← All use cases
Use case 06

Remuneration analyser

Ask questions of your real pay data in plain English and get answers that show their working. A blueprint for building it securely with your IT team in Microsoft Foundry (previously Azure AI Foundry).

Build with IT Microsoft Foundry Blueprint
Try itA scripted conversation. Watch itFive steps, animated. Build itA plan to take to IT. Check the dataWhat is safe, and where. Take itDownload the IT brief.
Status: this is a blueprint written from Microsoft's documentation. We have not yet built or tested it. The conversation below is a scripted illustration, not output from a real build.

What it does and why it matters

This is a remuneration health check assistant, built inside your organisation's own Microsoft Azure environment with your IT team. People in the reward team ask questions of the real pay data in plain English and get an answer with the working shown. Microsoft Foundry is the product it would be built in.

One design rule matters most: the AI does not do the maths. The AI turns the question into a query, ordinary code runs it over the data, and the answer comes back with the figures and the query, so a person can check it. Language models are good at understanding a question and unreliable at arithmetic, so the design keeps them apart.

Use this route when you need real, identifiable pay data, answers must be checkable and logged, access must be limited to named people, and it must stay inside your own environment. Do not use it for exploring ideas with invented data (use the pay gap analyser, use case 01) or for policy questions (use the team assistant, use case 05).

1Try it

Press Play to watch a short conversation with the analyser, or Show all to read it. This is a script, not a live system.

Replay Scripted illustration using the invented workforce from use case 01. Not output from a real build.

Scripted illustration using the invented workforce from use case 01. Not output from a real build. You can check the same figures in the live analyser in use case 01.

2Watch it

Five steps from a question to an answer you can check. Press Play, or step through at your own pace.

Step 1 of 5Invented data
Chat window
What is our gender pay gap on total remuneration, average and median?

Signed in with your work account. Nothing has been calculated yet.

You ask in plain English. No query language, no spreadsheet formulas. Just the question, as you would ask a colleague.

What is our gender pay gap on total remuneration, average and median?
Query written by the AI
SELECT gender,
       AVG(total_remuneration)    AS average_pay,
       MEDIAN(total_remuneration) AS median_pay
FROM   governed_pay_view
GROUP  BY gender

The AI writes a query. It has not touched the numbers. It has only turned your words into instructions that a database can follow.

governed_pay_view (a few of the rows)
employee_idgenderleveltotal_remuneration
E1042WLevel 3$112,300
E1043MLevel 3$118,900
E1107WLevel 4$149,500
E1118MLevel 2$93,800
E1190MLevel 5$201,400
$107,860 average total remuneration for the 139 women in the full data (invented)

Code runs it on your data. Ordinary code does the arithmetic, the same way every time. The AI is not involved in this step.

Reward team only
Results by level
GroupWomenMenGap
Level 420244.6%
Level 59123.5%
Level 626(calculated)
Hidden: small group

Guardrails apply. Small groups are replaced before the result reaches the AI or you. Only the named reward team group can ask at all.

13.3%average gender pay gap, total remuneration (invented)

Men $124,423, women $107,860. Median gap 13.0%.

Show working
SELECT gender,
       AVG(total_remuneration)    AS average_pay,
       MEDIAN(total_remuneration) AS median_pay
FROM   governed_pay_view
GROUP  BY gender

(124,423 − 107,860) ÷ 124,423 = 13.3%

An answer you can check. The figures and the query arrive together. A person can run the same query, or redo the sum, and confirm it.

All figures in this walkthrough are invented, for illustration only. Row values are samples.

3Build it

Unlike the other use cases, you do not build this one yourself. You build the plan, and your IT team builds the system. This section gives you the picture to show them, plain-English descriptions of each part, the decisions you need to make together, a technical panel for them, and a note to send with the brief.

Microsoft's product names and features change quickly. Every Microsoft statement on this page links to the page it came from and is marked checked November 2026. Where we could not verify something, it is a question for your IT team, not a claim.

The architecture, in one picture

Read it left to right. Everything inside the dashed line stays inside your organisation's Azure environment.

Inside your organisation's Azure environment People in thereward teamSigned in withtheir work account Chat interfaceWhere the questionis typed AI agentUnderstands thequestion and writesthe query Query orcode toolRuns the query.Code does the maths Governed pay datasetSmall groups hiddenEmployee IDs, not names Logging and reviewQuestions, queries and answers recorded, so they can be reviewed People in the reward teamSigned in with their work account Inside your organisation'sAzure environment Chat interfaceWhere the question is typed AI agentUnderstands the question, writes the query Query or code toolRuns the query. Code does the maths Governed pay datasetSmall groups hidden. IDs, not names Logging and reviewQuestions, queries and answersrecorded, so they can be reviewed

The same diagram as a list, for screen readers and for printing:

  1. People in the reward team sign in with their work account. They sit outside the boundary.
  2. Chat interface: where the question is typed and the answer is shown.
  3. AI agent: understands the question and writes the query. It does not calculate.
  4. Query or code tool: runs the query. Ordinary code does the maths.
  5. Governed pay dataset: small groups are hidden here, and employee IDs are used instead of names.
  6. Logging and review sits underneath all four parts and records questions, queries and answers.
  7. The boundary is your organisation's own Azure environment. It surrounds everything except the people.

What each part does

In plain English, with what Microsoft's documentation says about each one.

Sign-in and access: who gets in

People sign in with their normal work account, and access is limited to a named group. Microsoft Foundry uses Microsoft Entra (Microsoft's sign-in and identity service) and role-based access control, which lets you control who can create, run and manage agents.

Source: What is Foundry Agent Service, checked November 2026.

Chat interface: where people ask

This is the screen people type into. Your IT team chooses it. Microsoft's documentation describes publishing an agent to a stable endpoint (a fixed address other software can call) and sharing it through Microsoft Teams and Microsoft Copilot.

Source: What is Foundry Agent Service, checked November 2026.

AI agent: understands the question

An agent is an AI model paired with instructions and tools. It reads the question, decides what to ask for, and calls a tool. Your instructions tell it to write queries, show its working, and refuse certain requests. Microsoft calls the simplest kind a prompt agent: you set instructions, a model and tools, and Foundry runs it.

Sources: What is Microsoft Foundry and Agent Service overview, checked November 2026.

Query or code tool: where the maths happens

The tool runs what the AI wrote. Microsoft describes two options that fit. A code interpreter tool runs Python code in a sandbox (a sealed-off space) over files you supply. Custom tools let the agent call functions or services your team writes. Which is better depends on whether the tool should ever see individual rows. See the IT panel below.

Sources: Code Interpreter and Agent Service overview (tools and toolboxes), checked November 2026.

Governed pay dataset: what the tool can see

This is your data, prepared for this use. It holds employee IDs instead of names, and it hides small groups before anything reaches the AI. This part is your organisation's design, not a Microsoft feature. Microsoft's documentation says Agent Service can store files and conversations in your own Azure resources instead of Microsoft-managed ones, which matters if you need full data ownership.

Source: Use your own resources in Agent Service, checked November 2026.

Logging and review: the record

Microsoft's agent tracing captures user inputs, agent outputs, tool use and results, token use and timing. Traces are stored in Azure Monitor Application Insights. Microsoft warns that traces can contain sensitive information, so the logs need the same access controls as the data.

Source: Agent tracing overview, checked November 2026.

Safety filters: a different job

Foundry has guardrails that scan prompts and answers for harmful content and prompt attacks (attempts to trick the AI). These are useful, but they are not the small-group rule. Hiding small groups has to happen in the data. Microsoft marks guardrails for agents as preview.

Source: Guardrails and controls overview, checked November 2026.

The five decisions to make with IT

These are business decisions as much as technical ones. Settle them in a meeting before anyone builds anything.

  1. Where the data lives and how it is refreshed

    Which system holds the governed pay dataset, who prepares it, how often it is refreshed, and who can see it. Decide which fields go in and which stay out.

  2. Who gets access

    Which named group, who approves membership, and how often it is reviewed. Decide whether everyone in the group sees the same things.

  3. What the assistant may and may not answer

    Your small-group threshold, whether it may ever list individual employee IDs, and what it must refuse (for example legal advice or policy questions). Write these down so they can be tested.

  4. How answers are checked

    A test set of questions with known answers, worked out separately by a person. The assistant must get them all right, with its working shown, before it is used. The file sample-questions.md is a starting point.

  5. Who owns it after launch

    Who fixes it, who updates it when the data or the product changes, who reviews the logs, and who retires it. Without a named owner, it will drift.

For your IT team

The technical version

Verified capabilities, each with its source and marked checked November 2026. Names changed during this research: Microsoft's pages say the brand moved from Azure AI Studio and Azure AI Foundry to Microsoft Foundry, and the old Assistants-style agent API to the Responses API (Agents v2) with conversations, items and responses replacing threads, messages and runs. Confirm current names before you build.

What Microsoft documents

  • Agent types and tools. Foundry Agent Service offers prompt agents (configuration only), voice-based prompt agents and hosted agents (your own code in a managed container). Built-in tools include web search, file search, code interpreter and memory. Custom tools can be functions, OpenAPI specs and MCP servers. Tools can be grouped in a toolbox exposed as one managed MCP endpoint. Agent Service overview
  • Code interpreter. The model writes and runs Python in a sandbox. Files such as CSV and XLSX can be supplied. Microsoft says the sandbox runs in the same Azure region as the Foundry project and cannot make outbound network requests, and that it has additional charges beyond token fees. When code interpreter is used through a toolbox in a hosted agent, user isolation is not supported and all users in the same project share one container context. Code Interpreter
  • Authentication to tools. Options include key-based access, Microsoft Entra using the agent's or project's managed identity, and OAuth identity passthrough (on-behalf-of). Agents can have a dedicated Entra identity. Agent Service overview
  • Access control and network. Role-based access control through Microsoft Entra and Azure RBAC. Private networking for prompt agents, and bring-your-own virtual network for hosted agents. Agent Service overview
  • Your own storage. By default Agent Service manages storage for files, conversations and vector stores. A standard setup can use your own Azure Storage, Azure Cosmos DB and Azure AI Search so that files, conversations and vector stores are in your own resources. Use your own resources
  • Regions and residency. Australia East is listed as a region where a Foundry project can be created. Microsoft says model and tool availability varies by region and that its list is a snapshot to verify. Feature availability across regions. Deployment type sets where prompts and responses are processed: Global (any Azure region), Data Zone (within the US, EU or Asia Pacific zone) or Standard (within the Azure geography you specify, possibly moving between regions in that geography for operational purposes). Data at rest stays in the designated geography. Deployment types. Microsoft's general statement is that it will not store or process customer data outside the customer-specified geography without authorisation, with exceptions for global deployments. Azure data residency
  • Tracing and logging. Tracing is generally available for prompt and hosted agents. It captures inputs, outputs, tool calls and results, and token use, stored in Application Insights, and can include sensitive data. Microsoft has a how-to on restricting access to sensitive content in traces. Agent tracing overview
  • Guardrails. Controls for hate, sexual, self-harm, violence, prompt attacks, indirect attacks and protected material apply to agents. A personal information control and task adherence are listed for agents as preview. Tool call and tool response scanning for agents is preview. An agent's guardrail fully overrides its model's guardrail. Guardrails overview

Design suggestion (ours, not Microsoft's)

Small-group suppression and the removal of names must happen in the data layer. If the code tool ran over a raw individual-level file, the AI could in principle write code that exposes one person's pay, and an instruction in the prompt would be the only barrier. A safer pattern is a custom function or service, written by your team, that queries a governed view and returns only aggregated, suppressed results, with the query text returned alongside. Whether that is feasible is for your team to judge.

Open questions we could not verify

  • Are the model and the tool we want available in Australia East as a Standard (in-geography) deployment for our subscription?
  • Does the Asia Pacific data zone satisfy our residency requirement, and which regions does it cover?
  • Where are uploaded files, conversation state and traces stored in our chosen setup, and in which region?
  • Can code interpreter file access be limited per user, or per group, outside the hosted agent and toolbox case Microsoft describes?
  • Is the personal information guardrail for agents suitable while it is in preview?
  • Can hiding one small group be undone by subtraction from totals we still show?
  • What are the current costs? We have included none. Use Microsoft's pricing pages.
  • Should the assistant ever list individual employee IDs? This is a decision for the organisation first.

The full question list is in it-brief.md.

Take the brief to IT

Download it-brief.md from the last section of this page and send it with this note. Change the words in square brackets.

Subject: Request to scope a secure remuneration analyser Hi [name], I would like to scope a small assistant for the reward team. It would let a few named people ask questions of our real pay data in plain English. The AI would write the query, ordinary code would do the maths, and every answer would show its working. Small groups would be hidden in the data before the AI sees the result. The idea is to build it inside our own Azure environment. I have attached a one-page brief with the design rules and a list of questions for IT, security and privacy. Some of the questions are things I could not confirm from Microsoft's documentation, such as which region the data and the AI service would run in. Could we book 45 minutes to go through it? I would like to agree who owns the data, who can access it and who signs off before any real pay data is connected. I would also like to start with invented data to test it. Thanks, [your name]

Good habits

  • Start with invented or de-identified data. Connect real data only after sign-off.
  • Never let the model do the arithmetic. It writes the query, and code runs it.
  • Keep the test set and re-run it after every change to the model, the instructions or the data.
  • Read the working, not just the answer. Spot-check real answers against a calculation you do yourself.

4Check the data

What this touches

This is the one use case on the site designed for real, individual pay data, so it deserves the most care. It involves individual pay, gender, level and job family, which people rightly treat as highly sensitive.

What is safe

  • Only inside an environment your organisation controls and has approved for this data.
  • Employee IDs, not names.
  • Small groups hidden before the AI ever sees the result.
  • Access by named group.
  • Logging on.

Ask before you start

  • Has our privacy officer assessed this use?
  • Which region will the data and the AI service run in?
  • Who can see the conversation logs?
  • What is our rule for small groups?
  • Who signs off before real data is connected?

Privacy rules for employee records differ between private employers and government agencies, so check with your privacy officer. Start with the OAIC's guidance on the Australian Privacy Principles and on employment and privacy. General information, not legal advice.

5Take it

Download the IT brief Download the sample questions

Use the brief to start the conversation with your IT, security and privacy teams, and use the questions as the first draft of your test set. Nothing here has been built or tested yet.

What next: Back to all use cases