Claude Projects: Your Persistent Compliance Program Assistant

What This Builds

A persistent AI assistant that already knows your institution's compliance program — its regulatory framework, policy structure, risk profile, and key requirements — so every interaction starts from the right context. Instead of explaining "we're a community bank under the OCC with a BSA risk rating of moderate" in every chat session, you say that once, and the assistant remembers it for every conversation.

This replaces 15–30 minutes of context-setting per session and produces more institution-specific drafts from the first response rather than the third.

Prerequisites

• Claude Pro subscription ({{tool:Claude.price}}) — required for Projects feature
• Compliance program documentation to load (sanitized — no client data)
• 1–2 hours for initial setup and testing
• Basic comfort using Claude for drafting and analysis

The Concept

A Claude Project is like hiring a new compliance analyst who has already read every key document in your compliance program before their first day. You set it up once — loading in your regulatory profile, program overview, and policy framework — and then every conversation starts from that baseline. The assistant gives you institution-specific responses without you having to re-explain your context each time.

Build It Step by Step

Part 1: Gather Your Context Documents (Sanitized)

Before building the Project, collect these documents and remove all client data, account numbers, and personally identifiable information:

1. Compliance Program Overview — 1–2 page summary of your program structure, regulatory relationships, key policies
2. Regulatory Profile — Charter type, examining agencies, asset tier, key product lines, current risk ratings
3. Policy Index — A list of all active policies with their titles and last review dates (no policy text — just the index)
4. Key Regulatory Framework — A 1-page summary of the primary regulations governing your institution (BSA, Reg E, TILA, etc.)
5. Compliance Calendar — Key examination dates, annual testing deadlines, reporting cycles

Total: 5–10 pages of sanitized overview documents. Keep it lean — this is context, not a policy library.

Part 2: Create the Claude Project

1. Log in to Claude at {{tool:Claude.url}}
2. Click Projects in the left sidebar (available with Claude Pro)
3. Click New Project
4. Name it: "Compliance Program Assistant — [Your Institution Type]"
5. Click Add Context and upload or paste each document from Part 1

Part 3: Write the System Instructions

In the Project's Instructions field, write:

You are a compliance program assistant for [describe your institution generically, e.g., "a community bank under $2 billion in assets, chartered by the OCC, with a moderate BSA/AML risk profile"]. You have been provided with our compliance program overview, regulatory profile, and policy framework.

Your role is to help the compliance officer:
1. Draft policy language, examination responses, SAR narratives (sanitized), risk assessment narratives, and board reports
2. Analyze regulatory changes for impact on our program
3. Answer questions about regulatory requirements with appropriate caveats that legal review may be needed

IMPORTANT PRIVACY RULES:
- Never ask for or use real customer names, account numbers, or PII
- All examples should use generic descriptions only
- Always note when outputs require legal counsel review before finalization

When drafting compliance documents: use formal regulatory language, be specific and precise, and flag any content that needs institution-specific customization with [CUSTOMIZE] markers.

Part 4: Test the Project

Ask these test questions to verify the context is loaded correctly:

• "What are the primary regulations governing our compliance program?"
• "Draft a short explanation of our BSA risk profile for a new employee orientation"
• "If a new CFPB rule were announced tomorrow, walk me through how our program would respond"

Evaluate: Does the assistant reference your specific institution context? Does it give institution-specific answers rather than generic ones?

Part 5: Refine the Context

Based on your testing, add missing context or clarify the instructions:

• If the assistant doesn't know your examination cycle, add it to the context
• If it gives overly generic responses, add more specifics to your regulatory profile document
• If it's giving advice that doesn't match your institution type, clarify in the instructions

Real Example: Policy Update Workflow

Setup: You've loaded your compliance program overview, regulatory profile (OCC-chartered community bank, moderate BSA risk), and a policy index with 25 active policies.

Input (in any new conversation in this Project): "FinCEN just issued a new advisory on virtual currency red flags. I need to assess whether it affects our policies and draft a management alert."

Output (institution-specific, no re-briefing needed):

• "Based on your institution's profile as a community bank with limited virtual currency exposure, this advisory has moderate relevance. Specifically, your BSA policy's Red Flag section and your CDD procedures may need to reference this advisory. Here's a draft management alert: [draft follows, using your institution's program structure]"

Time saved: No 15-minute context-setting preamble. Immediate, institution-specific response.

What to Do When It Breaks

• The assistant forgets your context → It may be a long conversation that has pushed earlier context out — start a new conversation within the same Project (context documents reload)
• Responses feel too generic → Your context documents may not be specific enough — add more detail to the regulatory profile or compliance program overview
• The assistant makes up regulatory details → This is a known AI limitation — always verify specific regulatory citations against source documents before using in official materials
• Project becomes outdated → Refresh your context documents annually (after examination cycle, after major regulatory changes)

Variations

• Simpler version: Create a single "compliance context" document that you paste at the start of every Claude session instead of using Projects — achieves similar results without the Project setup cost
• Extended version: Add your 3–5 most-used policy templates to the Project context so the assistant can produce policy language in your exact house style from the first draft

What to Do Next

• This week: Build the Project with minimal context (just your regulatory profile and program overview) and test it on 3 real tasks
• This month: Refine the context based on what's missing, and establish a habit of always using the Project rather than starting fresh conversations
• Advanced: Add a second Project for examination preparation season, loaded with your prior examination reports and remediation documentation (sanitized)

Advanced guide for Compliance Officer professionals. These techniques use more sophisticated AI features that may require paid subscriptions.