Comp AI Starter Kit

A practical guide for Total Rewards professionals. No coding required.
Everything below is free.

Cowork is not a chatbot. It is an autonomous agent that lives inside the Claude Desktop app. Where regular Claude chat waits for you to type a message after every step, Cowork takes a goal and works through it independently. It plans tasks, reads your files, writes documents, runs scripts, and delivers finished work to your folders.

It connects to the tools you already use: Google Drive, Gmail, Notion, Calendar, Slack. Think of it as a junior analyst who can do the legwork but needs your judgement on every decision that matters.

For compensation teams, that means less time on the repetitive mechanics and more time on the work that actually requires a senior practitioner's brain.

This is where most people get Cowork wrong. They open it, type a question, get a decent answer, and move on. Then next week they open it again and it has forgotten everything. They skipped the setup that makes Cowork genuinely useful.

Your Project holds the context. Your Skill holds the procedure. Together, Claude produces work that is consistent, context-aware, and aligned with how your team actually operates.

Compensation work is cumulative. The decisions you made about salary bands in January inform the merit cycle in March which feeds the pay equity review in June. If your AI tool forgets everything between sessions, you are re-teaching it every time.

Instructions you write. A permanent file called CLAUDE.md that Claude reads at the start of every session. Put your comp philosophy, your naming conventions, your preferences here. Claude follows them every time without being asked. Your starter template is at the bottom of this guide.

Memory Claude writes itself. When you correct Claude, it saves that as a note and applies it in future sessions. Your corrections compound over time.

The limitation. Memory is scoped to the Project. What Claude learns in your merit cycle Project does not carry over to your job architecture Project. Think about your Project structure upfront.

Large language models predict the next word. They do not compute. When you ask Claude to calculate a compa-ratio in conversation, it is making a statistical guess. For anything involving multiple variables, regressions, or large datasets, it is unreliable.

When you describe an analysis to Cowork, it writes a Python script and runs it inside a sandboxed virtual machine on your computer. Same input, same output, every time. Auditable. Repeatable. Deterministic.

Your data is sensitive. Salaries, performance ratings, demographic breakdowns, disability status. This is GDPR-regulated personal data. The principle: ask the AI to write the code. Run the code yourself, on your machine. The AI never sees a single employee record.

Three things still travel to the AI: column names (metadata, not personal data, but they reveal what your organisation tracks), error messages (check before sharing as they sometimes include sample values), and small-group outputs (keep aggregations above around 10 people).

Always start with synthetic data. Ask Claude to generate a realistic dataset with the right structure but no real people. Iterate until the code works, then swap in the real file path.

Upload your cycle guidelines and timeline into a Project. Ask Cowork to draft cycle kickoff emails for managers, mid-cycle reminders for HRBPs, and post-cycle summaries for leadership. You review and send.

Upload your guidelines, merit matrix, and FAQ document with a simple instruction: answer only from these documents; if the answer is not here, say so. This alone saves hours during every cycle.

Give Cowork your anonymised summary data (not individual records) and ask it to generate discussion guides, flag outlier recommendations, and draft calibration talking points. A pack per department in minutes.

Export market data from your survey providers. Ask Cowork to identify roles below P25 or above P75, calculate gaps, and draft a summary with recommended actions. You own the interpretation. Cowork does the formatting.

Paste a job description and ask Cowork to suggest matching survey codes. Always review the matches yourself.

The Code Not Data approach may already satisfy your policy. If the restriction is about data exposure and your method ensures zero employee data enters the AI, you may be compliant. The AI is a code generator, not a data processor. Worth taking to your security team.

Shadow AI is already happening. Research shows employees in over 90% of organisations are using consumer AI tools for sensitive work, including pasting pay data into public interfaces. Approving an enterprise tool is not adding risk. It is governing risk that already exists.

If you need help building the internal case, bring it to Range. We have members who have had this conversation with Legal, IT Security, and Procurement and can share what worked.

This is the file that makes Cowork remember who you are and how to work with you. Paste it into a new Cowork session with this instruction: save this to my CLAUDE.md file. Fill in the brackets. Delete anything that does not apply to your role.

The barrier to using AI in compensation is not the building part (now Claude, Codex, Copilot can do that for you). It is knowing where to start and knowing how to keep your data safe while you do it. The judgement, the interpretation, the business context. That is yours. Cowork handles the mechanics so you can spend more time on the work that actually needs you.

Published by Giac Soliman, Founder of Range.