The Claude use cases that pay off in a business are the ones where Claude reads a lot and a person checks the result: finding discrepancies in documents, turning reports into analysis, drafting customer replies, and writing and reviewing code. Pick the model by task. Sonnet 5.5 or Haiku 5.5 covers everyday writing, Opus 5.5 covers analysis and daily coding, and Fable 5.2 is for the few large, open technical jobs.
I use Claude every day to build and run AIKI's own systems. The examples below are the ones Anthropic and its customers have published.
Claude use cases by department
| Department | Use case | Model that fits |
|---|---|---|
| Finance and accounting | Check attachments and statements for figures that do not add up | Opus 5.5 |
| Data and reporting | Write SQL and turn raw numbers into a report | Opus 5.5 |
| Legal and HR | Review documentation and flag what needs follow-up | Opus 5.5, with a person deciding |
| Sales and marketing | Research, first drafts, prototypes and campaign material | Sonnet 5.5 |
| Customer service | Draft replies and power a chatbot that hands over to a person | Sonnet 5.5 or Haiku 5.5 |
| IT and development | Write, test and review code; large technical projects | Opus 5.5, Fable 5.2 for the hardest jobs |
The model column is AIKI's suggested starting point. Anthropic itself recommends Opus 5.5 as the default for daily work, Sonnet 5.5 and Haiku 5.5 for simpler tasks in volume, and Fable 5.2 for long, demanding technical work.
Finance and accounting: finding what does not add up
The strongest finance examples are about catching errors a person could miss. In examples Anthropic published at the Fable 5.2 launch on 9 November 2026:
- Box had Claude go through the attachments to a loan application. It found that the salary deposits did not match the balance on the bank statement, so the case officer knew what to check.
- Fundamental Research Labs moved figures from an annual report into a forecasting model. Claude noticed that the regional numbers did not sum to the total, and got every cell right.
The pattern carries over to a small business: let Claude compare an invoice against the order or a budget against the actuals, and give the person the list of deviations to check. A person still approves anything that moves money.
Data and reporting: SQL that gives the right answer
A query that runs can still give the wrong total. In Anthropic's launch material, Oracle asked for SQL for a report on debt per country. Fable 5.2 returned the correct answer, while two other models wrote SQL that ran but summed the data wrongly and gave figures five to eight times too high.
Check one total by hand before the report goes to management, and ask Claude to walk through the query logic line by line.
Legal and HR: reviewing documentation
Legal AI company Harvey used Fable 5.2 to review documentation that employees have the right to work in the United States. Claude pointed out who needed follow-up straight away, what had to be corrected and who should do it.
HR teams can use the same approach on contracts and policy documents. In Europe this work usually involves personal data, so the data processing agreement and the provider's data retention terms come first (see the checklist at the end).
Sales and marketing: research, drafts and prototypes
Sales and marketing tasks are short and many, so the cheaper models fit: research on a prospect before a meeting, a first draft of an email, or a summary of a long tender document. Sonnet 5.5 costs a quarter of Fable 5.2 per token (2 and 10 dollars against 8 and 40 per million tokens in and out).
For visual work, Claude Design makes prototypes from a description, and the connectors for creative work let Claude work directly in the tools a marketing team already uses.
Customer service: drafts and chatbots
In customer service, Claude can draft replies for an agent to approve, or answer customers directly in a chatbot trained on your own content. Repetitive questions in high volume suit Haiku 5.5 or Sonnet 5.5. Complaints and complex cases should reach a person quickly.
Two rules apply in any setup. Under Article 50 of the EU AI Act, customers in the EU must be told from 2 August 2026 that they are talking to AI, unless it is obvious; in Norway the AI Act is not yet law (status for Norwegian businesses, in Norwegian). And the chatbot needs a clear route to a human. How to choose between a chatbot and staffed chat is covered in AI chatbot vs live chat, and AIKI builds AI chatbots for customer service.
IT and development: from daily coding to large projects
Software development is where I use Claude most. I counted what Claude Code did for me over the 14 days up to the Fable 5.2 launch: 1,199 sessions and about 328,000 tool calls from me and my agents. 87 percent of the calls were reading, writing and testing code, git and other terminal work. Opus 5.5 wrote 74 percent of the replies and Sonnet 5.5 25 percent.
On the largest jobs, the top model has published results: Uber ran a 10 to 12 hour security review with around 170 agents going through the code, and got a report, ready fixes and proposals for new guidelines.
Most steps run well on Sonnet 5.5 or Opus 5.5. My setup lets Sonnet 5.5 or Opus 5.5 do the work and ask Fable 5.2 for advice at the turning points: before a plan is locked, when something has been stuck for a while, and before a task is reported as done. Anthropic describes this as the advisor pattern. The setup and what it costs are in my Norwegian article on Claude Fable 5.2.
Claude Team and Enterprise plans
For the chat app, Anthropic sells two business plans (claude.com/pricing):
| Team | Enterprise | |
|---|---|---|
| Size | 2 to 150 people | Organisations that need more control |
| Single sign-on (SSO) | Yes | Yes |
| Model training on your content | None by default | None by default |
| SCIM, audit logs, role-based access | No | Yes |
| Custom data retention | No | Yes |
Teams of 2 to 150 people fit the Team plan. Enterprise is worth it when IT needs user provisioning, audit logs or its own retention rules. Claude also runs inside the tools the team already uses: Claude in Slack joins the channels, and Claude Code runs on developers' machines. If you are choosing between assistants, ChatGPT vs Claude compares them for business use.
What to check before rolling out Claude in a business
- Data retention. According to Anthropic's documentation on data retention, Fable 5.2 requires data to be stored for 30 days and is not available under a zero data retention agreement without special approval. If you use Claude through Amazon Bedrock or Google Cloud, the stored data stays with the cloud provider.
- Data processing agreement. If Claude will see personal data or confidential material, settle the agreement with the provider before the first real document goes in.
- The right model per task. The API list prices per million tokens in and out are 0.10 and 0.50 dollars for Haiku 5.5 (under 100,000 tokens in), 2 and 10 for Sonnet 5.5, 4 and 20 for Opus 5.5, and 8 and 40 for Fable 5.2 (Anthropic pricing). Defaulting everyone to the top model multiplies the bill.
- Cost per finished task. Measure what a finished task costs. An expensive model that gets it right the first time can cost less than a cheap one that needs five attempts.
- Training. Teams get more from Claude once they know which tasks to give it and how to check the output.
If you want help finding which tasks in your business are worth automating first, an AI consultant can go through the workflow with you, and AIKI builds the solutions through AI development and trains employees through AI courses.
Sources
- Anthropic: Introducing Claude Fable 5.2 and Claude Mythos 5.2, 9 November 2026 (customer examples from Box, Fundamental Research Labs, Oracle, Harvey and Uber; the advisor pattern)
- Anthropic: Pricing and models overview
- Anthropic: API and data retention
- Claude: Plans and pricing (Team and Enterprise features, read 11 November 2026)
- AIKI: EU AI Act for Norwegian businesses (in Norwegian, with primary sources for the Article 50 date)
- AIKI: Claude Code usage figures from my own sessions in the 14 days up to 9 November 2026, published in Claude Fable 5.2 (in Norwegian)
