData & AnalyticsLevel 5$234,9006316
MKT-BRD-4Brand ManagerMarketingLevel 4$118,3004315
RSK-MDL-6Model Risk SpecialistRiskLevel 6$264,90033

Start with a file per year. One row per job code, with the median and the two sample sizes: how many people and how many organisations sit behind the figure. Select all five files at once and the tool stacks them.

One job code: Software Engineer, Level 4
$162,400Median TFR, 2026
÷
$156,600Median TFR, 2025
− 1 =
+3.7%movement, 2025 to 2026

Work out each job's movement. This year's median divided by last year's, for the same job code. The tool does this for every code that appears in both years, and for any pair of years you choose.

Average movement of each family's job codes, 2025 to 2026
Data & Analytics
+7.8%
Technology
+4.5%
Risk
+3.8%
Customer
+0.5%
Marketing
−1.4%

Roll up to job families. A family's figure is the average of its job codes' movements, using only codes found in both years. That way a job being added to or dropped from the survey is not mistaken for a pay movement. Marketing, in gold, fell.

Biggest rises, 2025 to 2026, with every job code shown
Model Risk Specialist, L63 incumbents Small sample
+33.1%
Agile Coach, L64 incumbents Small sample
+14.7%
AI Engineer, L510 incumbents
+12.8%
Data Leadership, L629 incumbents
+12.4%

Watch the small samples. The two biggest "movers" rest on three and four people. With so few, one person joining or leaving swings the median. The tool flags any job code under your minimum, here 10 incumbents or 5 organisations.

Biggest movers, 2025 to 2026, with small samples left out
Moved up the most
AI Engineer, L5
+12.8%
Data Leadership, L6
+12.4%
Finance Leadership, L8
+10.0%
Moved down the most
Risk Leadership, L7
−4.4%
Marketing Leadership, L7
−4.4%
Content Producer, L2
−4.0%

Find the real movers. Leave out the small samples and the list changes: 23 job codes drop out and the picture is now Data & Analytics rising and Marketing falling. Click any job to see its full history, year by year.

All figures in this walkthrough are invented and match the tool's built-in sample at its starting settings.

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 invented data, never your real survey files. Expect two or three rounds of "that's not quite right, change X".

Step 1: Describe the tool

Say what the tool is, what goes in, and who will use it.

I work in a remuneration team in Australia and I am not a developer. Please build me a "market movement tracker" 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. Use no external libraries, fonts, images or links, and send no data anywhere. Draw the charts yourself. - The user loads several CSV files at once, one per survey year, or one file holding every year. Each row is one job code in one survey year. Needed columns: job_code, job_family, survey_year, tfr_median. Optional columns: job_title, level, incumbents, organisations, tfr_p25, tfr_p75. If a file has no survey_year column, read the year from a four-digit year in the file name. - A job code is one job at one level and keeps the same code every year. tfr_median is the median total fixed remuneration in annual dollars: base salary plus cashable benefits plus superannuation. - Build in an invented sample of five survey years and about 250 job codes across 10 job families and 8 levels, with believable patterns: one family rising fast, one falling, one flat, a few job codes that appear or drop out part-way, and job codes with very few incumbents whose medians bounce around from year to year. Add buttons to download the sample, one file per year. - Wherever levels are listed, put the most senior at the top. Work out seniority from pay, not from the level's name. - The audience is remuneration specialists. Use plain English and Australian spelling. Keep the design clean, with large readable headline numbers. 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 calculations

Do not let the AI guess the maths. Tell it exactly what to calculate.

Now add the calculations. Follow these definitions exactly. Do not substitute your own. 1. Job code movement between two years = (median in the later year ÷ median in the earlier year) − 1, shown as a percentage to one decimal. If the job code is missing from either year, it has no movement for that pair of years. 2. Movement measure: a three-way switch. "Latest year" compares the last two survey years. "Since first year" compares the first and last years. "Average per year" uses the first and last years and shows the steady yearly rate that gives the same total change: (later ÷ earlier) to the power of (1 ÷ number of years between them), minus 1. 3. Group movement (a job family, a level, or the whole market) = the average of the movements of its job codes, using only job codes present in both years. Count each job code equally. Add a switch to weight by the number of incumbents in the later year instead. Never compare the average pay of everything in one year with everything in another, because a change in which jobs were surveyed would look like a pay movement. 4. Small samples: two settings I can edit, minimum incumbents (start at 10) and minimum organisations (start at 5). A job code's movement is a small sample if it is under either minimum in either of the two years compared. Exactly on the minimum is not small. Add a three-way switch: "Show all" (no flags), "Warn" (flag them, still count them in the averages), and "Exclude" (leave them out of every average, count and list, and say how many were left out). 5. Notable movement: a setting I can edit, starting at 5%. Count the job codes that rose by that much or more, and those that fell by that much or more. Exactly on the threshold counts. 6. Overview: the headline market movement; a line chart of every job family as an index starting at 100 in the first year, built by linking each year's movement to the year before; a table of movement by job family or by level; and the biggest movers as two lists, the top 10 up and the top 10 down, ranked by percentage, switchable between job codes and job families. 7. Drill-down: clicking a job family or level shows it against the whole market, a heat map of year-on-year movement (levels down the side, pairs of years across the top), and a sortable table of its job codes. Clicking a job code shows its median for every year, the 25th to 75th percentile band if those columns were loaded, its sample sizes, and its change on the year before. 8. Every chart, table and figure must recalculate when any setting changes. 9. Add buttons to download any table as a CSV, and a print-ready one-page summary. Explain each definition in one plain-English sentence on the page.

Step 3: Protect the data

Survey data is licensed, so the tool should keep it on your computer and remind people of that.

Now set the rules for the data. - 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: "Your files are read by your own browser. Nothing is uploaded or saved." - Show a second notice: "Survey data is usually licensed. Check your survey licence before sharing this screen or its downloads." - Label the built-in sample clearly as invented wherever it is shown. - In every CSV download, put an apostrophe in front of any text that starts with =, +, - or @, so a spreadsheet cannot run it as a formula. - If a file is missing a needed column, say which file and which column in plain English, and load nothing. When you finish, list every place in the code that could store or send data, and confirm there are none.

Step 4: Make it prove itself

Load a tiny file where you can work out every answer by hand.

Please help me check that the calculations are right. 1. Create a test CSV with these rows (job code, family, year, median, incumbents, organisations): A1, A, 2024, 100000, 50, 10 A1, A, 2025, 110000, 50, 10 A1, A, 2026, 121000, 50, 10 A2, A, 2024, 100000, 20, 8 A2, A, 2025, 100000, 20, 8 A2, A, 2026, 95000, 20, 8 A3, A, 2024, 80000, 4, 3 A3, A, 2025, 80000, 4, 3 A3, A, 2026, 104000, 4, 3 A4, A, 2025, 50000, 30, 6 A4, A, 2026, 51000, 30, 6 B1, B, 2024, 200000, 100, 20 B1, B, 2026, 220000, 10, 5 2. Before I load it, tell me what the tool should show for the whole market: the latest-year movement with small samples shown, with small samples excluded, and weighted by incumbents; the movement since the first year; the average per year; and the number of notable rises and falls at 5%. 3. If any result does not match a hand calculation, say so plainly and fix the calculation. Do not adjust the test data to make it fit.

The answers you should get

Invented example
Whole marketJob codes comparedMovement
Latest year (2025 to 2026), small samples shown4+9.3%
Latest year, small samples excluded3+2.3%
Latest year, weighted by incumbents4+5.6%
Since first year (2024 to 2026)4+14.0%
Average per year4+6.6%
Latest year: 10%, −5%, 30% and 2%, averaged+9.25%
Notable rises at 5% (A1 and A3)2
Notable falls at 5% (A2, exactly −5%)1

Three rows are there to test the edges. A3 has only 4 incumbents, so it is the small sample, and leaving it out drops the answer from 9.3% to 2.3%. B1 has exactly 10 incumbents and 5 organisations in 2026, which is on the minimum and so not small. B1 is also missing in 2025, so it counts in "since first year" but not in "latest year".

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

Good habits

  • Check that job codes mean the same thing each year. If the survey provider re-levelled or renamed a job, the movement is not real.
  • Write down your definitions (which pay figure, which percentile, which minimum sample) before you compare two reports.
  • Look at a job code's full history before acting on one year's jump.
  • Keep the prompt you used so you can rebuild or audit the tool later.
  • Get someone else to check the numbers before anything goes to an executive or board.

4Check the data

What this touches

Market survey data: medians, percentiles and sample sizes by job code. It holds no employee records and no personal information. It is usually licensed, and the licence sets who can see it and where it can go.

What is safe

The finished tool runs in your browser and nothing is uploaded. Keep it and its downloads with the people your licence covers. When you build your own, give the AI assistant invented data only, never an extract from the survey.

Ask before you start

  • What does our survey licence say about sharing and AI tools?
  • Who here is allowed to see the survey data?
  • Where will the files and any downloads be stored?
  • Is this AI tool approved for licensed data?

Small samples matter for a second reason: a figure resting on a handful of people at a few organisations can come close to revealing what a single employer pays, which is why survey providers set their own minimums. This tool is market data only. If you later add your own pay data to compare against the market, it becomes employee data and the rules change: see the position in range analyser for how that is handled. 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 sample files can be downloaded from inside the tool.

What next: to see how your own people sit against their reference points, try the position in range analyser, or go back to all use cases.