Custom AI Assistant: Your Compensation Policy Expert

For Compensation and Benefits Analysts

Tools: Claude Pro | Time to build: 1–2 hours | Difficulty: Intermediate-Advanced Prerequisites: Comfortable using Claude for document drafting — see Level 3 guide: "Write Pay Equity Analysis Narratives"

What This Builds

Instead of starting from scratch every time a manager or HRBP asks a compensation question, you'll build a Claude Project pre-loaded with your company's compensation philosophy, salary structures, job architecture, and key policies. Every conversation starts with full context: it knows your bands, your philosophy, your approval thresholds. You go from "let me pull up the spreadsheet and check" to a 30-second answer.

Prerequisites

• Comfortable using Claude Pro for document analysis tasks (Level 3)
• Claude Pro or Team subscription — Projects require a paid plan ({{tool:Claude.price}}/month)
• Your company's comp documents ready to upload: salary structure, job architecture overview, comp philosophy statement, key policies (off-cycle increase, FLSA classification)
• 1–2 hours for initial setup; 15 minutes for periodic updates

The Concept

A Claude Project is like a standing briefing you give a knowledgeable advisor before every meeting. Instead of re-explaining your salary bands every time, the Project holds all the context permanently: your uploaded documents, your custom instructions, your preferred outputs. Every new conversation you start inside that Project automatically has access to all of it.

Think of it as hiring a temporary analyst who reads your entire comp policy manual on Day 1 and never forgets it. You don't explain your job architecture every time. You just ask the question.

Build It Step by Step

Part 1: Create the Project

1. Log in to Claude at {{tool:Claude.url}}
2. In the left sidebar, click New Project
3. Name it: "Compensation Analysis Assistant" (or your preferred name)
4. You'll see a Project page with tabs: Overview, Documents, Conversations

What you should see: A Project workspace where you can add documents and set custom instructions that apply to every conversation within the project.

Part 2: Upload Your Company Documents

Click Add to Project (or the document upload icon) and upload the following. Start with what you have; even 2–3 documents provide significant value:

Priority 1 — Upload first:

• Your current salary structure (grade ranges, midpoints by grade, as Excel saved as PDF or plain text)
• Your comp philosophy statement (even a 1-page document is valuable)
• Your job architecture overview (career levels, level definitions, if documented)

Priority 2 — Add next:

• Off-cycle increase policy
• FLSA classification policy and decision criteria
• Bonus and incentive plan overview (without individual employee data)
• Benefits summary (for quick reference on common questions)

What you should see: Documents listed in the Project sidebar. Claude can now reference these in any conversation within the project.

Important — Data sensitivity: Do NOT upload documents containing individual employee names, salaries, or protected class information. Upload structural documents (grade ranges, level definitions, policies), not employee data.

Part 3: Write Your System Instructions

In the Project settings, find Custom Instructions (or "Project Instructions") and write the following, customizing the [brackets]:

You are a compensation analysis assistant for [Company Name]'s People & Compensation team.

Your role: Answer compensation questions, draft communications, and analyze scenarios using the company documents uploaded to this project.

Company context:
- Industry: [your industry]
- Company size: [employee count]
- Pay positioning strategy: [e.g., "50th percentile for most roles; 75th percentile for critical tech roles"]
- Compensation year: [current year]
- HRIS system: [Workday/SAP/etc.]

How to respond:
- For questions about salary ranges: reference our salary structure document. Provide the grade range and flag if the requested salary is outside the normal band.
- For FLSA questions: reference our classification policy. Always recommend legal review for borderline cases.
- For policy questions: cite the relevant policy section in your answer.
- For off-cycle increase requests: explain the approval path from our off-cycle policy.
- Always note when a decision requires CHRO or finance approval per our policies.
- Do not speculate about specific employees or make commitments outside our documented policies.

Tone: Professional, helpful, specific. Use our internal terminology (e.g., "grade bands" not "salary ranges" if that's our term).

What you should see: Custom instructions that appear at the start of every Project conversation, giving Claude your full company context before you type your first question.

Part 4: Test and Refine

Open a new conversation inside the Project and test with 3–4 realistic questions:

1. "What's the grade range for a Senior HR Business Partner?"
2. "A manager wants to give an off-cycle increase to a high performer. What's the process and approval threshold?"
3. "We're adding a new role called 'AI Operations Lead.' What grade would you recommend and what FLSA classification?"
4. "Draft a note to the CHRO recommending a market adjustment for our Data Engineering team that's 18% below P50."

Review the answers. If responses are too generic, add more specific documents. If it references the wrong policy, clarify in your instructions ("for off-cycle increases, always reference the Off-Cycle Increase Policy document, not general guidance").

Real Example: A Manager's Off-Cycle Request

Setup: Your Project has the salary structure, off-cycle policy, and comp philosophy uploaded.

A typical conversation: You: "A manager is requesting an off-cycle increase for a Software Engineer II currently at $98K. She got an outside offer for $115K. Grade 7 range is $82K–$118K. What can we do?"

What the assistant returns:

• Confirms she's within the grade range (no structural issue)
• References the off-cycle policy: retention increases up to $X require HR manager approval; above $X require CHRO approval
• Notes she's currently at 88% of midpoint ($111K), suggesting she's below market
• Suggests a market correction framing rather than "counter-offer matching" to avoid precedent issues
• Drafts a 3-sentence recommendation memo for manager submission

Time saved: 20 minutes of policy lookup, email drafting, and calculation, done in 30 seconds.

What to Do When It Breaks

• Gives wrong grade range → Verify your salary structure document was uploaded correctly. Try asking: "What documents do you have access to in this Project?" It should list your uploads.
• Ignores policy document and gives generic advice → Add a specific instruction: "When asked about [topic], always reference the [Document Name] first before giving general guidance."
• Makes up approval thresholds → Add specific approval amounts directly to your custom instructions rather than relying solely on document extraction.
• Responses are too long → Add to instructions: "Keep answers under 200 words unless asked for a full draft document. Use bullet points for policy summaries."

Variations

• Simpler version: Use a single conversation with the same pasted context at the start each time. No Project setup needed; just paste your comp philosophy + grade ranges in message 1.
• Extended version: Add your manager self-service guide, total rewards FAQ, and open enrollment timeline. This turns it into a complete total rewards knowledge base for the whole HR team.

What to Do Next

• This week: Set up the Project with your top 3 documents and test with 5 real questions you've been asked recently.
• This month: Add remaining policy documents; share access with your HRBP team (if on Claude Team plan).
• Advanced: Add a "monthly update" prompt at the start of each conversation that reminds you to update the salary structure when you do your annual range refresh.

Advanced guide for compensation and benefits analysts. Claude Projects require a Pro or Team subscription. Do not upload documents containing individual employee names or salary data.