Use cases by function6 use cases

Internal AI tools your team will actually open

Your employees are already using AI with company data. KPMG and the University of Melbourne found that 48% of employees have uploaded sensitive company information to public AI tools, and that bans do not work: 67% of employees at companies that ban generative AI report doing it anyway. The useful question is not whether people use AI at work but whether they can use it with company data safely.

The builds here answer that in different ways: an assistant that answers from SharePoint and Drive and respects who can see what, an MCP server that gives Claude, ChatGPT or Copilot scoped access to your ERP and CRM, a private AI gateway with single sign-on, redaction and budgets, and reporting that explains itself, from a Monday brief to plain-English questions over your database.

What they share is an approach. They live where people already work, usually Teams or Slack. They cite their sources or say they do not know. They enforce permissions at the moment of each question. And they are measured against real questions from your staff, collected before the build starts.

Internal tools and knowledge systems I can build

What makes an internal AI tool get used

  • It lives where people already work

    A separate portal gets visited twice. An assistant in Teams or Slack gets used daily. Custom internal Slack apps also keep the full API rate limits that unlisted third-party apps lost in 2025.

  • It cites, or it says it does not know

    Every answer links to the document it came from. When nothing relevant is found, the assistant says so, which is what earns trust for the answers it does give.

  • Permissions are enforced per question

    The assistant searches only what the person asking is allowed to open, checked at query time against the same groups your file permissions use.

  • Numbers come from code

    In reports and database answers, figures are calculated by queries you can inspect. The model explains them; it never computes them.

  • It is tested on real questions

    A set of real questions with known good answers runs on every change, so improvements are measured rather than assumed.

Frequently asked questions

Can I connect our ERP or CRM to ChatGPT or Claude?

Yes, through the Model Context Protocol (MCP). Many systems now have official MCP servers, including HubSpot, Slack, Xero and QuickBooks Online. For systems without one, such as SAP Business One or an internal database, a custom MCP server exposes a small set of scoped, logged tools that Claude, ChatGPT and Copilot Studio can all use.

How do I stop employees pasting company data into ChatGPT?

Give them something better. Bans are ignored, as the KPMG research shows. A private gateway with company single sign-on, approved models, redaction of personal data and secrets, and assistants that already know your context removes the reason to use a personal account.

Do we need Microsoft 365 Copilot or a custom assistant?

If most staff already have Copilot licenses and your knowledge lives in Microsoft 365, start there. A custom assistant makes sense for staff without licenses, for knowledge spread across several systems, for strict EU hosting requirements, or when you need behavior Copilot does not offer, such as answering only from approved documents.

How do I build an internal AI assistant for company documents?

Connect the document sources, mirror their permissions, index the content with metadata about owner and freshness, retrieve only what the asking user may see, and require a citation for every answer. Then test it on a few hundred real questions before rollout. The permission model is the part to get right first.

Also browse: Outlook and Microsoft 365 · Slack · Gmail and Google Workspace · SaaS · all use cases