AI knowledge assistant: preparing the base and launch criteria
What to prepare in your documentation before launching a corporate knowledge AI assistant, how to check its answers and what the launch costs
A corporate knowledge AI assistant answers employees from company documentation, so before launch you need clear sources, owners who keep them current and rules for checking answers. In the AKORDO case, an assistant built on OpenAI finds answers in Confluence through RAG search and works in Slack. In the catalogue, a corporate knowledge AI agent costs $1,600-3,300 and takes an indicative 3-6 weeks.
Which questions the assistant should cover
Start with a list of questions employees now ask colleagues or look up in documents by hand. In the AKORDO case, this was a first line of internal support where people searched corporate documentation themselves.
Group the questions by topic and mark the document that holds the answer to each. The assistant will not cover a question without a document until someone writes the answer down.
Sources, currency and owners
For each source, assign an owner who updates the document, and mark outdated pages. The assistant repeats what it finds, so an old instruction produces an old answer.
Before launch, prepare:
- a list of sources: wiki, internal instructions, policies;
- the owner of each source and how often it is updated;
- documents that not every employee may see;
- sample questions with correct answers for testing.
In the case, the knowledge base updates automatically and n8n orchestrates the processes. Agree the update routine for your own base before launch.
Access and handing over a difficult question
Decide where employees will ask questions. In the case, the assistant was connected to Slack, and the team kept its usual way of working.
Describe where the assistant sends a question it found no answer to, and which expert answers it. Under the AKORDO catalogue's principle, AI takes repetitive questions off the team and passes to a person whatever needs a live decision.
Checking answers before launch
Take the prepared sample questions and compare the assistant's answers with the correct ones. Record which source it used and where it got things wrong.
The AKORDO case runs answer scoring with continuous improvement of the AI. Agree the scoring method before launch, so employees know how to report an inaccurate answer.
If the team does not work with AI yet or is wary of it, the catalogue has training in the safe use of AI: $300-700, an indicative 1-2 weeks.
What the corporate knowledge case implemented
The AI assistant for corporate knowledge consists of an assistant built on OpenAI, RAG search across Confluence and internal documentation, Slack integration, automatic knowledge base updates, answer scoring and process orchestration through n8n.
According to the case, employees get answers in seconds instead of searching documentation by hand, new employees are onboarded faster, and the company is building a single reliable source of corporate knowledge.
This stack describes one implementation. Define the sources and the access channel for your company after a review.
Documentation as the basis of the work appears in other cases too: in product consulting for fintech teams, the AKORDO team prepared technical and product documentation, and in the AI audit for a fintech company it analysed customer support processes.
The article on AI agents for business gives wider context on AI agents and business knowledge.
Budget and expectations
According to the AKORDO service catalogue:
| Service | Indicative cost | Indicative timeline |
|---|---|---|
| Corporate knowledge AI agent | $1,600-3,300 | 3-6 weeks |
| Team training in the safe use of AI | $300-700 | 1-2 weeks |
AKORDO quotes the exact amount after a free consultation. The client pays for licences, API, hosting and telephony separately, and catalogue timelines are indicative and not contractual.
Questions before you start
Which tools did the case use?
OpenAI, RAG search across Confluence, Slack and n8n, in the roles described on the case page.
Does everyone need the same stack?
No. The case describes one implementation, and you define the sources and the access channel for your company after a review.
How much does the launch cost and how long does it take?
$1,600-3,300, an indicative 3-6 weeks according to the catalogue. The client pays for external costs separately.
Discuss your knowledge base through the AI solutions service or a free AKORDO consultation.