Claude reads the paperwork. Your team approves the posting.
Claude is Anthropic's family of AI models. It reads invoices, remittances and emails well, keeps to the fields you ask for, and can point to the line each number came from. As an Anthropic Claude Partner, QServices builds Claude agents that run in Microsoft Foundry on your Azure, draft the ERP entry and wait for a named person to approve it.
Illustrative. Claude drafts; nothing posts until your AR team approves.Illustrative
DefinitionClaudenoun · AI model family
Also calledAnthropic ClaudeClaude APIClaude in Microsoft Foundry
Claude is a family of large language models made by Anthropic. In October 2026 the line-up is Haiku 5.5 for high-volume reading, Sonnet 5.5 and Opus 5.5 for reasoning and drafting, and Fable 5.1 for long, demanding agent work. All four read text and images, call tools and work with up to a million tokens of context.
Where to run it
Four ways to run Claude. One usually fits a Microsoft shop.
The model is the same everywhere. What changes is whose cloud it runs in, who bills you and which features you get.
Our default
Microsoft Foundry, hosted on Azure
You run Microsoft 365, Dynamics or Azure and want Claude on your Azure invoice.
Fit with MicrosoftBest
Data stays in your cloudStrong
FeaturesGood
You keep
Prompts and replies stay in Azure. Entra ID and Azure roles control access. Billed in Claude Consumption Units on your Azure invoice and counted toward your Azure commitment.
Watch for
No batch discount, Files API, code execution or Managed Agents. Pay-as-you-go quota for Opus 5.5 starts at 40 requests a minute (10,000 on Enterprise Agreements). Not on CSP subscriptions.
Anthropic API
You want every feature first, including half-price batch jobs and Managed Agents.
Fit with MicrosoftSome
Data stays in your cloudSome
FeaturesBest
You keep
The full feature set. No training on your data by default, and zero data retention on request.
Watch for
A separate contract and bill outside Azure. Keeping processing in the US costs 1.1x.
Amazon Bedrock
Your systems already run on AWS.
Fit with MicrosoftLow
Data stays in your cloudStrong
FeaturesGood
You keep
Claude runs on AWS-managed infrastructure under AWS data terms.
Watch for
Regional endpoints cost 10% more than global ones.
Google Cloud
Your data platform is on Google Cloud.
Fit with MicrosoftLow
Data stays in your cloudGood
FeaturesStrong
You keep
Since 22 October 2026, Claude operated by Anthropic can be bought through Google Cloud Marketplace with Google IAM.
Watch for
The newest route: check the features and regions you need.
Routes, limits and prices as published by Anthropic and Microsoft, checked 23 October 2026.
Which model
Pick the model by the job, not the headline.
Most back-office agents use two: a cheap, fast model to read every document, and a stronger one for the exceptions.
Model
Price per million tokens (in / out)
Best at
Where we use it
Haiku 5.5
$0.10 / $0.50 for prompts up to 100,000 tokens
Classification and extraction at volume
Reading invoices, remittances and emails
Sonnet 5.5
$2 / $10
Reasoning and drafting at a fair price
Matching exceptions, drafting replies and journals
Opus 5.5
$4 / $20
Anthropic's suggested starting point for most work
Hard exceptions and multi-step agents
Fable 5.1
$10 / $50
Long, demanding agent work
Rarely needed in the back office; keeps data 30 days
Claude API list prices, October 2026. Batch requests are half price and cached prompt reads cost a fraction of normal input. Claude 4.7 and later count about 30% more tokens for the same text, so compare costs on your own documents.
Who decides
What Claude does alone, and what waits for a person.
Anthropic's own guidance for agents is to keep people in control and ask before high-stakes actions. We write it down with your controller before we build.
ActionClaude does it aloneClaude drafts, a person approvesPeople only
Read invoices, remittances and emails
Read-only access to the mailbox and documents it needs.
Pull out the fieldswith a citation to the source line
Structured outputs keep every field in the schema; citations show where each number came from.
Match to open invoices and draft the entry
The draft sits unposted in the ERP.
Post cash, a payment or a journal
It moves money, so a named person on your team approves it.
Write off a short payment or issue a credit
A business decision, not data entry.
Change bank details or customer master data
A common fraud target. Claude can flag it, never change it.
Your data
What happens to the documents Claude reads.
Tick what your security team needs before the first agent goes live.
0 of 11 settled
With Anthropic
In Microsoft Foundry
In Copilot Studio and Microsoft 365 Copilot
From Anthropic's privacy centre and API documentation and Microsoft Learn, October 2026.
How a Claude agent goes live
From your documents to a live agent.
1Weeks 1 to 2
Assessment
We measure the documents, volumes and exceptions, and test Haiku and Sonnet on a sample of your real files.
Fixed price
2Weeks 3 to 6
Build on Foundry
The agent, its schema, the ERP lookups and the approval queue, in your Azure subscription.
3Weeks 7 to 8
Shadow run
Claude drafts alongside your team; nobody acts on it until the accuracy is proven.
4Then
Live, with an approver
Every draft waits for a named person, and we report cost per document and accuracy each month.
FAQ
What buyers ask about Claude.
Will Anthropic train Claude on our invoices and emails?
Not by default. Anthropic doesn't use data from its commercial API to train models. API data is deleted within 30 days unless its safety systems flag it, and zero data retention is available on request.
Can we run Claude inside our Azure tenant?
Yes. Claude in Microsoft Foundry, hosted on Azure, has been generally available since June 2026. Prompts and replies stay in Azure, access goes through Entra ID, and it's billed on your Azure invoice. Anthropic remains the data processor.
Which Claude model should a finance agent use?
Usually two. Haiku 5.5 reads and classifies documents at volume for $0.10 per million input tokens; Sonnet 5.5 or Opus 5.5 handles the exceptions and drafts the entry. We test both on your own files first.
Can Claude post directly to our ERP?
It can call your ERP's API, but we don't let it post. Claude drafts the entry, a named person approves it, and every step is logged.
What does Claude cost to run?
You pay per token. Reading a typical invoice with Haiku 5.5 costs a fraction of a cent; batch jobs on the Anthropic API are half price. On Azure, Claude is billed in Consumption Units of $0.01 each.
Should we use Claude or OpenAI's GPT models?
Both are strong. We test both on your documents and pick per task, and on Azure either can run in Microsoft Foundry under the same controls. See OpenAI for business agents.
Is QServices an Anthropic partner?
Yes. QServices is an Anthropic Claude Partner, as well as a Microsoft Solutions Partner and an OpenAI Select Partner.
QServices Inc. undertakes every project with a high degree of professionalism. Their communication style is unmatched and they are always available to resolve issues or just discuss the project.
Turn the Microsoft Licenses
You Already Own Into an AI Workplace
Join our live webinar on Sept 24 and see five ways to automate meetings, approvals, and
Claude
