# AI for Knowledge Management

Author: Marin Niehues, Niehues Consulting · Published: 2025-06-02 · Updated: 2026-11-16 · AI Adoption & Operations
Source: https://www.niehuesconsulting.com/blog/ai-knowledge-management/

AI for knowledge management makes existing company knowledge findable within the daily workflow, provided the sources are maintained and responsibilities are settled. The technology is ready. What separates a prototype from a tool that people use every day is the organization: knowledge needs owners and a quality standard. A fixed routine for keeping it current comes on top. This article describes a three-stage approach, places the tools in context and sets out the cost and timeline of a knowledge assistant. This article provides an overview and does not constitute legal advice.

## The types of knowledge an organization must distinguish

An organization holds more knowledge than its people can use. The reason lies in the form the knowledge takes. Knowledge management research has long distinguished explicit knowledge, which can be written down, from tacit knowledge, which stays bound to people. For the use of AI, three types follow, and each requires different handling.

- Documented knowledge sits in policies and manuals. AI unlocks it through semantic search and an answer that cites its source.
- Case knowledge arises in tickets and project reports. It is extensive and rarely cleansed. It contains the solutions to questions that have been answered before.
- Experiential knowledge sits in the heads of specialists and can only partly be written down. AI unlocks the part that interviews and structured handovers have documented. For the rest, the assistant serves as a signpost to the right person.

The work follows availability. Documented knowledge comes first because it exists and can be checked. Case knowledge follows once the quality of the first use case is established. Experiential knowledge requires a dedicated initiative with the specialists concerned.

The benefit is evidenced. A field study of more than 5,000 customer service agents shows that an AI assistant which makes the knowledge of experienced colleagues available during conversations raises productivity by 14 percent on average. For less experienced staff, the gain was 34 percent. AI for knowledge management therefore shortens onboarding above all.

**The quality of the source sets the ceiling for the quality of the AI answer. No model compensates for outdated or contradictory documents. Knowledge management with AI therefore begins with maintaining the sources, and the choice of model comes afterwards.**

## How do you introduce AI in knowledge management?

You introduce AI in knowledge management in three stages, Understand, Build and Roll Out, and each stage delivers a result on which the executive team decides.

Stage one, Understand: the analysis starts with the knowledge flows of the business units. Which questions cost time every day? Where do decisions wait for information that only one person can provide? From the answers, a business case emerges for each application, valuing the time saved and the shorter onboarding. A use case with a clearly named business unit and a manageable set of sources comes first.

Stage two, Build: the prototype works with the real documents and the real questions of the business unit. Subject matter experts define a test catalog in advance. An answer counts as correct when it states the facts accurately and names its source. For documented knowledge, [retrieval augmented generation](/blog/retrieval-augmented-generation-enterprise/) is at the technical center: the model receives the relevant passages for each question and formulates its answer on that basis. The solution takes over the access rights of the source systems and checks them with every question. Overly broad permissions in the source system thereby become visible, so a review of permissions belongs before launch.

Stage three, Roll Out: the transfer into operations defines who maintains the sources and who decides on new ones. Employee enablement is added. People learn which questions the assistant answers and when it still makes sense to check with experts. Adoption follows when the subject matter owners maintain the test catalog themselves.

**What a test catalog contains**
A test catalog consists of real questions from the business unit with the correct answer and the corresponding source. To start, 50 to 100 questions covering the range of daily work are sufficient. The catalog is defined before building and rerun with every change to the system. This keeps quality measurable even as models and documents change.

## Which tools are available for AI in knowledge management?

Three groups of tools are available for AI in knowledge management: assistants in office software, AI features in wiki and ticketing systems, and custom RAG solutions. Assistants such as Microsoft Copilot access mailboxes and document stores and are suitable when knowledge sits mainly there. What matters in the roll-out is covered in the article on [introducing Microsoft Copilot](/blog/introducing-microsoft-copilot/). AI features in wiki and ticketing systems unlock the content of that particular system. A custom RAG solution pays off once knowledge from several systems has to come together or the business unit wants to define sources, answer format and testing itself.

The choice follows a simple rule. If more than half of the relevant sources sit in one system, the initiative starts with that system's built-in AI feature. If knowledge is spread across three or more systems with different rights, a custom solution is the shorter route to reliable answers.

## Decision criteria for the executive team

Before approving an initiative, the executive team clarifies five points:

- Value: what time and which errors does missing knowledge cost today, measured against a specific process?
- Sources: are the documents in a condition from which reliable answers can emerge?
- Access: who may see which knowledge, and does this also apply to the assistant's answer?
- Accountability: which function owns the sources and the assistant?
- Operation: who funds and steers the application after the prototype?

**Routes to knowledge access**
Criterion | Classic intranet search | AI assistant without an operating model | With Niehues Consulting
Answer format | List of hits for users to evaluate | Formulated answer, sources unchecked | Answer with source citation, measured against the test catalog
Currency | Dependent on individual maintainers | Random | Fixed ownership per source with a set rhythm
Access rights | Managed at system level | Often unclear | Taken over from the source systems and checked before launch
Operation after the prototype | In place | Open | Transfer with responsibilities and metrics

## How much does an AI knowledge assistant cost, and how long does implementation take?

An AI knowledge assistant for one business unit reaches regular operation after three to six months, and its cost consists of the build, ongoing license or operating costs and the maintenance of the sources. The prototype with its test catalog is ready after four to eight weeks. Operationalisation with permission checks and integration adds two to four months. External support is typically billed in the German market at around 1,000 to 2,500 euros per consultant day.

Set the calculation before approval. The number of users multiplied by the time saved per week, the working weeks, the fully loaded rate and the usage rate gives the annual benefit. Ongoing license or operating costs and the maintenance time of the source owners are deducted. The one-off build divided by the remaining net benefit gives the payback period.

**Worked example with assumed values**
- 200 users each save 30 minutes a week, which over 45 working weeks comes to 4,500 hours.
- At 60 euros an hour and a usage rate of 60 percent, this gives 162,000 euros a year.
- Ongoing costs of 40,000 euros and maintenance of 20,000 euros leave a net benefit of 102,000 euros.
- A build of 80,000 euros therefore pays back in just over nine months.

## Operation: how a prototype becomes a tool

Success is decided in the phase after the prototype, where a large share of AI initiatives stalls. An assistant that convinces in testing loses value in daily use as soon as documents age and nobody reports errors. The organization therefore needs three roles: the source owner, who keeps the content of their area current; the person accountable for the assistant, acting as product owner; and a function that evaluates user feedback.

Steering relies on a few metrics that the business unit itself understands:

- Usage rate per business unit
- Share of answers with source citations that subject matter experts rate as correct
- Time to information compared with the baseline before roll-out
- Number of sources reviewed or corrected each quarter

Source quality remains the biggest lever, as the article on [data quality and AI success](/blog/data-quality-for-ai-success/) shows.

## Niehues Consulting as a partner for AI in knowledge management

[Niehues Consulting](/) is a consultancy for AI management and organization design based in Düsseldorf. Founder Marin Niehues works from the delivery perspective: he builds organizations that deliver and brings AI into operation. Projects range from VC-funded scale-ups to DAX 40 corporations, with engagements in 20+ countries on six continents.

For knowledge management with AI, Niehues Consulting works according to the Niehues Delivery Model. In Understand, business analysis settles the test question of who decides, who delivers and who uses, because without named source owners no reliable assistant emerges. In Build, the prototype must prove its benefit in the real process with real data, measured against the test catalog. In Roll Out, the solution moves into operation with internal or external teams, and the test question is: are users enabled and processes adapted? The full path is described in the article [From pilot to scale](/blog/from-pilot-to-scale-ai-operations/).

The entry point is the free initial consultation of 30 minutes. It clarifies the use case and the state of the sources. [Book an initial consultation](/#kontakt).

**What the executive team decides**
Knowledge management with AI succeeds as a delivery topic. The executive team selects a use case with measurable value and names the owners of the sources. It requires a test catalog and then steers operation through metrics.

**Sources**
- Brynjolfsson, Li, Raymond: Generative AI at Work, NBER Working Paper, 2023
- Bundesverband Deutscher Unternehmensberater: Honorare in der Unternehmensberatung, 2022

## Frequently asked questions

### What is AI-supported knowledge management?

AI-supported knowledge management makes existing documents and experience accessible through search and answer assistants. The assistant formulates an answer from the relevant sources and states where it comes from. Quality depends on how well the sources are maintained and on clear responsibilities.

### How does a company get started with AI in knowledge management?

Start with a single use case in one business unit whose sources are manageable. A short analysis is followed by a prototype in the real workflow, measured against a test catalog. Expansion to further areas then pays off.

### Which data may an AI assistant use?

An AI assistant may use the data that the person asking can access in the source system, provided data processing agreements and internal policy cover the provider. For the design, this means the assistant takes over the access rights of the source systems, so that nobody sees content through an answer that is closed to them there. Case-by-case legal assessment lies with data protection and legal advisers.

### How do you prevent wrong answers from a knowledge assistant?

You limit wrong answers with maintained sources, answers that cite their sources and a test catalog that is rerun with every change. Errors cannot be ruled out entirely, so the assistant openly reports when it finds nothing and refers to experts. For high-risk questions, sign-off by a person remains part of the process.

### How do you capture tacit knowledge with AI?

You capture tacit knowledge by first documenting it in interviews and structured handovers and only then making it available to the assistant. AI helps to prepare conversations and reveal gaps in the content. What cannot be written down, the assistant makes accessible by pointing to the right person.

### How does the executive team recognize success?

By the share of correct answers with source citations and by the usage rate. Time to information compared with the baseline is added. These metrics show whether the assistant is effective in daily work and whether source maintenance is functioning.
