AI Knowledge Base
One source of answers, so staff and customers are told the same thing.
What is AI knowledge base?
An AI knowledge base is a single answer source built from your existing documents, connected to every channel where questions arrive, so that updating one document changes the answer everywhere at once.
When answers live in different heads and old documents, customers get told different things by different people, and neither the customer nor the business can tell which version was right.
We build one knowledge base from material you already have, then put it behind whatever people already use: the chatbot, the support inbox, the staff intranet.
When something changes, you update it once and every channel serves the new answer immediately. That single-update property is the whole reason for building it.
Why staff and customer content stay apart
Duara builds knowledge bases from Nairobi for organisations where the same question receives different answers depending on who is asked.
We separate staff-facing and customer-facing content from the start, and we do not merge them later on request. Internal notes about pricing flexibility, difficult accounts or margin belong nowhere near a customer channel, and a single pool with filtering applied afterwards eventually leaks.
Who this is for
- Staff give inconsistent answers
- Product and price information is scattered
- New starters take months to become useful
- The same questions circulate endlessly
Industries we serve
- Telecommunications
- Insurance
- Retail and eCommerce
- Utilities
- Education providers
- Healthcare groups
Problems this solves
These four appear together in any organisation where knowledge has never had a single owner.
- Different answers to the same question. Two staff give a customer different information, and neither knows which one was correct.
- Documentation nobody trusts. A manual exists but is out of date, so people ask a colleague instead, which guarantees it stays out of date.
- New starters dependent on interruption. Onboarding takes months because everything must be learned by asking somebody who is busy.
- Knowledge concentrated in a few people. A handful of long-serving staff hold the answers, and their absence stops work.
What we deliver
A single knowledge base assembled from your existing material, serving every channel, with separate internal and customer content and reporting on the questions it cannot answer.
One knowledge base built from your existing documents
The same answers served to chat, support and staff tools
Separate internal and customer-facing content
Update once and it applies everywhere immediately
Reporting on questions with no good answer yet
Source links so any answer can be traced and verified
Features
- Built from what you already have. Existing manuals, policies and past replies become the source, so nothing has to be written from scratch first.
- Separated audiences. Internal and customer content are held apart structurally, not filtered at the point of answering.
- Single point of update. Change the source document and every channel serves the new answer, which is the entire reason for a central base.
- Gap reporting. Questions asked that could not be answered are listed, which is a far shorter and more useful list than documenting everything.
- Traceable answers. Every response links to the document it came from, so staff can verify before repeating it to a customer.
Benefits
- Consistent answers everywhere. The same question receives the same answer regardless of who or which channel is asked.
- New staff become useful faster. Answers are available without interrupting a colleague, which shortens onboarding materially.
- Updates actually propagate. A price or policy change reaches every channel at once rather than gradually and incompletely.
- Gaps become visible. You learn what customers ask that you have never documented, which is usually a short and surprising list.
Technologies we use
- Document ingestion. Existing manuals, policies, past replies and product data pulled in as the answer source.
- Audience separation. Internal and customer content held in structurally distinct sets rather than tagged within one pool.
- Retrieval with citation. Answers assembled from source documents with the origin returned alongside.
- Channel connectors. The same base serving chat, helpdesk and internal tools without duplicating content.
- Gap analytics. Logging of unanswered questions to drive what gets written next.
Content structure matters more than model choice. A well-organised base answered by a modest model beats a disorganised one behind an expensive one.
How the process works, step by step
A knowledge base is usually answering questions within four to five weeks, with content organisation taking most of the time.
-
Step 1
Free content audit
We look at what documentation exists, how current it is, and how much would need rewriting before it is usable.
-
Step 2
Audience split
We separate internal from customer-facing material, which is easier to do at the start than to retrofit.
-
Step 3
Ingestion and structuring
Documents are brought in and organised so retrieval returns the right passage rather than a vaguely related one.
-
Step 4
Internal launch
Staff use it first, which surfaces wrong and missing answers without a customer seeing them.
-
Step 5
Customer channels
The customer-facing set is connected to chat and support once the internal set has settled.
Quality assurance and how we work
What we commit to on a system that becomes your single source of truth.
- Internal and customer content stay separate. They are held in distinct sets rather than one pool with filtering, and we decline requests to merge them later.
- Staff use it before customers do. Wrong and missing answers surface internally first, where the cost of an error is a correction rather than a complaint.
- Every answer cites its source. Staff can verify before repeating anything to a customer, which is what makes the base trusted rather than tolerated.
- Gaps are reported, not hidden. Questions the base could not answer are surfaced weekly, because that list is the most useful output it produces.
- You own the content. Everything remains yours in its original form, exportable, and never used to train public AI models.
Frequently asked questions about AI knowledge base
How do we keep it current?
You edit the source documents. Gap reporting also shows what people asked that it could not answer, which tells you exactly what to write next.
Can some content stay internal?
Yes, and internal and customer content are held in structurally separate sets rather than one pool with filtering applied.
Why not use one pool with permissions?
Because filtering applied after retrieval eventually leaks. Internal notes about margin or difficult accounts belong nowhere near a customer channel.
What if our documentation is out of date?
That is the first thing we audit. Where the existing material is too stale to use, we say so before building on it.
Do we have to write everything first?
No. It is built from what you already have, and gap reporting then tells you the short list of what is genuinely missing.
How long before the knowledge base is usable?
Four to five weeks, with content organisation taking more time than the technical build.
Why do staff use it before customers?
So wrong or missing answers surface where the cost is a correction rather than a complaint.
Can it serve our chatbot and helpdesk together?
Yes, and that is the point. One base behind every channel means an update propagates everywhere at once.
Does it show where an answer came from?
Yes, with a link to the source document, so staff can verify before repeating anything to a customer.
What is a content gap?
A question people ask that no document currently answers. The list is usually shorter and more surprising than businesses expect.
Can it handle product and pricing data?
Yes, though live pricing is better connected to the source system than copied, so it cannot drift.
Does it search Swahili documents?
Yes, including questions asked in Swahili against documents written in English.
Who should own it internally?
A named person, usually in operations or support. Knowledge bases owned by a committee stop being updated.
What if two documents disagree?
The conflict is flagged rather than silently resolved, because picking one arbitrarily is how wrong answers become authoritative.
Can we export the content?
Yes, at any time and in its original form. You are not locked into the system holding it.
Does it replace our intranet?
It can sit behind it. Most clients keep the intranet and let this answer the questions people would otherwise ask a colleague.
How do new staff benefit?
They get answers without interrupting anybody, which is usually the largest single reduction in onboarding time.
What happens when a policy changes?
You update the one document, and every channel serves the new answer immediately rather than gradually.
Is the content used to train AI models?
No. It stays yours and is never used to train public models.
What does the knowledge base cost to maintain?
Document volume and question rate determine it. We project the monthly figure from your own numbers before anything is agreed.
Glossary of terms
Knowledge base terms in plain language.
- Knowledge base
- A single organised source of answers, used by every channel rather than duplicated per channel.
- Single source of truth
- The principle that a given fact lives in exactly one place, so updating it updates everything.
- Retrieval
- Finding the relevant passage in your documents to answer a specific question.
- Citation
- The link back to the source document, allowing an answer to be verified.
- Content gap
- A question people ask that no document currently answers.
- Audience separation
- Keeping internal and customer-facing content structurally apart rather than filtering one pool.
Related services in AI Implementation
Talk to us about AI knowledge base
Ask three staff the same customer question and compare the answers. That exercise usually settles whether you need this.
Start your project