Claude for business: Claude Team, Claude Enterprise, pricing and privacy

The essentials in 5 points

• Rolling Claude out to a team is an architecture decision, not a license purchase. Five questions get settled before you launch.
• The choice of plan depends on real integrations and the level of centralized administration expected, not on the pricing grid. Enterprise starts at 20 seats.
• Where your data is processed: by Anthropic for the Claude apps, with no European option as of 16 October 2026; for the API, a European region is possible with Google Cloud or Amazon.
• The real risk is shadow deployment: personal accounts, sensitive data in unvetted tools, access invisible to the IT department.
• The answer is a five-step plan: scope, choose inference, connect at least privilege, govern access, train.

The short answer

The questions that matter are not about the product but about its integration: which plan for which usage, which connections to internal data, where inference runs, how to govern access, and how to train teams. Settling these before you launch avoids the most expensive scenario, the one where everyone adopts the tool on their own, with no framework and no visibility.

The way forward is simple: scope usage and the level of governance, choose the inference path that fits your cloud, connect to internal data at least privilege, centralize access administration, then train. The rest of this article walks through each of these decisions for a CIO preparing a team rollout.

Claude Team and Claude Enterprise: plans and prices

Update of 16 October 2026: public prices taken that day from claude.com/pricing and its French version, excluding taxes, per seat per month. Euro amounts in brackets are those shown to a visitor in France, where Enterprise stays priced in dollars. These prices should be rechecked before any purchase.

For Claude Enterprise, the seat price excludes consumption, billed on top at API rates, so it does not compare directly with a Team seat.

Plan	Price	For whom	What it adds
Claude Team, Standard seat	$20 (€18) annually, $25 (€21.04) monthly	2 to 150 people	Central billing and administration, single sign-on (SSO), Claude Code and Cowork, Claude in Slack, user-level and organization-level spend controls, usage analytics, organization-wide Skills deployment, up to $500 per month of pooled API credits, no training on your content by default
Claude Team, Premium seat	$100 (€90) annually, $125 (€105.23) monthly	Heavy users	Five times the usage capacity of a Standard seat
Claude Enterprise	$20 per seat, billed annually only, plus consumption at API rates, 20-seat minimum	Organisations that need fine-grained governance	Role-based access, account provisioning from your directory (SCIM), audit logs, compliance API for monitoring, IP allowlisting, configurable data retention, Claude Security in beta

Skills are written procedures Claude follows for a recurring task. Team’s API credits, offered since 7 October 2026, let you run your own applications and agents on Anthropic’s developer platform. Claude Security, reserved for Enterprise, scans the organisation’s code for vulnerabilities and suggests patches for review.

Enterprise also brings domain capture, which attaches accounts opened with the company’s email domain to the organisation, network-level access control, and a HIPAA-ready offering, HIPAA being the US health data regulation.

For reference, the individual Pro plan costs $20 (€18) per month billed monthly, $17 (€15) annually, and the Max plan is offered from $100 (€90) per month. These plans do not provide the central administration expected for a team rollout, nor a data processing agreement in the organisation’s name.

What the Claude offering includes in October 2026

The pricing page opens with an announcement: “Claude Cowork is now just Claude”. Cowork was the mode in which Claude carries out a whole task: it reads files, chains the steps and hands back a document. Since 16 September 2026, Cowork and chat have been merging into one Claude.

The rollout started with Pro and Max, as the banner still says. Team and Free come next, and Anthropic commits to notifying Enterprise admins at least 30 days before anything changes. On the pricing page, the Team plan still lists “Claude Code and Claude Cowork”: both come with the seat. The Claude Cowork guide covers task execution and its settings.

Claude Docs, Claude Slides and Claude Design, which produce documents, decks and designs inside the conversation, came out of beta on 8 October 2026. They are on every plan, including Free.

Claude Code, Anthropic’s coding agent, is included in every paid plan, from the Team Standard seat up. It draws on the same usage allowance as chat.

Team or Enterprise: the 20-seat threshold and data retention

Enterprise requires at least 20 seats. Below that, Team is the only plan administered by the organisation; above 150 people, Enterprise is the only option. In between, the choice turns on governance.

Three needs tip the balance towards Enterprise: creating and removing accounts automatically from the directory, traceability through audit logs, and control over how long data is kept. Team offers no retention setting. For an organisation bound by an archiving policy or professional secrecy, that point alone often settles the question.

The economics differ too. A Team seat includes a usage allowance; beyond it, extra consumption is paid at API rates. An Enterprise seat comes with actual consumption billed on top, which administrators cap with spend limits. Moving from Team to Enterprise also means losing the monthly API credits, which the pricing grid reserves for Team.

An estimate at 20 and then 50 seats

The following calculation rests on a working assumption, not on Anthropic data: consumption of $20 to $100 per user per month at API rates, depending on how intensively people use Claude. Prices are for annual billing.

Headcount	Team Standard, allowance included	Enterprise, seats	Enterprise, seats and consumption
20 seats	$400 per month (€360 excl. VAT)	$400 per month	$800 to $2,400 per month
50 seats	$1,000 per month (€900 excl. VAT)	$1,000 per month	$2,000 to $6,000 per month

On the Team side, the only addition is consumption beyond the included allowance. A pilot of a few weeks replaces this assumption with a real measurement.

Enterprise can be bought self-serve or through Anthropic’s sales team. Only the latter opens invoicing with deferred payment terms, purchase through AWS Marketplace, trials, consultation and dedicated support above certain spend thresholds. US-only inference is reserved for usage-based Enterprise contracts: older contracts billed per seat only do not have it.

Privacy, data and GDPR

Five points to check before any rollout:

• Model training. On Team, on Enterprise and for the API, Anthropic states that it does not train models on your content by default. Necessary, but not sufficient for data covered by professional secrecy.
• The data processing agreement. Anthropic’s Data Processing Addendum, effective 24 February 2025, is part of the commercial terms and includes the European standard contractual clauses for transfers outside the EU. In the European Union, services are provided by Anthropic Ireland Limited.
• Storage and processing. Storage is where data is kept; processing is where the model computes its answer. As of 16 October 2026, the Claude apps offer no European option. The only restriction available is US-only inference, on Enterprise, at a 10% premium; it does not change where data is stored. Through Anthropic’s direct API, storage is available in the US only, and inference can be set to “us” or “global”.
• Going through a cloud provider. For API use and in-house applications, Claude models are available through Amazon Bedrock, Google Cloud Vertex AI and Microsoft Foundry on Azure. Today, this route is the only way to keep processing in Europe, at a 10% premium: the details by provider are below.
• Retention. On Enterprise, the account owner can set a retention period of at least 30 days; without a setting, data is kept indefinitely. Team offers no retention setting. Memory across conversations is off by default on Team and Enterprise.

Compliance of the use itself remains your organisation’s responsibility: legal basis, entry in the record of processing, informing the people concerned, and the list of data allowed in the tool.

The five questions to settle before deploying

These five questions belong to leadership; the AI training for executives and boards covers them in half a day.

Which plan, for which usage? The question is settled from real usage, not from reading the pricing grid. The Microsoft 365 connectors and the Google Workspace add-on are available on paid plans, which already covers a large share of office needs. Plans designed for organizations stand apart through centralized administration: authorizing connectors, controlling access, visibility into usage. Start from the question “who needs access to what, and who administers it” to choose, not from the feature list.

Which integrations to internal data? Three mechanisms complement each other, each under least privilege: a connector opens only what the use case needs.

The Microsoft 365 connector now goes beyond search. Since 7 July 2026, Claude can draft and send email, manage calendar events, and create and update files in OneDrive and SharePoint. These write tools require consent from a Microsoft Entra administrator, Microsoft’s directory, and then enabling for the organization; Teams remains read-only.

Inside the Office apps, Claude for Microsoft 365 carries context from Outlook to Word, Excel and PowerPoint: according to Anthropic, a change propagates to every open file.

Since 6 October 2026, the Claude for Google Workspace add-on and the Google Docs, Sheets and Slides connectors are in beta on all paid plans. The add-on puts Claude in a sidebar next to the open file; the connectors create and edit Google files from Claude. On Team and Enterprise, the account owner must enable these connectors first.

For everything else in your information system, the MCP protocol, an open standard for connecting a model to tools, links Claude to your databases, your business tools, and your internal services.

Where does inference run? This is the most structural decision for an IT department. The direct Anthropic API is the simplest path. To stay inside an already validated cloud, Claude models are available through Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry on Azure. That last option removes the classic “new vendor” objection for organizations already on Azure: Claude fits into their environment, under the contracts and controls already in force.

A requirement to process data in Europe narrows the choice. Amazon Bedrock serves recent models in its European regions, including Paris, Frankfurt and Ireland, except Claude Fable 5.1, which is US-only on regional endpoints. On Google Cloud, regional endpoints such as europe-west1 stop at Claude Sonnet 4.6; only the “eu” multi-region endpoint serves recent models. Microsoft Foundry only offers a dedicated US zone: it does not provide processing in Europe.

How do you govern access and permissions? A team rollout only holds if administration is centralized. Administrator authorization of MCP connectors lets you validate once, at the organization level, the tools and data users connect to, without device-by-device setup. That is what moves you from individual tinkering to governed deployment: the IT department decides, and the user inherits the approved access on first connection.

How do you train teams? Without training, a technical rollout leads to weak adoption and risky usage. Training is not about learning where to click, it aligns teams on approved use cases, admissible data, and verification reflexes. It also makes the official path easier to follow than the workarounds. For business teams, the AI agents training takes place in the chosen tool, starting from their own tasks.

The mistakes of shadow deployment

The first risk lies less in choosing the wrong plan than in not choosing one at all. In the absence of an official path, staff adopt the tool on their own, with personal accounts. Conversations and files then escape all administration, and the IT department has neither visibility nor any means to act.

The second mistake follows close behind: dropping sensitive data into unvetted tools. A hurried employee pastes a contract, an HR file, or an extract from a customer database into a consumer account, outside any data processing agreement. The problem comes not from the tool but from the lack of a framework around it. The only durable defense is to offer an official alternative, administered and documented, that is more convenient than the workaround.

The third mistake is opening access that no one keeps track of. Connectors wired by hand, permissions granted case by case, and nothing to inventory them. Centralized access administration exists precisely to prevent this. A governed deployment does not remove usage: it makes usage visible and access revocable.

A five-step deployment plan

1. Scope usage and governance. Inventory who needs to do what with Claude, which data is involved, and what level of administration the organization requires. This scoping determines the plan, not the other way around. It is the purpose of an AI audit: five axes reviewed, costed roadmap.
2. Choose the inference path. Decide between the direct API and deployment inside your cloud (Bedrock, Vertex, Microsoft Foundry on Azure) based on your data residency requirements and the contracts you have in place.
3. Connect at least privilege. Enable the Microsoft 365 or Google Workspace connectors and the MCP connections you need, each access limited to the strict requirement. Start narrow, widen once usage is proven.
4. Centralize access administration. Authorize connectors at the organization level, so users inherit the right access on first connection, with no individual setup.
5. Train and support. Align teams on approved use cases and verification reflexes, then track adoption and adjust: this is the purpose of an AI enablement programme for teams. Follow-up continues after go-live.

Claude and the other business assistants

Several sources of intelligence are better than one, and many organisations run two assistants side by side. The same questions of pricing, data and governance are covered, vendor by vendor, in the guides to Mistral for business, ChatGPT Business, Gemini for business and Copilot Studio. For work delegated to Claude, the Claude Cowork guide takes this one further.

Where Colombani.ai fits

To set up a first use case shared by the whole team, the First agent offer trains your team on a case chosen with them and puts the agent into service on Claude.

For business teams and executives adopting Claude, the Claude training programmes offer a choice of courses for each audience. Each course page sets out its programme, prerequisites and delivery options. To compare providers before choosing, see how to choose Claude training in France.

These decisions are made better with someone who has already led this kind of rollout. Colombani.ai supports the scoping on the consulting side: architecture choices, inference path, access governance, and a deployment plan fitted to your information system.

For the technical teams who will run Claude day to day, the Claude Code in production course covers advanced usage, integration with internal systems, and the right security reflexes. It is delivered by a Qualiopi-certified provider; its courses are eligible for an OPCO funding request. The trainer, Ulysse Trin, holds the Claude Certified Architect Professional certification, issued by Anthropic.

Before standardising on Claude Code, it helps to know where it stands against the other command-line agents: the Claude Code, Gemini CLI, Codex, Hermes and OpenClaw comparison gives the 2026 verdict on cost, context, MCP and memory.

Sources

• Claude, plans and pricing and French version in euros, accessed 16 October 2026: prices, plan contents, Team API credits, Enterprise options
• Claude, release notes: Microsoft 365 write tools (7 July 2026), memory off by default on Team and Enterprise (25 August 2026), monthly API credits (7 October 2026), accessed 16 October 2026
• Claude, Cowork and chat are one Claude, 16 September 2026
• Claude, Docs, Sheets and Slides in Google Workspace, 6 October 2026
• Claude, Dashboards and Motion; Docs, Slides and Design out of beta, 8 October 2026
• Claude for Microsoft 365, accessed 16 October 2026
• Anthropic, enable US-only inference for your organization, accessed 16 October 2026
• Anthropic, custom data retention on Enterprise, accessed 16 October 2026
• Anthropic, API data residency, accessed 16 October 2026
• Anthropic, Claude in Amazon Bedrock, on Google Cloud and in Microsoft Foundry, accessed 16 October 2026
• Anthropic, Data Processing Addendum, effective 24 February 2025
• Anthropic, Getting started with Claude Code, accessed 16 October 2026
• Anthropic, LLM gateway configuration, accessed 16 October 2026
• Anthropic, Corporate proxy configuration, accessed 16 October 2026