Using Claude in business: a guide to getting started (2026)
This guide was published on November 1, 2026, and is based on the official Claude Help Center (support.claude.com), Anthropic announcements, and our Claude deployments with business teams. Anthropic updates Claude monthly; the settings names and offerings described here are those from the end of October 2026.
To use Claude in a company, five things are needed: a team offer (Team up to 150 people, Enterprise beyond), about ten settings made before opening access, clear rules of use on the data, a configuration by profession (projects, instructions, skills) and training on real cases. The tool itself can be mastered in an hour. What differentiates purchased licenses from used licenses is everything else that surrounds them. This guide walks through these steps in order, with precise settings, the first ten use cases to suggest, and the most common mistakes we see.
The essentials in five lines
- Claude is Anthropic's AI assistant. In a business setting, it is used to write and analyze long documents, produce files, and delegate entire tasks with Cowork.
- A personal account is not enough For work data: only the Team and Enterprise offerings provide centralized administration and a commitment not to train models on your data.
- Administrative settings are performed before the first user: single sign-on, Coworkconnectors, spending limits and role-based access.
- Claude configures himself by profession thanks to projects, instructions and skills, which avoid having to re-explain the context with each request.
- Adoption hinges on training : workshops on real cases from each team, then a measurement of usage at 30 and 90 days.
What Claude can do for a team
Claude is no longer just a chat window. Since September 16, 2026 on Pro and Max plans, and since October 23, 2026 on Enterprise plans, the chat and Cowork They now form a single Claude: you submit your request in the same place, and Claude decides whether to respond directly or launch a longer process in the cloud, using your files and connected tools. Specifically, a team can entrust him with:
| Function | What it allows | Example in business |
|---|---|---|
| Conversation and writing | Write, rephrase, summarize, translate | Transforming meeting notes into a shareable report |
| Document analysis | Read multiple long documents at once, up to 1 million context tokens | Compare three contracts and list the differing clauses. |
| File creation | Produce downloadable documents, spreadsheets and presentations | An Excel budget tracking file with formulas |
| Cowork (delegation) | Perform several steps leading to the deliverable, even with the computer closed. | Prepare the monthly report from five exports. |
| Projects | Grouping instructions, reference documents, and conversations on the same topic | A "Tenders" project including the response guidelines and client references |
| Skills & Tools | Record a repeatable procedure that Claude applies on demand | Formatting the summary notes according to the management template |
| Connectors | Connect Claude to Microsoft 365, Google Workspace, Slack, Salesforce and other tools | Find the latest version of a document in SharePoint |
| Docs, Slides and Design | Create and edit formatted documents, presentations, and visuals | A committee presentation based on a memo |
| Claude Code | Coding, automating, analyzing data in the terminal or application | A script that consolidates files every Monday |
For an overview of all our resources on the tool, guides, kits and training courses, see our page Claude in business.
Step 1: Choose the right offer
The first instinct is often to let everyone take out a Pro subscription on their credit card, then reimburse them through an expense report. It's quick, but the organization then has no visibility into the accounts, no administration, and no contractual guarantee on the data. For professional use, the choice is between two offers:
| Team | Enterprise | |
|---|---|---|
| Size | From 2 to 150 guests | Large organizations, with no ceiling |
| Price (excluding taxes) | Standard seat at $25 per month, $20 paid annually; premium seat at $125 ($100 annually) | Annual seat plus consumption at the API rate |
| Administration | Console, SSO, domain capture, spending limits | In addition: customized roles, access to role-based templates, group budgets |
| Conformity with laws and regulations | No training on your default data | In addition: Compliance API, audit logs, retention options |
| Cowork | Included, enabled by default, beta web and mobile versions | Included is cloud execution, which must be enabled by the administrator. |
Two tips from our deployments. Start with monthly payments to gauge actual adoption, then switch to annual billing: you save 20%, but you commit to a specific number of seats. And reserve premium seats for the few people who regularly reach their limits, usually those who delegate a lot. Cowork or use Claude Code. Details of each price are in our guide. Claude's Prize in 2026.
Step 2: Make the adjustments before opening access
Claude's default settings are designed for an individual user who wants to get started quickly. In an organization, about ten parameters should be decided before inviting the first collaborator, in the organization's settings:
| Setting | Default value | Our recommendation |
|---|---|---|
| Single Sign-On (SSO) and Domain Capture | To configure | Activate from the start: any account created with the company address joins the organization |
| Cowork | activated | Keep it, but open it first to a pilot group |
| Execution of Cowork in the cloud | Enabled on Team, disabled on Enterprise | To enable: Cloud sessions are tracked by the compliance API; scheduled tasks run without a computer being turned on. |
| Automatic approval of actions | authorized | Reserved for trained users; on Enterprise, it is not started by default. |
| "Always allow" for connector tools | disabled | Leave it disabled: every write action (send, modify, delete) requires validation. |
| Available connectors | According to the offer | Open the reading section first (SharePoint, Drive), then the writing section. |
| Plugins | Next is the switch Cowork | Define which ones are installed by default, available, or prohibited. |
| Spending limits | No | A limit per organization and per person, with alerts. |
| Access to models (Enterprise) | All models | The "Efficient" mode limits high-volume populations to Sonnet and Haiku. |
| Data Retention (Enterprise) | Standard | To be aligned with your archiving policy |
One important point to know regarding security: sessions Cowork Sessions launched locally on a workstation retain their history on that device and do not follow the standard retention policy. Enterprise administrators can retrieve them via the compliance API, for a maximum of 30 days. This is yet another reason to favor cloud sessions. We detail the legal implications in our article. Claude's Security and Compliance in Business: GDPR, AI Act and Governance.
Step 3: Establish the rules of use
Employees aren't asking for a twenty-page rulebook. They want to know, before their first request, what they're allowed to put in Claude. Four rules are enough to get started, which you can incorporate into your AI charter if you have one:
- The authorized data. A simple list, by sensitivity level: what goes into Claude without question (public documents, internal notes), what requires caution (customer data, anonymized HR data), what never goes there (according to your classification).
- The verification. All figures, legal references, and quotes produced by Claude are verified before publication. The AI remains a first draft, and the person who submits it is responsible for its content.
- Transparency. We say when a deliverable has been produced with Claude, at least internally, and always when it goes to a client.
- The actions. Sending an email, modifying a shared file or publishing content on behalf of someone is validated by humans, one action at a time.
These rules also fulfill a legal obligation. Since February 2, 2025, Article 4 of the AI Act requires organizations using AI to ensure their staff have a sufficient level of AI proficiency. Our check article 4 of the AI Act allows you to check in a few minutes where your organization stands.
Step 4: Configure Claude for each job
This is the most frequently overlooked step, and the one that most significantly impacts the quality of the responses. A "naked" Claude knows neither your company, nor your profession, nor your templates. Three levels of configuration provide this context once and for all:
personal preferences
In their profile settings, each user describes their role, team, and preferred response format in a few lines: length, tone, language, and style. Claude applies these preferences to all conversations. Furthermore, the system's memory function retains relevant information from one conversation to the next; each user can review and delete their saved information.
Team projects
A project groups together instructions, reference documents, and all conversations on a single topic. On Team and Enterprise, a typical project is shared with colleagues. For example, a purchasing team can create a "Supplier Consultations" project containing the purchasing policy, the specifications template, and the evaluation grid: every buyer working on it starts from the same foundation, without having to copy anything.
The skills
A skill is a written procedure that Claude applies when the situation arises: formatting a summary note, analyzing a contract according to your grid, preparing a report in your own format. It can be written out, or recorded by showing the procedure to Claude once. Grouped in a plugin, a team's skills are then shared with all its members from the console. Our article Claude Skills: 10 skills ready to install gives examples to follow.
Le Claude Configuration Kit provides, for each major job, the preferences, project instructions and the first skill to install.
Step 5: Start with ten quick-win use cases
A successful deployment begins with time-saving practices from the very first week, on tasks that employees already perform. Here are the ten we most often recommend at launch, tested in training sessions with business teams:
| job | Case studies | Example of a request |
|---|---|---|
| All | Meeting minutes | "Based on these notes, write a report including decisions, actions, responsible parties, and deadlines." |
| All | Response to a sensitive email | "Provide a firm but cordial response to this email, 120 words maximum." |
| Board | Summary | "Summarize these three reports on one page for a management committee, with three options and their risks." |
| Finance | Commentary on discrepancies | "Explain the discrepancies between budget and actual figures in this file, line by line, and highlight those that exceed 10%." |
| HR | Job description | "Draft a job description based on this information, using our attached template." |
| Legal | Contract review | "Compare this contract to our model and list the clauses that deviate from it, by level of risk." |
| Sales | Appointment preparation | "Prepare a profile on this prospect using their website and latest annual report, with five questions to ask." |
| Marketing | Content variation | "Adapt this article into a LinkedIn post, a newsletter, and three headlines." |
| Purchasing | Offer analysis | "Compare these three supplier offers in a table according to our evaluation grid." |
| Project | Progress report | "Based on this schedule and these emails, draft the weekly update for the sponsor." |
For each use case, the quality of the response depends primarily on three elements of the request: the context (who you are, who the deliverable is for), the expected result (format, length, tone), and the attached documents. Here is a complete request, which you can adapt:
I am a management controller in an industrial company with 800 employees. Here is the export of actual results as of the end of September and the 2026 budget. Write a one-page commentary for the finance director: the three most significant discrepancies, their probable cause based on the descriptions, and the questions to ask the relevant managers. Highlight any lines where the data seems inconsistent.
Other ready-made applications, categorized by profession, are in our AI prompt library.
Step 6: Moving from conversation to delegation with Cowork
Once the initial uses are established, the next benefit comes from delegation: entrusting Claude not with a question, but with an objective, and letting him carry out the steps. Preparing a report from five exports, conducting competitive intelligence and formatting it, sorting and renaming a hundred supporting documents: these are tasks that take several hours. Cowork runs autonomously, in an isolated virtual machine, and continues even if you shut down your computer.
Four rules make delegation reliable:
- Describe the deliverable, not the steps. "An Excel spreadsheet per supplier with the annual total and the evolution over two years" rather than a list of transactions.
- Provide the sources. The files, folder, or connector where the data is located.
- Establish a checkpoint. "Show me the plan before you start," or "Stop before you send anything."
- Review the result like that of a competent new colleague: Claude indicates what he has done and checked, but the final validation is up to you.
For missions ready to copy, see Claude Cowork Examples of tasks by profession and our guide Configure Claude CoworkIf your teams are hesitating between this and OpenAI's tool, see our comparison. Claude Cowork or ChatGPT Work details the differences.
Claude's best practices in business
These ten rules are featured in all our training courses. They fit on one page, which can be displayed or included in the welcome message for new users:
- One conversation per topic. A conversation that gets longer consumes more and more and mixes contexts.
- The context first. Who you are, who the deliverable is for, and what you will do with it.
- The expected format, explicitly. Length, structure, tone, file type.
- Documents rather than summaries. Providing the source is better than describing it.
- Ask for the hypotheses. "Tell me what you assumed" reveals the weak points in an answer.
- Iterate rather than start over. “Shorter”, “more factual”, “add an example” correct better than a new request.
- The right model for the task. Sonnet or Haiku for reformulating and classifying, Opus for complex analyses; see the right AI model for each task.
- One project for each recurring issue, with its instructions and reference documents.
- One skill for each repeated procedure more than twice a month.
- A human verification of everything that comes out of the team: figures, references, quotes, recipients.
Common mistakes
- Open all licenses at once, without a pilot or ambassadors. Six weeks later, a third of the accounts were never used.
- Training in the tool rather than in its uses. A menu demonstration doesn't tell a buyer what Claude changes in his day.
- Allow personal accounts to coexist with the team offer. Domain capture solves the problem.
- Open the writing of connectors from day onebefore users know how to review a proposed action.
- Use the most powerful model by reflex. on Enterprise, where consumption is billed: rewriting an email costs five times more with Fable than with Sonnet, for the same result.
- Do not measure anything. Without usage indicators, it is impossible to know which teams to support.
Train your teams in Claude
Learning Claude doesn't take days. The approach that works best in our deployments has three phases: a short awareness session for everyone (one hour, often a webinar), workshops tailored to specific roles and focusing on participants' real-world cases, and then a network of ambassadors who share skills and best practices within their teams. The key is customization: a workshop built around the use cases of a legal team is not the same as one for a sales team, even with the same tool.
Our Claude's training in the workplace This follows this logic: each session starts with the participants' real tasks, alternates between practical exercises and feedback, and ends with requests and skills that everyone can reuse the very next day. For delegation, the Claude's training Cowork goes further on long-term missions and their supervision.
Measuring adoption at 30 and 90 days
Three indicators are sufficient for management: the percentage of active licenses each week, the number of projects and shared skills, and the time saved reported on priority use cases. The administration console provides the first three. On Enterprise, the "Smart Reports," in beta since September 10, 2026, also analyze usage and cost per team and identify recurring requests to be converted into shared skills. Our Claude's Usage Analysis Kit proposes the tables and alert thresholds to be reviewed.
F.A.Q
Can Claude be used for free in a business setting?
Technically yes, a free version exists, but it's not suitable for work data: no administration, no contractual guarantee, limited use, and neither Cowork nor Claude Code. For a team, the Team offer starts at $20 excluding tax per person per month with annual payment.
Is Claude compliant with the GDPR?
For Team and Enterprise offerings, Anthropic commits by default not to train its models on your data, encrypts data in transit and at rest, and offers retention options and a data processing agreement. Compliance then depends on your usage: what data you store, and on what legal basis.
What is the difference between Claude Team and Claude Enterprise?
Team covers 2 to 150 people with a fixed price per seat. Enterprise is aimed at large organizations, combines an annual seat with usage-based billing, and adds custom roles, a compliance API, group budgets, and access to role-based templates.
Does Claude work with Microsoft 365?
Yes. The Microsoft 365 connector allows Claude to search SharePoint, OneDrive, and Outlook, and, since July 2026, to send emails and edit files, with administrator permission. Claude is also available in Microsoft 365 applications starting with the Pro plan. If your organization already uses Microsoft 365 Copilot, please note that Copilot Cowork It also works with Anthropic models.
How do I learn to use Claude?
The most effective approach is to practice on your own tasks: start with a real deliverable, ask Claude for it, compare, and iterate. Anthropic's Claude Academy training courses provide the foundation; in-house training built on your team's use cases saves several weeks of trial and error.
How long does it take to deploy Claude?
The initial setup takes a day. A pilot program with one or two groups lasts four to six weeks, the time needed to develop the first projects and skills and to train ambassadors. The rollout then follows over three months, team by team.
Sources
- Claude's Help Center, Use Claude Cowork on Team and Enterprise plans, accessed on November 1, 2026.
- Claude's Help Center, version notesSeptember and October 2026.
- Anthropic, Claude Cowork and chat come together for Enterprise, 23 October 2026
- Anthropic, Making Claude Cowork ready for enterprise.
- Anthropic, Claude's Offers Page, accessed on November 1, 2026.
- Regulation (EU) 2024/1689 on artificial intelligence (AI Act), Article 4 on AI governance.
To learn more: our comparison Claude or ChatGPT for workand all our free resources on the page Claude in business.