Solutions / Data

AI on company data

Your team gets answers from your own documents and data, with the source cited.

How we deliver it

• A document assistant: staff ask questions of your policies, contracts and wiki, and every answer cites the document it came from.
• Natural-language search across business data, such as orders, tickets and customers, instead of building a report.
• Document automation: incoming emails, invoices and forms are read and filed into the right system, with a person approving the exceptions.
• The AI you already pay for, set up properly: Microsoft 365 Copilot or Gemini for Workspace, with permissions that match who can see what.

Models follow how sensitive the data is: hosted APIs such as Anthropic Claude, OpenAI or Azure OpenAI by default, and open-weight models on infrastructure you control when the data must not leave it.

Where your data lives is your decision, per project: in your own cloud or Microsoft or Google tenant, hosted by us, or sent to a model provider’s API.

How it works

We start from one useful question your team asks every week, not from a model. The answer has to come from your own documents, respect who may see what, and show where it came from, so people can check it. Where your data may go decides the setup.

How an answer is made

The assistant does not answer from memory. It finds the passages in your documents that match the question, keeps only those the person asking is allowed to read, drafts an answer from them alone, and shows the sources next to it.

Automation with a person in the loop

Reading emails, invoices and forms and filing them in the right place saves the most time. Anything the system is unsure about goes to a person, so a wrong guess never lands in your books.

What decides it

The first use case

One question or one document flow that costs real time every week. It proves the value, and the setup, before anything is extended.

Where the data may go

Your own tenant, servers we host for you, or a model provider’s service. Contracts, regulation and how sensitive the documents are decide, case by case.

Permissions

The assistant must never show someone a document they could not open themselves. That depends on access being right to begin with, which is single sign-on and well-kept groups.

Checking it is right

A set of real questions with known answers, run before launch and after every change, so quality is measured instead of assumed.

Cost per use

Hosted models charge for every question; self-hosted ones cost servers. Usage decides which is cheaper, and it is worth estimating before choosing.

Hand over or run

Your choice, per solution. We hand it over with documentation your team can work from, or we keep it running: upgrades, monitoring and incidents, with one of us on call 24/7.