Solutions · Artificial intelligence

Claude

Claude is the artificial intelligence assistant developed by Anthropic. In companies it is used to query internal documents, to write and review texts and to add automated steps to processes that are manual today. It is available both as an application for people and as a service that the systems you already use can call via API.

How it works

Claude at a glance

An assistant on company documents

You upload contracts, specifications, manuals, procedures or minutes and ask questions in plain language. Claude answers based on those texts and points to the passages it drew the answer from. It is the most concrete use case for those whose information is scattered across folders, email attachments and document management.

Writing and reviewing texts

First draft of an offer, rewriting a procedure in understandable language, summary of a meeting, consistency check between a document and its previous versions. The text produced remains working material: responsibility for what leaves the company stays with the person who signs.

Analysis of long texts

Claude works with large context windows, meaning it can keep many pages under review in the same request. It helps when two versions of a contract must be compared, relevant clauses extracted from a specification or a set of documents summarised without splitting them by hand.

Integration into processes via API

Besides use through the interface, Claude can be called from software: classifying incoming emails, extracting fields from a PDF and passing them to the ERP, preparing a reply that an operator then approves. This is where artificial intelligence stops being a personal tool and becomes part of a workflow.

Where data ends up

It is the first question to ask, before choosing the model. Anthropic’s business plans and API access provide that submitted content is not used to train the models, with administrative controls over users and retention. Free and personal versions follow different terms: they are not the same thing and should not be used with company data.

What Claude is not

It is not a source of truth and does not replace a management system. It can produce wrong answers stated confidently, especially about data it does not have in front of it. It should be placed where errors are visible and can be corrected by the people doing the work, not where output goes straight to a customer or a payment without checks.

In detail

Claude features

Claude is used in three different ways, which often coexist in the same company. The first is the web or desktop application, which people use every day on texts and documents. The second is shared workspaces, where a department’s or project’s reference documents are collected and standing instructions are given, so answers are consistent and do not depend on how each person phrases the question. The third is the API, which a developer uses to embed the model in an existing application.

Choosing among these three is not a technical decision but an organisational one: it determines who sees what, how much it costs and how governable it is. On data residency and processing it is worth reading our page on Digital sovereignty: some sectors and some types of data call for different choices from those made for generic use, and the decision must be taken before going into production, not after.

  • Conversation on uploaded documents: you attach text files, PDFs and spreadsheets and ask Claude to work on that material rather than on its general knowledge.
  • Workspaces with standing instructions: the context is defined once (who we are, what tone we use, which documents apply) and every conversation in that space inherits it.
  • Extended context: many pages can be assessed in the same request, useful for comparing versions and analysing contractual or technical documentation.
  • Analysis and transformation of structured text: extracting fields, reclassifying, normalising lists that are retyped by hand today.
  • Support for writing code and queries: useful for those in the company who maintain reports, integrations and small automations, with human review before going live.
  • API with parameter control: you decide which model to use, how long the output should be and how to handle cases where the model cannot answer.
  • Administrative user management in business plans: creating and revoking access, visibility on usage, separation between test and working environments.
  • Business contractual terms: for business plans and the API, submitted content does not feed model training, and retention policies can be defined.
With Aesir

How we put it into operation

We do not start by activating licences. We start from the process: which activity takes up time today, who does it, on which documents, and what happens when the result is wrong. That shows whether Claude is really needed, in which form (assisted use by people or automation via API) and with which data perimeter. Many requests that arrive as “we want artificial intelligence” are better solved by fixing a workflow or an archive; when the case holds up, the setup is sized and the plan suited to the number of people and the type of data handled is chosen.

Implementation follows: configuring workspaces and reference instructions, controlled document upload, defining who accesses what, integration with existing systems where planned, and training the people who will use it, which is the most often underestimated part. Then we stay: monitoring real usage, correcting instructions when answers are not as expected, updating when models and terms change, ongoing support with one contractual point of contact for licences, integration and support.

Claude is part of Artificial intelligence. See also the Aesir method and our success stories.

Want to know whether Claude makes sense for your company?

We start from the process, not the product. A first meeting with no commitment.

Request a call