P
AI for Compensation and Benefits Analyst

Custom AI Assistant: Build a Compensation Policy Claude Project

Claude

For Compensation and Benefits Analysts

Tools: Claude | Time to build: 2-3 hours | Difficulty: Intermediate-Advanced Prerequisites: Comfortable using Claude for document analysis — see Level 3 guide: "How-To Guide: Write Pay Equity Analysis Narratives"

What This Builds

A Claude Project configured as your team's internal Compensation Policy Assistant. Instead of HR Business Partners emailing you the same 15 questions every week ("Is $X competitive for this grade?" "What's our market position target?" "Can we offer above the range?"), they query the assistant directly and get consistent, policy-grounded answers. You load your salary structure, comp philosophy, grade bands, and key policies once. The assistant handles the routine questions.

Prerequisites

  • Claude Pro or Team subscription at {{tool:Claude.price}}/month (Projects feature required)
  • Your current salary structure in a readable format (Word, Excel exported to PDF, or plain text)
  • Your compensation philosophy document (even a one-page version works)
  • Core comp policies: off-cycle increase policy, promotion criteria, geographic pay adjustment rules
  • Basic comfort using Claude for writing and analysis tasks

The Concept

A Claude Project is like having a new team member who has memorized your entire compensation handbook before their first day. Every time someone starts a conversation with your Project, Claude already knows your salary bands, your market target (P50? P75?), your increase approval thresholds, and your FLSA classification framework. You set it up once. Claude applies that context automatically to every conversation, every time.

Think of it as the difference between calling a consultant who has to relearn your business every call versus a seasoned internal colleague who already knows your policies cold.

Build It Step by Step

Part 1: Gather and format your knowledge documents

Before setting up the Project, collect and format the documents Claude will use:

Documents to include:

  1. Salary structure: Grade bands, salary ranges, pay grade definitions. Export from Excel as a table in Word or paste as plain text. Include midpoints, minimums, and maximums.
  2. Compensation philosophy: Your company's market positioning (P50 base, P75 TCC?), internal equity principles, geographic pay approach.
  3. Key comp policies: Off-cycle increase criteria and approval levels, promotion pay guidelines, signing bonus/relocation policy, pay transparency policy.
  4. Job architecture: Job families, levels, FLSA classification by grade (if documented).
  5. Common Q&A: If you have an existing FAQ for HRBPs, include it. It teaches Claude the format of questions it will receive.

Formatting tips:

  • Use clear headers (##) so Claude can navigate the documents
  • Convert Excel ranges to plain text tables: Grade 7: $72,000 – $95,000 – $118,000 (min-mid-max)
  • Remove any personally identifiable data. This is policy content only, not employee records.

Part 2: Create the Claude Project and upload documents

  1. Log in to {{tool:Claude.url}} and click Projects in the left sidebar
  2. Click Create Project
  3. Name it: "Comp Policy Assistant: [Company Name] [Year]"
  4. Click Add content → Upload files and upload each document you prepared in Part 1
  5. Alternatively, use Project Instructions (the text field) to paste content that's short enough to type directly

What you should see: Each uploaded document appears in the Project's knowledge panel with a preview. Claude can now reference these documents in any conversation within the Project.

Part 3: Write the Project Instructions (system prompt)

The Project Instructions are the most important part: they tell Claude who it is, what it knows, and how to respond. Click Edit Project Instructions and write:

Copy and paste this
You are the Compensation Policy Assistant for [Company Name].

## Your role
You help HR Business Partners and hiring managers answer compensation policy questions. You are knowledgeable, direct, and precise. You only advise on our company's actual policies and salary structures — you do not speculate or invent policies.

## What you know
- Our salary structure: grades, ranges, midpoints (see Salary Structure document)
- Our compensation philosophy: market positioning, internal equity principles (see Comp Philosophy document)
- Our key policies: off-cycle increases, promotion guidelines, offer approval thresholds (see Policies document)
- FLSA classification by grade level (see Job Architecture document)

## How to respond
- Always cite which policy or salary structure range your answer comes from
- If a question is outside our documented policies, say "This isn't covered in our current policy documents — please connect directly with the Compensation team for guidance."
- For offer pricing questions, provide the grade range and market position context, but remind the user that offers above range require Compensation team approval
- Do not share individual employee salary information — you have access to ranges and structures, not individual data
- Be direct. HRBPs are busy — give the answer first, then the explanation.

## Tone
Professional but conversational. These are your teammates — not an external audience.

Part 4: Test and refine with real questions

Before sharing with the HRBP team, run 10 test questions through the Project yourself:

  1. "What is the salary range for a Grade 8 Manager?"
  2. "Our candidate has 12 years of experience and competing offers. We want to hire at $105K for a Grade 7 role. Is that within policy?"
  3. "What's the difference between our P50 base target and TCC target?"
  4. "Can a manager approve an off-cycle increase without comp approval?"
  5. "Is a Senior Marketing Analyst typically exempt or non-exempt?"

What good output looks like: Direct answers with specific range data cited, clear escalation guidance ("this requires Compensation approval") on out-of-policy scenarios, and appropriate "I don't know / contact the team" responses for gaps.

What to refine: If Claude invents a policy that isn't in your documents, add a line to your Instructions: "If a question refers to a policy not in the knowledge documents, say so explicitly — do not infer or create a policy." Retest until hallucination is eliminated.

Real Example: HRBP Scenario

Setup: You've uploaded your salary structure (Grades 1–15, min-mid-max), your comp philosophy (P50 base, P75 TCC for engineering), and your off-cycle increase policy (requires VP approval + Compensation sign-off for >5%).

HRBP input: "I have a Software Engineer 3 who just got a counteroffer from Meta at $185K base. We're at $162K which I think is midpoint for Grade 11. What are our options?"

Claude's output (good): "Grade 11 Software Engineer range is $145,000 – $180,000 – $215,000 (min/mid/max). At $162K, your employee is at 90% compa-ratio — below midpoint. Market context: our P75 TCC target for engineering means we should be competitive at $185K when total compensation is factored in.

Options within policy: An equity adjustment to $175–185K would bring them to midpoint-to-above-midpoint compa-ratio and is approvable at VP level + Compensation sign-off (over 8% — see Off-Cycle Increase Policy). Alternatively, review if their level is appropriate — if they're performing at a G12 level, a promotion analysis would be the right path.

Action needed: Connect with Compensation to run a retention analysis before the employee responds to the counteroffer."

Time saved: The HRBP got a grounded, policy-specific answer in 30 seconds instead of emailing you and waiting for a response.

What to Do When It Breaks

  • Claude cites a wrong salary range → Check which document version was uploaded. Outdated salary structures cause this. Update the file in Project Settings → re-upload.
  • Claude invents a policy → Tighten your Instructions: add "Do not infer policies not present in the knowledge documents." Retest.
  • Claude refuses to answer legitimate questions → Add the specific question type to your Instructions as an explicit example of what it should handle.
  • HRBP gets inconsistent answers in different conversations → Check that all conversations are happening inside the Project (not in a regular chat). Project context only applies inside Project conversations.

Variations

  • Simpler version: Use a single Claude conversation with a long system prompt pasted at the start — works for personal use, but doesn't scale to a team
  • Extended version: Add a separate "Benefits Knowledge" section with plan summaries and open enrollment FAQs, effectively creating a combined Total Rewards Assistant

What to Do Next

  • This week: Build the Project, test with 10 questions, refine Instructions
  • This month: Share with 2–3 HRBPs as a pilot. Collect questions it answers poorly. These become your next refinement round.
  • Advanced: Connect your company's HR portal to Claude via API so HRBPs can query the assistant directly from their intranet without switching tools

Advanced guide for Compensation and Benefits Analyst professionals. Claude Projects require a paid subscription. Keep salary structure and policy documents current, as stale data produces inaccurate answers.