If your team works in Microsoft 365, test Microsoft 365 Copilot first. It may already cover answers from company documents and help inside Office. A custom AI assistant becomes worth considering when suitable extensions still leave specific workflow, interface or hosting requirements unmet.
The answer is already somewhere: in the quality manual, an email or the folder only one colleague knows about. With AI knowledge management, your team asks directly: “What approval do I need for this order?” What matters is whether the system finds the current answer and a source people can check.
Copilot Chat, Microsoft 365 Copilot or a custom assistant?
- Copilot Chat: For web research and working with files you provide. Included with eligible Microsoft 365 subscriptions without an extra Copilot licence fee. The base chat does not automatically search all company knowledge; attached files, open content in supported apps and connected agents can supply internal sources. Microsoft explains the differences.
- Microsoft 365 Copilot: For everyday work with company knowledge, email, meetings and Office apps. The paid licence adds work context and features in Word, Excel, PowerPoint, Outlook and Teams. You also need licences for the apps you use; Teams requires a Teams licence. Microsoft's feature comparison.
- Extend Copilot: For external knowledge, check Copilot connectors first; for specific tasks, consider an agent in Copilot Studio. A CRM connection or another workflow step does not by itself require a separate application.
A custom assistant fits when those options leave an important requirement unmet. That might be an interface inside your business software, precise approval rules across several systems or a required hosting model. Ask for a trial that demonstrates the gap. Development, testing, hosting and maintenance then belong in the price; a custom solution is not automatically cheaper or more secure.
Already using ChatGPT Business, Enterprise or Edu? Also test Company Knowledge with your supported sources and their permissions. A ready-made knowledge platform is another option when search and source management are the main job.
What does Copilot cost for a business?
Checked 9 September 2026. Microsoft direct prices for the US market, in USD per user per month; local prices and applicable taxes vary:
- Microsoft 365 Copilot Business: regular price $21 with an annual commitment, paid yearly, for up to 300 users. Microsoft advertises $18 for the first year: available 1 July–31 December 2026 to existing customers with an eligible Business plan. The monthly subscription costs $25.20. Business prices and promotion terms.
- Microsoft 365 Copilot, enterprise offering: $30 with an annual commitment, paid yearly. This is a different licence offering from Copilot Business. Enterprise price.
Both prices cover the add-on. The base subscription costs extra. Copilot Business requires Microsoft 365 Business Basic, Standard, Premium or Apps for Business. Eligible enterprise prerequisites include Microsoft 365 E3/E5 and Office 365 E1/E3/E5. Check existing bundles before buying another licence. Microsoft's licence prerequisites.
Agents have separate billing rules. Certain internal agent use under a licensed Copilot user's identity is included within fair-use limits. Other usage and independently triggered flows can consume Copilot Credits. Budget for the actual usage and features. Copilot Studio billing.
Compare the total cost with the time your team loses today. The process cost calculator shows the annual effort. Use the supplier comparison template to assess existing tools and custom solutions against the same requirements.
Three things a good knowledge system must do
Find the right source. Your team needs the current work instruction, not a similar-sounding PDF from three years ago. A good system connects existing repositories and keeps its search index current. The original remains the authoritative source.
Make the answer verifiable. “According to the manual” is little help. A link to the right section is useful. Employees can check the statement immediately and read the source themselves for important decisions. If the documents do not support a reliable answer, that must be clear.
Respect permissions. Anyone who cannot open the original document should not learn its contents through AI. Access rights must therefore limit which sources are selected in the first place. A friendly answer interface alone cannot solve this.
How your documents become an answer
The system makes your existing sources searchable, such as SharePoint, a wiki or a file server. When someone asks a question, it retrieves the relevant passages and passes them to a language model. The model uses them to compose an answer and points back to the sources. This method is called retrieval-augmented generation, or RAG.
Your company knowledge does not need to be trained into a model. Source quality, search and the handling of conflicting or missing information are what matter. The model running in the background is one part of the solution.
Test the questions that cost you time today
Start with one department and the questions that keep coming up there. Give the system real documents and review the results with the people who use them:
- Does it find the right, up-to-date information?
- Does the answer cite a specific source?
- Do confidential contents remain inaccessible to unauthorised users?
- Does the answer help someone complete the next task faster?
Include a question the documents cannot answer. This is where you see whether the system behaves reliably.
Hosting, data processing and ongoing operation belong in the decision from the start. We integrate suitable existing tools or build the connections you need, with clear roles and controlled data flows. Explore our solutions for knowledge management and internal assistants.
From the first workshop to knowledge in daily use
At Kirsch & Drechsler, an opportunity assessment and workshops grew into an ongoing partnership: internal knowledge management, a website chatbot and joint development of CherryGate. It shows how individual AI ideas can grow into a foundation the company keeps using. See our work with Kirsch & Drechsler.
Ready to introduce this to your team? Our internal AI assistant service covers source integration, access controls, testing and team introduction.
