In every SA business of 20+ people, the same five questions get asked dozens of times a week. "What's the leave policy?" "Where's the latest pricing sheet?" "Who do I contact at supplier X?" "What's the SOP for refunds?" Senior staff become a knowledge bottleneck — they answer the same questions over and over while their actual work piles up. New hires take three months longer to ramp than they should because nobody has time to explain things twice.
An AI team assistant fixes this without forcing you to write a wiki. We connect a language model to the documents and systems you already have — SharePoint, Google Drive, Confluence, Slack, your existing wiki, your CRM — and constrain it to answer only from those sources. The result is an AI that knows your business, cites its sources, and says "I don't know" when it doesn't.
What we build
Document grounding (RAG)
Answers come from your actual SharePoint, Google Drive, Confluence, Notion, internal wikis. Every answer cites the source.
Slack & Teams integration
Native bot in Slack and/or Microsoft Teams. Ask in a channel or DM, get answers without context-switching.
Permission-aware
Mirrors your existing access rights. Staff who can't see HR docs can't get HR info from the AI either.
System lookups
Beyond docs — looks up CRM contacts, ticket status, project info, pricing. Reads structured systems via API.
Escalation paths
When the AI doesn't have an answer, it routes the question to the right human and learns from the response.
Onboarding accelerator
New hires get a knowledgeable assistant from day 1. Reduces time-to-productivity meaningfully.
Why "just use ChatGPT" doesn't work
Three reasons generic AI fails as an internal assistant:
It doesn't know your business. Generic AI was trained on the internet, not your policies. Ask about your leave entitlement and you'll get a generic answer about typical company policies — confidently wrong.
It hallucinates. When generic AI doesn't know something, it makes something up that sounds plausible. For internal questions where there's a real right answer (in your handbook, SOP, contract), that's worse than useless — it's actively dangerous.
It can't access live data. "What's the status of our Q3 project?" "How many tickets is support handling today?" "What does our pricing currently look like?" Generic AI has no answer; a properly-built team assistant does.
Who it's for
- 20+ employees. Below this, the manual answer-the-question approach still works.
- Decentralised knowledge. Information lives across SharePoint, Slack, wiki, email, plus institutional memory in senior staff's heads.
- Repetitive internal questions. If staff are asking the same questions repeatedly, AI pays back fast.
- Document discipline (or willingness to develop it). The AI is only as good as its sources — if your policies aren't written down anywhere, fix that first.
Sectors where this typically applies: professional services firms, law firms (case knowledge bases), accounting practices (tax guidance, client procedures), manufacturing (SOPs, product specs), healthcare practices (HPCSA-aligned policies, billing rules), non-profits (programmes, donor info), education (teacher resources, policies), retail chains (operating procedures), and any company past 50 employees where institutional knowledge is starting to slip.
Typical SA deployment
Representative implementation: SA accounting practice with 35 staff. Figures illustrative.
Baseline: partners interrupted ~12 times per day each by junior staff asking procedural and tax-treatment questions. Most questions had documented answers in the firm's internal wiki and SARS guidance library, but staff couldn't find them. Onboarding time for new juniors averaged 4-5 months to full productivity.
What we built: Slack-deployed AI assistant trained on the firm's wiki, SARS guidance docs, internal billing procedures, client-folder structure conventions, and historical email threads (anonymised). Permission-aware so juniors couldn't query partner-only material.
Representative results after eight weeks:
- Partner interruptions: down ~65% (mostly the easy stuff)
- New-hire onboarding time-to-productivity: 4-5 months → ~3 months
- Average answer accuracy on documented topics: 94%
- "I don't know" rate (rather than guessing): ~7%, all properly escalated
- Source-document gaps surfaced for cleanup: 47 in first month (huge value beyond the AI itself)
What it costs
Implementation. R24,000 to R90,000 depending on data sources, integration depth, and access-control complexity. Single-source Slack bot pulling from one document repository starts at R24,000-R35,000. Multi-source, permission-aware, with API integrations to CRM/ERP/ticketing lands R55,000-R90,000.
Monthly running. R2,000-R8,000 typically. AI usage is mostly the language-model API; cost scales with query volume. We don't charge a SaaS layer.
Hosting options. Standard deployments use vetted ZA-region cloud. For highly sensitive corpora (legal, medical, financial advisory) we offer fully-private deployment with no external API calls.
POPIA & access control
- Permission inheritance. Connects to existing SSO (Azure AD, Google Workspace, Okta). Access rights flow through.
- Source citation. Every answer links to the source — no black-box outputs.
- Audit log. All queries logged with user, timestamp, sources retrieved. Subject to retention policy.
- Data residency. ZA-region by default; private deployment available for sensitive workloads.
How to start
30-min call to map your knowledge sources and biggest internal-question patterns. Within a few days a written proposal with phased rollout (typically: top-5 question categories first, broader rollout once accuracy is proven).
Email info@faautosolutions.com or use the contact form.