Manufacturingexample use case
Company document search — an answer with its source in seconds
Policies, price lists, contracts and procedures live in SharePoint, Google Drive, Teams and a network drive — and people still ask colleagues because they cannot find anything. An internal AI assistant (an AI chatbot over your company documents) answers questions asked in plain words directly in Teams or Slack, and attaches the document and page every answer came from.
Sound familiar?
- A new technician or colleague asks the same thing the previous one did — the answer exists, nobody can find it.
- Knowledge is scattered across SharePoint, Google Drive, Confluence and email.
- You tried uploading PDFs to ChatGPT and it started inventing numbers and clauses.
- Security or legal stopped the project because nobody could say who gets access to what.
How it works
1. Inventory and connect the sources
We connect SharePoint, Google Drive, Confluence, Notion, network drives or a database. We find what is outdated and what contradicts itself — usually the most important finding.
2. Permission-aware indexing
Documents are processed so the system remembers who may see them. An answer is never built from a source the user cannot access.
3. An answer with a citation — or “I don’t know”
Every answer comes with the document, page and excerpt. When there is no source, the system says so instead of making one up.
4. Deployed where people already work
In Teams, Slack, the intranet or your own app. A tool that needs a separate login never gets adopted.
What the system handles
- PDF, Word, Excel, PowerPoint, scans, wiki pages, tickets and emails.
- Slovak, Czech, English and German — the question and the documents can be in different languages.
- Document versions: prefers the newer policy and flags a conflict with the older one.
- Audit log: who asked what and what they received — for compliance and ISO audits.
- Runs in an EU region or fully on-premise if data must not leave the company.
Human oversight and safety
- No answer without a source — a base rule, not an optional feature.
- Permissions come from your identity system (Microsoft Entra ID, Google Workspace), never configured a second time by hand.
- Quality is measured on 50–150 real questions from your company before launch and after every change.
- Content owners see what people ask and where material is missing — and can fill the gaps.
What result is realistic
2–4 h
saved per person per week
For a manufacturing or service company with scattered documentation we plan for 2–4 hours saved per week per person who currently searches or asks colleagues. The effect is even bigger in onboarding: a new hire gets an answer with a source immediately, not after a week.
The figures are indicative for typical volumes. We give a precise estimate for your company after analysing the process.
Indicative scope
- Source inventory and architecture design
- €1,200 – €2,400
- Pilot over one data source (Teams or Slack)
- €4,000 – €8,000
- Production rollout with permissions and audit
- from €9,000
- Operations, reindexing and evaluations
- from €450 / month
Prices are indicative and exclude VAT. We give an exact figure only after analysing your process — not off the cuff on the first call.
Frequently asked questions
Why not just Microsoft Copilot or ChatGPT Enterprise?
For many companies that is enough and we will tell you so. A custom solution makes sense when you need sources outside Microsoft 365, hard citations, your own rules for “I don’t know”, an audit trail for compliance, or no data leaving the company.
What if our documents are outdated or contradict each other?
It shows up in the first phase and is the most common finding. The system flags conflicts and prefers newer sources, but good retrieval cannot fix bad content. A clean-up list is part of the deliverable.
How do I know the system is not making things up?
Every answer has a citation you can verify in one click. On the test set we measure how many answers are factually correct and how often the system correctly says “I don’t know”. You see the numbers before launch.
Can it run without the cloud?
Yes, we can deploy a smaller model on your infrastructure. Answers are somewhat weaker than with the largest models — we measure the difference on your questions so you decide on numbers.
What is RAG and do I need to understand it?
RAG (retrieval-augmented generation) is the technical name for exactly this solution: the model first retrieves the relevant parts of your documents and only then answers from them. You do not need to understand it — what matters is that answers have a source.
Related service
Answers from your documents
We build this solution as part of our service „Answers from your documents — with a cited source“.
Search and chat over your own documentsRead more
- Company document search: why nobody can find anything and how to fix itCompany document search fails in SharePoint, Google Drive and Teams for the same reasons. When to tidy up, when Copilot will do, when to build an AI assistant.
- RAG without illusions: why your company chatbot lies and how to fix itUploading a PDF to a chatbot takes five minutes. Building a knowledge base you can trust takes considerably longer — and the difference is three things a demo never shows.
- AI agents vs. chatbots: what your company actually needs in 2026A chatbot answers. An agent acts. The difference is whether AI can see your systems and do something in them — and that is exactly where the return is decided.
More solutions
WholesaleAutomating email orders into your ERP with AI
~70% · of orders handled without a human
AccountingAutomated invoice and delivery note processing with AI
< 3 months · typical payback period
E-commerceAI customer support automation — without the risk of AI sending nonsense
~40% · shorter first-response timeIf your people spend time looking for information that already exists in the company, this is one of the fastest-returning AI projects there is.
You do not need to know whether you need AI, automation or a new system. Show us the process that slows you down — we will tell you what can be automated and whether it pays off.