Service · New
AI Implementation & Development
Practical AI for South African businesses. The boring layer that makes the exciting stuff safe.
Microsoft 365 Copilot · Power Automate · Custom GPTs & Claude Projects · RAG knowledge bases · POPIA-aligned AI governance · Cost & risk monitoring.
Book an AI assessment →What we mean by AI implementation
Boring infrastructure for exciting tech.
Most South African mid-market businesses we talk to don’t need a custom LLM. They need Microsoft 365 Copilot deployed properly — with adoption tracking, governance, and the POPIA dossier their auditor will eventually ask for. Plus a few targeted Power Automate or Copilot Studio workflows that take real, measurable work off the team’s plate.
That’s a 60-90 day engagement, not a 12-month transformation. It costs what an extra junior staff member would cost. And done right, the team gets back 3-6 hours per week, every week.
This is the bit most consultancies skip: nobody wants to write the acceptable-use policy. Nobody wants to map the data flows that POPIA cares about. Nobody wants to track which prompts staff have run through which models or sit through the monthly hallucination spot-check. So we do. And then the exciting AI work your team is actually doing has a foundation that doesn’t fall over the first time the Information Regulator writes.
What we cover
Six things, done with care.
Each one is a real engagement with measurable deliverables — not a slide deck.
Microsoft 365 Copilot — deployed properly
Tenant readiness check (SharePoint permissions are usually a mess and Copilot inherits that). License sizing — most clients should NOT roll out to everyone. Adoption tracking. Power user training. Quarterly utilisation review. Without these, the R430/user/month spend becomes shelfware in three months.
Targeted workflow automation
Power Automate, Copilot Studio, AI Builder. The boring wins: invoice data extraction, meeting summaries posted to channels, IT helpdesk triage, candidate CV screening, supplier onboarding checklists. Each automation has a measurable hour-saved number we agree upfront and report on quarterly.
Custom GPTs & Claude Projects
Department-scoped AI assistants. A Finance assistant trained on your Chart of Accounts. An HR assistant on your policy library. A Sales assistant on your past proposals. Quick to build, immediate value, easy to retire when better tooling arrives. We avoid heavy custom LLM training — rarely the right answer for mid-market.
RAG knowledge bases
Retrieval-augmented generation done sensibly: your SharePoint or document store wired to a chat interface that cites its sources. Use cases: sales playbooks queried by reps, support runbooks queried by helpdesk, internal “how do I” questions answered without escalation. Document permissions are respected end-to-end — no leakage of sensitive content into responses.
AI governance & POPIA dossier
The boring layer: acceptable-use policy (written, signed, in induction). Data flow map (what user inputs go where, what training data was used, retention period). Vendor due-diligence pack (OpenAI, Anthropic, Microsoft, Google — their SOC 2s, sub-processors, data residency). Audit trail of prompts and outputs on sensitive workflows. The thing your auditor will ask for in 2027.
Cost & risk monitoring
Per-user / per-workflow cost reporting. Sensitive data exposure scanning (DLP across Copilot interactions). Hallucination spot-checks on high-stakes use cases (regulatory advice, financial commentary, clinical content). Model-update impact assessments (when GPT-5 ships, what changes for the workflows that depend on it).
What runs while you sleep
Quiet, governed, auditable.
AI without governance is shadow IT. We’ve set the recurring checks up so the boring assurance work happens without anyone asking.
- ✓ Copilot license utilisation reviewed weekly (resize before quarterly true-up)
- ✓ AI tool access per user audited (who can hit which models)
- ✓ Sensitive data exposure scans (DLP across Copilot prompts)
- ✓ Cost monitoring per workflow — flagged if spend triples month-on-month
- ✓ Hallucination spot-checks on high-stakes outputs
- ✓ POPIA AI audit log review (records subject access requests can resolve)
- ✓ Model-update notifications + impact assessment (GPT-5, Claude 4.X, Gemini 2.X)
- ✓ Vendor lock-in risk review — document portability across providers
What you can hand to the board or the auditor
Evidence, not vibes — AI edition.
Quarterly AI report
One A4 page for the board. License usage by user, top 5 workflows by hour-saved, total spend, sensitive-data incidents, governance status. No jargon, no hand-waving.
POPIA-aligned AI deployment dossier
Acceptable-use policy, data flow maps, vendor sub-processor lists, audit logs, breach response addendum. Updated every quarter. Ready to hand to the Information Regulator if they ask. Most consultancies skip this entirely — we lead with it.
Honest about the service age
This is a young service line at Burika. Built on 20 years of M365 expertise but the AI-specific playbook has hardened over the last 12 months. We take selective clients while the patterns mature — means more attention per engagement, fewer of them concurrently.
What we won’t do
We won’t sell you a custom LLM you don’t need. We won’t promise AGI or “transformation.” We won’t hide what each model vendor does with your data. If broken processes are the real problem, we’ll tell you to fix those before automating them — even if it costs us the engagement.
“We were drowning in user tickets and 3am server alerts. Two years on Burika’s managed support, our team hasn’t seen an after-hours incident — and our staff actually like calling IT now.”
— IT Manager, EurekaIT, Johannesburg
“Burika audited our setup, fixed three POPIA gaps we didn’t know we had, and now monitors everything quietly. Our MD finally stopped asking me “are we covered?””
— Operations Director, Joubea