What does a knowledge assistant cost in Australia?

FC-26 Knowledge Assistant Sprint runs $18,000–$35,000 AUD depending on corpus size, integration depth, and pilot scope — fixed scope with 50% on signature. Assessments from FC-07 ($1,990) de-risk the approach first.

Do we need Azure for RAG?

Azure AI Foundry is a common choice when you are already on M365. Other stacks are valid on GCP or Supabase for app-embedded assistants. Architecture follows identity and data location — not hype.

How accurate are RAG answers?

Accuracy depends on document quality, chunking, and retrieval tuning. Pilot with real questions from staff, measure citation correctness, and expand corpora gradually — not big-bang on day one.

← All articles

AI on your data · 8 July 2026 · 9 min read

Vector RAG knowledge assistants explained for business owners

Vector RAG turns your documents into answers staff can trust — with citations and permissions. Here is how it works, what it costs, and when it beats Copilot alone.

RAG — retrieval-augmented generation — is the pattern behind most "chat with your documents" products that actually work in business. Your files are chunked, embedded into a vector database, and retrieved at question time so the model answers from your content — not from generic training data.

For Brisbane and SEQ service businesses, the use case is concrete: dispatch asks about site access rules, finance asks about billing exceptions, new hires ask about SOPs — without paging the one person who remembers.

How vector RAG works (plain language)

Documents are split into sections and converted to numerical embeddings. A question is embedded the same way; the system finds the closest chunks and passes them to the language model as context.

Good implementations return citations — file name, page, section — so users can verify. Permissions filter chunks before retrieval so users only see what they could open in SharePoint or your app.

When RAG beats Copilot

Copilot excels inside M365 apps for individual drafting. RAG excels when answers must come from a defined corpus — job packs, safety procedures, client history — with audit-friendly citations.

If the problem is "find the right paragraph in 5,000 PDFs," RAG. If the problem is "rewrite this email," Copilot.

Build vs buy vs sprint

Off-the-shelf tools rarely map to your permission model. Custom builds take months when scoped poorly. FC-26 Knowledge Assistant Sprint delivers a pilot-ready assistant in 4–6 weeks: corpora scoped, Azure AI Foundry or equivalent, handoff documentation.

Start with FC-07 if you are unsure whether RAG, Copilot, or automation wins first.

Security and governance

Data residency, retention, and logging must be designed — not assumed. No training on client data without explicit policy. Monitor prompts and block exfiltration patterns.

North Ark builds permissions-aware assistants after tenant readiness — see connect documents to AI safely and our AI knowledge assistant page.

Frequently asked questions

What does a knowledge assistant cost in Australia?: FC-26 Knowledge Assistant Sprint runs $18,000–$35,000 AUD depending on corpus size, integration depth, and pilot scope — fixed scope with 50% on signature. Assessments from FC-07 ($1,990) de-risk the approach first.
Do we need Azure for RAG?: Azure AI Foundry is a common choice when you are already on M365. Other stacks are valid on GCP or Supabase for app-embedded assistants. Architecture follows identity and data location — not hype.
How accurate are RAG answers?: Accuracy depends on document quality, chunking, and retrieval tuning. Pilot with real questions from staff, measure citation correctness, and expand corpora gradually — not big-bang on day one.

You want employees to query company data Consult service

Next step

vector RAG knowledge assistant — ready to act?

Knowledge assistant sprint

Or book a free fit call →