AMG Companies

Claude AI
Usage Agreement

Your license to Claude comes with a commitment — to use AI responsibly, protect sensitive data, and build solutions that scale.

Read & Acknowledge ↓
🛡️ Effective April 2026
Philosophy Claude Models Data Tiers Data Masking Guidelines Agreement
01

The AMG AI Philosophy

Build engines, not one-time answers

🚫

The "AI Well" Trap

Going back to Claude every week with the same question — "summarize last week's sales" or "reformat this spreadsheet" — wastes your time and your license.

Example of what NOT to do
You: "Hey Claude, can you reformat this
sales recap for me again this week?"

This creates zero lasting value. Next week you'll be right back here doing the same thing.

✅

The Scalable Approach

Ask Claude to build you a reusable script, template, or automated workflow that solves the problem permanently. Then you only go back for new problems.

Example of what TO do
You: "Build me a Python script that reads
the weekly sales export and generates a
formatted recap email automatically."

Now the work is done once, and it runs every week without you — or Claude — lifting a finger.

Automate Repetition

If you're asking Claude the same type of question more than twice, ask it to build a tool that answers it automatically going forward.

Create Living Documents

Don't ask Claude to draft something from scratch every time. Have it create templates with variables you can fill in, or scripts that generate documents from data.

Build Your Own Toolbox

Over time, your team should accumulate a library of reusable scripts, templates, and workflows — all built with Claude's help but running independently of it.

Ask Claude How to Automate

Not sure what can be automated? Ask Claude. It can identify repetitive patterns in your work and suggest automation opportunities.

💡
The goal: Claude should make you more capable over time — not more dependent. Every interaction should leave behind something reusable: a script, a template, a documented process.
02

Understanding Claude Models

Choosing the right model for the right task

ℹ️
Why it matters: AMG's enterprise agreement includes access to multiple Claude models, each billed differently. Picking the right model for your task saves cost and gets you better results. Using a premium model for a simple job is like printing a one-page memo on a commercial printing press.
⚡
Claude Haiku 4.5
Fastest & Most Affordable
•• Low Cost
Haiku is Claude's lightest, fastest model. Responses arrive nearly instantly and costs are a fraction of Sonnet. It is optimized for straightforward, high-volume tasks where speed matters more than deep reasoning.
Strengths
  • Near-instant responses — ideal for time-sensitive tasks
  • Lowest cost of all current models
  • Excellent at following clear, structured instructions
  • Reliable for summarization, classification, and formatting
Limitations
  • Less capable at multi-step reasoning or nuanced analysis
  • May miss subtleties in complex or ambiguous prompts
  • Not recommended for strategy, legal, or financial interpretation
Best Use Cases at AMG
  • Quickly cleaning or reformatting data exports
  • Generating repetitive emails, SKU descriptions, or short copy variants
  • Summarizing routine meeting notes or recap emails
  • High-volume tasks in custom scripts (if built by IT)
Input~$0.80 / 1M tokens
Output~$4.00 / 1M tokens
Roughly 750,000 words per $1 at input rates — the most cost-efficient option.
🧠
Claude Sonnet 4.6
Best All-Around — Recommended Default
••• Mid Cost
Sonnet is the sweet spot of the Claude lineup — significantly more intelligent than Haiku but far more affordable than Opus. For the vast majority of AMG employee tasks, Sonnet delivers the best balance of quality, reasoning depth, and cost. It is the default model on claude.ai for enterprise accounts.
Strengths
  • Strong multi-step reasoning and contextual understanding
  • Excellent at writing, editing, analysis, and synthesis
  • Handles long documents and complex instructions reliably
  • Capable of coding, data analysis, and structured output
  • Supports a 1 million token context window
Limitations
  • Slower than Haiku on simple tasks (though still fast by any standard)
  • For the most demanding research-grade reasoning, Opus has an edge
Best Use Cases at AMG
  • Drafting presentations, proposals, and polished communications
  • Analyzing sales reports, brand performance, or market data
  • Building reusable scripts, templates, and Claude Projects
  • Complex research, competitive analysis, or synthesis tasks
  • Everyday default — if you’re unsure which model to use, start here
Input~$3.00 / 1M tokens
Output~$15.00 / 1M tokens
A detailed 5-page report costs well under $0.10 at typical usage — the best value for complex work.
🌟
Claude Opus 4.6
Most Intelligent — Premium Model
••••• Premium
Opus is Anthropic’s most capable model and the gold standard for intelligence across the industry. It excels at tasks that require extended, multi-layered reasoning, deep synthesis, complex coding, or careful judgment under ambiguity. It is also the most expensive — best reserved for work where quality of reasoning genuinely matters.
Strengths
  • Industry-leading performance on complex reasoning and coding benchmarks
  • Exceptional at synthesizing large, conflicting, or ambiguous information
  • Supports extended thinking (deeper, step-by-step reasoning before answering)
  • Best-in-class for long, nuanced documents and high-stakes outputs
  • 1 million token context window — can hold an entire season's data
Limitations
  • Significantly more expensive than Sonnet or Haiku
  • Slower response time due to deeper processing
  • Overkill for most everyday tasks — ROI drops when used for simple jobs
Best Use Cases at AMG
  • High-stakes strategy documents, board presentations, or executive briefs
  • Complex financial modeling or multi-variable scenario analysis
  • Deep competitive intelligence synthesis across many sources
  • When Sonnet has given you a result that’s “good but not quite right”
  • Building sophisticated automation logic or multi-step agent workflows
Input~$15.00 / 1M tokens
Output~$75.00 / 1M tokens
~19× more expensive than Haiku, ~5× more than Sonnet. Use intentionally.
🪙
What is a "token" — and how does your usage add up?
Understanding the unit of cost behind every Claude interaction
Claude doesn't read text the way we do — it breaks everything down into tokens, which are small chunks of text roughly equal to ¾ of a word. As a practical rule: 1,000 tokens ≈ 750 words, or about one and a half pages of a Word document.
How a sentence is tokenized
Thequickbrownfoxjumpedoverthelazydog
↑ Each colored block is one token. 9 words = ~12 tokens (punctuation, spaces, and word endings count too).
1
You write a prompt — every word consumes input tokens. A longer, more detailed prompt costs more to send.
2
Claude reads your context — any files, documents, or prior conversation you paste in are also tokenized and billed as input.
3
Claude writes a response — every word it generates consumes output tokens, which are priced higher than input tokens across all models.
📧
Short email draft
~300–500 tokens total
📄
5-page report analysis
~3,000–5,000 tokens total
📊
Large spreadsheet paste
~10,000–30,000 tokens
📚
Full season's order data
~100,000+ tokens
💡 Practical takeaway: Most everyday tasks — drafting an email, summarizing a report, analyzing a spreadsheet — use far fewer tokens than you'd expect. The cost per interaction is fractions of a cent on Haiku and Sonnet. Tokens add up primarily when you paste in very large documents or have long back-and-forth conversations with extensive context.

🧩 Not sure which model to use?

Is your task repetitive, simple, or high-volume?
→ Yes  →  Use Haiku
↓ No
Does it require careful reasoning, drafting, or analysis?
→ Yes (most tasks)  →  Use Sonnet — your everyday default
↓ Yes, and it’s genuinely complex or high-stakes
Is this a board-level deliverable, deep strategy work, or an edge case where Sonnet fell short?
→ Yes  →  Use Opus — premium, use deliberately
💡
Cost is shared: Under AMG’s enterprise agreement, all licensed users draw from the same usage pool. Using Opus for tasks Sonnet can handle equally well is not just inefficient for you — it’s an unnecessary cost to the company. Choose the lightest model that gets the job done well.
03

Data Classification Tiers

Know what you can — and cannot — share with Claude

ℹ️
How your data is handled today: Prompts and responses are processed on Anthropic's servers. Logs are retained for up to 7 days, then deleted. Your data is not used to train Claude.
  • Sales recaps & brand analysis
  • Draft presentations & marketing copy
  • General research & competitive data
  • Publicly available retail / market data
  • SKU names & style descriptions

No restrictions. Use freely to draft, analyze, and iterate.

  • Wholesale pricing (non-contract)
  • Margin analysis (aggregated)
  • Brand performance summaries
  • Inventory / ATS data
  • Shipment & logistics planning

Permitted, but use good judgment. Aggregate numbers when possible and avoid including identifying account details unnecessarily.

  • Named vendor / factory terms
  • Customer-specific cost sheets
  • Royalty rates & license details
  • AR aging & customer credit terms
  • Chargeback dispute details

Manager approval required — or use Data Masking (see Section 3) to anonymize before sending to Claude.

  • Employee SSNs, payroll, or salary data
  • Customer credit card / banking data
  • Signed contracts under NDA
  • Personal health or benefits information
  • Passwords or system credentials

No exceptions. No masking workaround. This category of data must never leave AMG systems under any circumstances.

🔍 Quick Check — Is my data safe to use?

04

Mandatory Data Masking

Protect sensitive info before it ever leaves your machine

⚠️
Required for all Tier 3 data. Before sending any sensitive business data to Claude, you must run it through a masking script that replaces identifiable names, figures, and terms with anonymized placeholders.

How the Masking Workflow Works

1

Generate Your Masking Script

Ask Claude to build you a reusable masking script tailored to your data type.

💬 What to say to Claude
"Build me a Python script that reads a CSV file,
replaces all vendor names with Vendor_001, Vendor_002
etc., rounds dollar amounts to the nearest $5, removes
dates, and outputs a masked version."
↓
2

Run the Masking Script Locally

Execute the script on your machine. It produces a masked file (safe to send) and a local key file (stays on your machine).

Original (stays local)
Supplier Alpha Inc.$4.37 FOB
Beta Manufacturing$6.12 FOB
→
Masked (sent to Claude)
Vendor_001$4–5 FOB
Vendor_002$6–7 FOB
↓
3

Use Claude with Masked Data

Paste or upload the masked file. Claude can still analyze patterns and give advice — it just doesn't see real names or exact figures.

↓
4

Unmask Locally

When you get Claude's output, run the unmask script on your machine to restore the real values in the final document.

05

Practical Guidelines

Quick reference for everyday use

✅ What You Can Do Now

  • Draft, edit, and improve presentations, emails, and reports
  • Analyze sales recaps, brand summaries, or retail data
  • Run margin scenarios using aggregated numbers
  • Research competitors, retail trends, and market data
  • Build customer-facing materials (line sheets, pitch decks)
  • Ask Claude to build reusable scripts & templates

⚠️ Requires Approval or Masking

  • Customer-specific pricing or cost sheets
  • Vendor or factory cost details — use ranges
  • Named customer credit terms or AR details
  • Royalty rates and license specifics

❌ Never — No Exceptions

  • Employee personal information (salaries, SSNs, benefits)
  • Customer credit card numbers or bank details
  • Signed contracts subject to NDAs
  • System passwords, API keys, or security credentials

❓ Not Sure?

If you are unsure whether data is safe to use with Claude, contact IT (Paul Arking) before proceeding. This policy applies to all AMG divisions: USA Legwear, PVH Legwear, BR Intimates, and shared corporate functions.

06

Your Agreement

Acknowledge & activate your Claude license

By checking each item below and signing, you confirm that you have read, understood, and agree to follow this Claude AI Usage Agreement in full. Your Claude license will be activated upon submission.