Microsoft Copilot for Compensation

Why this guide exists

Millions of people use Microsoft Copilot every day. Thousands of them are reward professionals. And almost every tutorial you can find treats Copilot as if it lives in a vacuum. Here is how to summarize a document, here is how to draft an email. Zero understanding of what a compensation analyst actually does with their day.

Reward work is different. You handle personal data that is regulated under GDPR and monitored by auditors. You run merit cycles where a single bad formula cascades into forty wrong recommendations. You brief board committees where a wrong number is a career defining mistake. And you do it all inside the Microsoft stack because your employer bought it and locked the door on everything else.

This guide is for you. It mirrors the Claude Cowork for Compensation guide we published earlier, but re-grounded for the tool you are actually allowed to use at work.

Before Copilot can help you, you need to understand three things the average tutorial never explains: which Copilot you have, what your company can see, and where Copilot is safe to use for comp grade data. The rest of this guide then walks through the workflows that matter in a typical rewards year.

The License Minefield

Microsoft Copilot is not a single product. It is a brand stretched across at least four different tools with four different pricing models and four different data boundaries. Most people do not know which one they are using, and that is dangerous when you work with salary data.

The four flavours, translated

Consumer Copilot (copilot.microsoft.com, signed in with a personal Microsoft account). Free. Looks identical to the work version. Not covered by any enterprise agreement. Prompts and uploads may be used to train future models. Never paste comp data into this. Ever.

Microsoft 365 Copilot Chat (same URL, but signed in with your work Entra ID). Free to anyone with a work account. Covered by Microsoft's Enterprise Data Protection commitment, meaning prompts stay inside your tenant and are not used to train models. But it only grounds answers against the public web. It cannot see your SharePoint, your Outlook, your OneDrive, or your Teams chats. Think of it as ChatGPT with a legal shield, not as an employee.

Microsoft 365 Copilot (the paid SKU, roughly $30 per user per month depending on your EA). This is the full version. It reads your SharePoint, your Outlook threads, your OneDrive, your meeting transcripts, your Teams chats, and it can operate inside Excel, Word, Outlook, PowerPoint and Teams. Everything in this guide assumes this license. If you do not have it, ask your CIO for a seat (or at minimum a pilot) before you spend another cycle drafting manager emails by hand.

Copilot Studio. A separate paid product for building custom agents that connect to bespoke data sources. Outside the scope of this guide, but worth knowing the name exists when your IT team starts talking about it.

How to check which one you have in 30 seconds

1. Open Copilot (web app or in Teams).
2. Look at the top right corner. If you see your work email and a small badge that says Protected or Work, you are on a protected in your tenant license.
3. Now look for a toggle near the prompt box that says Work and Web. If that toggle exists and Work is selectable, you have Microsoft 365 Copilot, the paid SKU. If it is missing or greyed out, you are on Copilot Chat only.

The Data Boundary: What Your Company Can Actually See

This is the section that matters most for compensation teams, and the one every generic Copilot tutorial skips.

The training session that inspired this guide addressed it directly:

Yes. Your company actually can. Microsoft is saying this data is protected, so it is not going to the internet. But your company could flag what you typed and come back and say, why did you type that? Don't type my boss is a jerk, what should I say to him? Don't type how much is my coworker making? Don't ask for their social security number. All of that could be flagged.

For a compensation analyst this is not a minor footnote. It is the entire posture. Your Copilot prompts and outputs sit inside your tenant's audit log. They can be pulled by Purview for eDiscovery, reviewed by HR during an investigation, and seen by your admin with the right role. That is a feature, not a bug. It is what makes the tool legally acceptable for comp work in the first place. But it changes how you phrase things.

What lives inside the protected boundary (usually safe)

• Prompts and responses when signed in with your work account on M365 Copilot or Copilot Chat.
• Uploads into a Copilot conversation. Microsoft commits they are not used to train models.
• Content Copilot reads from your SharePoint, OneDrive, Outlook, Teams via the Graph.
• Generated files (Word, Excel, PowerPoint outputs).

What leaves the boundary or gets flagged (be careful)

• Anything typed in a browser signed into a personal Microsoft account instead of your work one. Check the avatar in the top right every time.
• Web grounded prompts. When you toggle to Web, Copilot sends a search query to Bing. The query itself is logged, and the websites it fetches see a request originating from Microsoft. Do not paste salary band midpoints into a prompt in Web mode.
• Anything a reasonable Purview admin would find suspicious if they audited. Show me what Sarah from Finance is paid is worse than Show me compa ratios across Finance.
• Prompts typed into third party plugins and connectors that are not on your tenant's approved list.

The three rules of thumb for comp

1. Individual level data stays out of the prompt box. Attach it as a file in a scoped to Work conversation where permitted. Do not retype salaries or employee IDs into an open ended prompt.
2. Small group aggregations are still identifying. Why does the output show a 23% gap for the three female directors in Market X? is re-identifiable even if no names are used. Keep group questions at n=10 or higher, or use anonymised summary data.
3. Use Temporary Chat for one offs. If you are experimenting with a prompt about sensitive territory and do not want the conversation stored or used to update your saved memories, open a Temporary Chat. It ignores your custom instructions, is not saved to history, and does not feed the memory system. Useful when you want to try something without it influencing future sessions.

Setup for a Rewards Practitioner

Copilot, like every other AI tool, is only as good as the context you give it. Five minutes of setup at the start of your first session saves hours across every cycle after that.

Custom Instructions

Settings → Personalisation → Custom Instructions. This is how you tell Copilot who you are and how you want to be spoken to. These instructions are applied to every outside Temporary chat you open.

A starter pattern for a senior reward practitioner:

I am a Total Rewards professional responsible for compensation
design, merit cycles, pay equity analysis, and market benchmarking.

When you answer me:
* Default to plain practitioner language, not HR consultancy jargon.
* When presenting data, show percentiles as P25, P50, P75 (not 25th
  percentile).
* When I ask for analysis, identify the relevant framework first (for
  example cascading bands, compa ratio, range penetration) before
  producing numbers.
* When I ask you to calculate something, never guess. Write a script
  or ask me to run it in Excel with Copilot.
* Flag regulatory implications for EU Pay Transparency, UK Gender Pay
  Gap, and SEC pay ratio where relevant. Do not volunteer a legal
  opinion. Flag it and let me decide.

Change the specifics to match your role (in house versus consulting, country focus, level seniority), but this structure holds up across rewards contexts. Save it and forget it.

Saved Memories

Saved Memories sit next to Custom Instructions but work differently. Custom Instructions are rules. Memories are facts Copilot should remember about you.

A sensible starter set:

• Your role and scope. I am Head of Total Rewards at a 2,400 person fintech, reporting to the CPO. I cover EMEA and APAC, not North America.
• Your compensation philosophy in one paragraph. We pay at P60 of a FAANG tier peer group for engineering, P50 for all other functions. We run one merit cycle in April and one promotion cycle in October. No off cycle adjustments above 5% without my explicit approval.
• Your vendors and data sources. Survey providers: Mercer TRS, Radford Global Technology, WTW. Equity admin: Carta. HRIS: Workday.
• Naming conventions. We use the bands L1 through L9, not job grades. L5 is our default IC career level.

This pattern is the Microsoft equivalent of a Claude CLAUDE.md file. It is how you stop re-explaining your organisation every morning.

The Agents Worth Using

Copilot ships with a library of prebuilt agents on the homepage (Tools → All agents). Most are noise. Four of them are genuinely useful for comp work. The honest ranking:

1. Prompt Coach, use this first

Prompt Coach is the one almost nobody uses and everybody should. You describe vaguely what you want (help me write a pay equity memo for the board), and it walks you through goal, context, source, and expectations before producing a polished prompt you can then send to any other agent.

The point isn't the prompt. The point is that Prompt Coach forces you to articulate what you actually want before you hand it off to an AI. Reward professionals spend their lives translating fuzzy executive briefs into defensible analyzes. Prompt Coach is that same muscle, applied to Copilot.

Use it when: you are about to ask for something substantive and you are not sure how to scope it. Pay equity commentary. Board memos. Salary band refresh methodology.

2. Researcher, for regulation and market context

Researcher gives you a five to seven page expert grade research report with citations. You set a goal (produce a research report on the operational implications of the EU Pay Transparency Directive for multinational employers with a European workforce), and it crawls the public web, cross references sources, and writes the report for you.

For comp this is the one agent you will use most. Reward practitioners are drowning in regulatory change. The EU Pay Transparency Directive, the UK Section 78 gender pay gap review, US SEC pay ratio updates, Indian equity taxation reform. A Researcher report gets you a briefing you can circulate to your HRBPs in ten minutes instead of five hours.

Caveat: Researcher grounds in public web data. It will not know your internal salary bands, your board papers, or your previous filings. Pair its output with your own M365 Copilot prompts that ground in your tenant.

3. Analyst, for unstructured survey data

Analyst is a multi step data analysis agent. You upload a file (CSV, Excel, PDF with tables) and ask open ended questions. It is more capable than Copilot in Excel for anything that crosses multiple sheets or mixes unstructured commentary with numbers.

Where it shines for comp: Analyst handles survey comment fields, exit interview themes, and pulse survey feedback well. You can give it 400 verbatim responses to a pay satisfaction question and ask what are the three dominant themes, ranked by frequency, with a representative quote for each.

Where it does not shine for comp: anything involving actual employee level salary calculations. See the final section on the maths problem.

4. Idea Coach, for cycle comms brainstorming

Idea Coach is a brainstorming partner. You describe a constraint (I need five ways to frame a 2.8% total merit budget positively to a managerial audience that is expecting 4%), and it generates variants. It is useful when you are staring at a blank page in an Outlook draft.

The ones you can ignore

Career Coach, Writing Coach, and Learning Coach are fine but not uniquely useful for rewards work. Skip them.

Excel: The Merit Cycle Cockpit

Excel is where compensation actually happens. Merit matrices, salary band tables, pay equity regressions, range penetration analyzes. And with Copilot's Wave 3 release, Excel moved from the spreadsheet you memorized VLOOKUP for to a workspace an AI can edit directly.

Open an Excel file. Look for the Copilot icon in the ribbon. If the menu says Edit with Copilot, you have Wave 3. That is where everything below happens.

The dummy data trick

The single most important habit when using Copilot in Excel: always ask it to generate a working sheet with dummy data first. Not because the dummy data matters, but because it forces Copilot to build every formula, every data validation, every conditional format rule in front of you while you watch.

Once the sheet is obviously correct (dropdowns behaving, status colors firing, days until due calculating) you clear the dummy rows and paste in your real data. Same structure, same formulas, now with full clarity about how it was built.

For comp work this matters because you are often the only person who will ever re-open the sheet six months later and need to defend the logic. I asked an AI and pasted the result is not a defense. Here is the scaffolded version the AI built with synthetic data, and here is the validation I ran before loading the real population is.

What to build with one prompt

Copy and paste any of these into Edit with Copilot inside a blank workbook and Wave 3 will scaffold the whole sheet.

Merit cycle tracker

Create a merit cycle tracker for my People team. Columns I need:
employee ID, manager name, current base (USD), proposed merit %,
proposed promotion increase %, proposed total new base, compa ratio
after increase (range midpoint goes in a separate sheet), status,
HRBP reviewer, decision deadline. Add conditional formatting: red
if the combined merit + promotion is above 12%, amber if compa ratio
after increase is above 1.20, green if all checks pass. Add a
summary bar at the top with total spend, average increase, and the
number of cases flagged for review. Use dummy data for 15 employees
spanning three managers.

Salary band table with range penetration

Create a salary band table. Columns: band code (L1 through L9),
role family, country, currency, minimum, P25, midpoint, P75,
maximum, range spread %, midpoint to midpoint progression %. Add
a second sheet that takes an employee dataset (dummy data for 20
rows) and calculates range penetration and compa ratio against
the right band. Flag anyone below the minimum or above the maximum
in red. Add a small dashboard showing average penetration by
band and by country.

Gender pay gap working file

Create a working file for a UK Gender Pay Gap analysis. Use dummy
data for 200 employees with columns: employee ID, gender, base
annual salary, bonus in last 12 months, hours worked band,
quartile assignment. Calculate the mean gap, median gap, mean and
median bonus gaps, and the proportion of men and women in each
quartile, exactly as the UK reporting regulations require. Surface
every calculation as a named formula so an auditor can trace it.
Include a "methodology notes" sheet explaining every decision.

Iterating mid build

Wave 3 lets you ask Copilot to fix what it just did. The compa ratio column is showing asterisks, please fix. I want the status column to be a dropdown with Approved, Pending, Rejected, Withdrawn. Add a pivot on the Insights sheet that groups by manager. It edits in place, formulas and all.

You are still the reviewer. Click into each formula and read it. AI generated IF statements look right more often than they are right. Especially for any calculation that feeds a board deck.

The dashboards you should never build by hand again

Copilot can build an Insights sheet in seconds that takes you three hours in native Excel. Prompt: Add a sheet called Insights that summarizes the data in the merit tracker. Include: distribution of merit increases by band, top five outliers for manager review, total spend versus budget envelope, and a callout listing any employees whose combined merit + promo is above 12%. Include one chart per section.

This is the highest leverage workflow for anyone running a comp cycle. You do not need to be a dashboard person anymore. You need to be the person who reviews what the dashboard produced.

Word: Policy & Board Briefings

Word in Wave 3 is the fastest path from 150 pages of source material to a two page board briefing a non specialist can read.

Three moves that change how you draft

1. Generate a brief from a source doc

Drop the source material into OneDrive. Open Word. Click Draft with Copilot and point it at the source. Example prompt:

Create a two page executive briefing based on the attached pay transparency impact assessment. Audience: our Remuneration Committee. Structure: context (half a page), the three most material decisions the committee needs to make (one page), recommended position (half a page). Plain practitioner English. No jargon.

It will produce a first draft inside the Word document itself (not in a side panel). You edit in place.

2. Rewrite for a different audience

The second killer feature. You wrote a dense technical pay equity memo for your Legal team. Now you need the same content for your ExCo in practitioner language, and for your HRBPs in what do I do on Monday format. Wave 3 will rewrite a document for a different audience without losing the underlying logic.

Select the full text. Ask Copilot to Rewrite this document for an HRBP audience who does not have a statistics background. Keep every factual claim, but explain the methodology using analogies a non specialist would understand. Preserve the recommendations exactly.

3. Restructure without cut and paste

Wave 3 can reorder sections without you manually dragging them around. Move the executive summary to the end, lead with the financial impact, then the three scenarios, then the recommendation. Watch it reorganize in real time.

A pattern for the annual comp philosophy refresh

Every two or three years, rewards teams have to refresh the written compensation philosophy. The hardest part is not the writing. It is translating a set of loosely aligned decisions from ExCo and the CPO into a single coherent document without losing anyone's commitments.

Ask Copilot to Read the attached three documents (Board minutes from March, ExCo offsite notes from May, and the current comp philosophy from 2023). Produce a draft refreshed compensation philosophy that preserves every explicit commitment already made, surfaces any contradictions between the sources, and flags them at the end of the document for the CPO to resolve.

This is the kind of work that used to take a senior reward person two weeks of facilitated workshops. Copilot does not replace the judgment. It replaces the synthesis labour, which is the part that bored you anyway.

PowerPoint: Exec Comp Decks Without The Weekend

The board pack cycle is the single worst use of a reward leader's time. Thirty slides, eight revisions, two weekends, and half of it is reformatting.

Copilot in PowerPoint collapses the mechanical work. Judgment stays with you.

From Word to slides in one step

You already have the narrative in a Word document. Open PowerPoint. Click New presentation from file. Point at the Word doc. Copilot produces an outline, you refine the structure, and it generates a full deck with speaker notes.

Speaker notes are the part that matters. The slide is the scaffold. What you say alongside it is the actual briefing, and Copilot drafts it with the same rigour it applied to the Word doc. Every slide comes with a paragraph you can say out loud.

Summarize the 80 slide beast

The inverse move is equally useful. Someone sends you an eighty slide strategy deck and you need to know what it says about rewards before your one on one with the CPO tomorrow morning. Open the deck. Click the Copilot pane. Ask What does this presentation say about compensation, pay, equity, or rewards? Quote directly and give me the slide number for every reference.

Copilot pulls exact quotes with slide numbers. You read four slides instead of eighty.

Brand kits save you from the template war

Every internal rewards deck has a brand kit somewhere. Fonts. Colors. A title slide someone made in 2019 that nobody remembers how to edit. Upload the brand kit to Copilot when you create the presentation. The output inherits the fonts, the color palette, and the title slide treatment automatically.

This ends the era of the reward lead who spends Sunday evening restyling thirty slides so they pass the CPO's eyebrow test.

Where to stay disciplined

Do not let Copilot invent numbers in slide bodies. If it drafts a slide that says our gender pay gap reduced by 3.2pp, verify that the 3.2pp came from a spreadsheet you built, not from a language model's intuition. Pair the deck generation with a source file grounded in Excel. Say The numbers in this deck must come from the attached Excel workbook. If a figure is not in the workbook, leave it blank and add a [DATA REQUIRED] tag.

That single instruction prevents the most common AI board pack failure mode.

Outlook: Cycle Comms Without The Slop

Reward teams write more email than almost anyone else in People. Pre cycle kickoff, mid cycle reminders, letter generation, post cycle manager communications, exec comp statements. Each one has to be accurate, compliant, and on brand.

Copilot in Outlook helps, but only if you fix the default voice first. Left on defaults, it produces the I hope this email finds you well sludge that every LinkedIn post complains about. Two settings fix that.

The Custom Instructions that kill the AI tone

Outlook → Settings → Copilot → Draft with Copilot → Custom Instructions. Paste something like:

Open every email with Hi [name], and nothing else. Do not write
I hope this email finds you well or I hope you're doing great.

Get to the point in the first sentence. State the reason you are
writing.

Use bullet points for anything that is a list. Never wrap three
items into prose.

No em dashes. No moreover, furthermore, in this fast paced
world. No throat clearing.

Close with Thanks, Giac. Do not use Best regards or Warm
regards.

When I paste existing text, preserve my voice. Do not smooth out
my phrasing unless I explicitly ask for a tone change.

Replace the name and your preferred closer. Save once, and every subsequent Draft With Copilot attempt starts from that baseline.

The prompts that replace your cycle email pack

Pre cycle kickoff for managers

Draft a pre cycle kickoff email to all people managers. Context: merit cycle opens 15 March, closes 12 April. Use the attached cycle guidelines and merit matrix. The three things every manager needs to know by Monday: their budget envelope, the calibration deadline, and where to find the FAQ. One short paragraph per point. Keep the total under 200 words.

Mid cycle nudge for HRBPs

Draft a mid cycle status email to HRBPs. I want one short paragraph celebrating that 60% of decisions are in, then a bullet list of the five departments furthest behind, then the deadline and the escalation path. Tone: supportive but direct, not breathless. No friendly reminder.

Post cycle summary for leadership

Draft a post cycle summary for the ExCo. Structure: outcome versus plan (three bullets), pay equity impact after adjustments, three lessons we carry into the next cycle. Attach the merit tracker and reference the key tab names. Length: 300 words maximum.

Inbox triage when you come back from leave

The side panel Copilot (not the in email version, but the persistent pane to the right of your inbox) can run prompts across your whole mailbox. This is what saves you the first day back from annual leave.

Review the emails I received in the last ten days. Identify anything related to the Q2 merit cycle, pay equity review, or the new EU Pay Transparency filing. Group by topic. For each, tell me what action is required, who is waiting on me, and the current deadline. Prioritize by deadline.

Two minutes of output replaces two hours of scrolling. The accuracy is about 90%, which is good enough for triage. You still open the five most important threads yourself to confirm.

The supporting data move

When you are drafting a budget ask or a policy proposal and you want external citations to back it up, switch the side panel to Web mode and ask it to find supporting data. I'm asking for budget to buy a second compensation survey. Help me find credible industry statistics on how often reward teams switch between single provider and multi provider benchmarking, and what the impact is on benchmark match rates. Cite your sources. Copilot returns claims with receipts. You verify them.

Teams: Calibration Meetings On Autopilot

Most of the pain in a calibration meeting is not the conversation. It is the preparation beforehand (who's flagged, what the comp philosophy says, what the budget allows) and the follow up afterwards (who agreed what, who owns the action, when the next review happens).

The Facilitator agent in Teams handles both ends.

Setting up the Facilitator for a calibration session

1. Schedule the calibration meeting in Teams.
2. Open the meeting options. Go to Copilot and other AI.
3. Toggle on Record and transcribe automatically.
4. Toggle on the Facilitator agent.
5. Apply.

During the meeting, Facilitator transcribes every discussion, identifies action items, assigns them to the named owners, and drafts follow up meetings where the conversation implies one is needed. After the meeting, the chat contains a clean summary, a list of decisions, and a list of actions with owners.

Why this matters for calibration specifically

Calibration meetings produce decisions that need to be defended for twelve months, sometimes five years when a promotion is appealed. The Facilitator output is an auditable record of what was actually said, by whom, in what sequence. No more reconstructing a discussion from memory three weeks later when a manager queries a decision.

One discipline. Tell your attendees that the session is being transcribed. People behave differently when they know. For calibration this is a feature, not a bug. You want the discussion to stay rigorous because it is on the record.

Meeting prep in thirty seconds

Double click any meeting in your calendar. At the top of the side panel, click Chat with Copilot to get ready. It cross references the attendees, the meeting topic, your recent emails with that group, and any files shared in the chat, and produces a briefing note. For calibration meetings specifically, you can ask: Prepare me for the Engineering EMEA calibration at 11am. Flag any outlier recommendations in the attached merit tracker, list the three people whose proposed increase is above 10% and why, and surface any prior decisions about these employees from our last cycle.

Notebooks: Your Private Policy Library

Notebooks are the least used, most underrated feature of Copilot for a compensation function.

A Notebook is a bounded set of files. When you ask a question inside a Notebook, Copilot grounds its answer only in those files. The open internet is not consulted. Your wider tenant is not consulted. Only the documents you put in.

Why this matters for comp: you have a growing library of internal policy documents (compensation philosophy, bands, merit guidelines, equity policy, off cycle exception rules, PTO policy, localisation principles) and the number one recurring question from managers and HRBPs is what does our policy say about X? Without a Notebook, Copilot might answer from the public web, which is wrong. With a Notebook, it answers only from your documents.

What to load in your first Notebook

• Current compensation philosophy document
• Current salary bands (all countries)
• Current merit matrix and merit guidelines
• Promotion framework
• Equity grant policy
• Off cycle exception rules
• Comp cycle timeline
• Manager FAQ from the last cycle

Once loaded, share the Notebook with your team. Now every HRBP and people manager who asks what's our policy on X? can get an answer grounded in the actual documents, with citations back to the source page.

The Manager FAQ workflow

Every cycle, managers ask the same twenty questions. What if someone is above the band max? Can I give a merit increase and a promotion in the same cycle? How do I handle an employee on parental leave? Load every policy document into a Notebook, then build a public FAQ page that says: Ask the questions below to this Notebook. It will answer only from our current policies.

You have just replaced two days per cycle of answering the same questions manually.

What Copilot Still Can't Do For Comp

All of the above is genuinely useful. It also has firm limits. Here is where you must not rely on it.

1. It cannot do reliable maths in a chat window

Language models predict words. They do not compute. When you ask Copilot in chat what is the average compa ratio across my 400 person engineering org?, it is guessing based on patterns in its training data. For anything involving regressions, weighted averages, cohort comparisons, or anything board relevant, that is a career ending way to work.

The solve: push the calculation into Excel (via Edit with Copilot, which writes real formulas) or into the Analyst agent with a proper data file. Never accept a number Copilot produced in conversational prose. You did not read this section in the Claude Cowork guide by accident. The rule is identical, because the cause is identical. LLMs do not compute.

2. It cannot replace your audit trail

Copilot keeps its own history. Your compensation audit does not use Microsoft's history. Your audit lives in the actual working files: the Excel workbook, the Word document, the board paper. When Copilot edits those files, the audit trail is whatever versioning your SharePoint is configured to keep. Check with your IT team that version history is enabled with at least a 90 day retention window on any library Copilot writes to, and that deletion events are logged.

The prompt history inside the Copilot pane is useful for you to retrace your steps. It is not a compliance artifact. Treat it as scratch paper.

3. It should not see raw market survey data

Survey vendors (Mercer, Radford, WTW, Korn Ferry) license their data to you under strict terms. Those terms almost always prohibit submitting the raw data to a third party service. Microsoft 365 Copilot is grounded in your tenant, but your lawyers may still consider raw survey submissions a license violation. Default position: analyze survey outputs in Excel locally, and use Copilot on the summarized outputs and the methodology document, not the raw cut files.

If you are on a cross border engagement or a regulated industry (financial services, healthcare, defense), escalate to Legal before you point Copilot at survey data at all.

4. It does not know which employees exist (yet)

Copilot in Excel sees what is in the file you opened. It does not see your HRIS. There is no native Workday or SuccessFactors connector that gives Copilot a live view of your workforce. When you run a merit cycle analysis, you are still responsible for exporting the right population from the HRIS with the right effective date, the right cost centers, and the right exclusions. Nothing in Copilot solves that for you.

This will change over the next two years as Microsoft ships native HRIS connectors via Copilot Studio. Until then, the HRIS extract step remains your job.

Setup Checklist

• Confirm which Copilot you have. If it says Protected in the top right and you can toggle between Work and Web, you are on Microsoft 365 Copilot.
• Write a Custom Instructions block describing your role, voice preferences, and calculation discipline. Copy the template in this guide and adapt it.
• Add three Saved Memories: your comp philosophy paragraph, your vendor stack, and your naming conventions.
• Build your first Notebook with your policy library: philosophy, bands, merit guidelines, promotion framework, manager FAQ.
• Set Outlook's Draft with Copilot custom instructions so every email does not start with I hope this email finds you well.
• Pick one recurring workflow to migrate first. The merit cycle tracker is the highest leverage. Build it in Excel with Copilot, using dummy data, in one sitting.
• For your next calibration meeting, enable the Facilitator agent in Teams. Tell attendees it is on.
• For any question involving numbers that matter, write a formula or a script. Never paste a chat window number into a board pack.
• Flag to Legal before you point Copilot at raw market survey data. Default to working with summarized outputs instead.
• End every session by asking Copilot to update its memory with what it learned. Review what we did today and save anything I corrected into saved memories so you don't repeat the mistake next time.

One Last Thing

Most reward professionals stuck inside the Microsoft stack have been told Copilot is either incredible or useless. Neither is true.

Copilot is a genuinely capable tool that will not calculate your compa ratios for you, will not understand your survey license terms, and will not replace your judgment on pay equity. Used with discipline (the right license, the right toggle, the right custom instructions, and a healthy paranoia about numbers it invents in chat) it collapses the mechanical work that has nothing to do with why you got into this profession.

The rewards year has maybe six weeks of judgment heavy work and forty six weeks of mechanics. This guide is about giving you those forty six weeks back.

If you are figuring out what this looks like in practice and want people to compare notes with, that is what Range is for. We are a community of reward practitioners building with AI together, trading prompts, lessons, and workflow patterns every week. Member funded and practitioner led. A room of operators trying to get better at the craft.