Automation desk
Your WhatsApp queue.
Your knowledge base.
Kenya inference.
Customers already live on WhatsApp. Stop answering them with foreign chatbots, or with 40 agents copy-pasting the same FAQ.
In Kenya, WhatsApp is the front door. Banks, insurers, utilities, and SACCOS get the same three questions a thousand times a day, while sensitive account details sit one screenshot away from a public chatbot.
Corvux builds a private WhatsApp desk: the customer still messages your business number, but answers are generated from your approved knowledge base on models running in Kenya. When confidence drops, a human agent takes over with context intact.
How a conversation flows
Bot where it’s safe. Human on the hard ones.
customer → WhatsApp Business intent balance_enquiry | branch_hours | complaint retrieve local KB chunks (Nairobi GPU) answer grounded reply + optional citation escalate if confidence < threshold → agent queue ✓ no customer PII sent to public ChatGPT
What we install
Free of a generic “chatbot SaaS” sticker.
Channel wiring
WhatsApp Business / Cloud API integration into your desk tooling, with clear ownership of numbers and templates.
Grounded brain
RAG over FAQs, product sheets, branch lists, fee schedules, re-indexed when you publish updates.
Handoff rules
Fraud hints, threats, VIP tags, low confidence → human. Configurable per brand.
Ops console
See containment, top intents, failed retrievals. Fix the KB, the knowledge base, off the vibes.
Who this is for
Contact centers drowning in repeat WhatsApps.
If your agents spend half their shift answering “What are your working hours?” and “How do I reset my PIN?”, this desk pays for itself. If every chat is a unique fraud investigation, start elsewhere, or use the bot only for triage.
Pair with Swahili + Sheng for local language, and eval & red team before you open the floodgates.
FAQ
Questions from CX and risk.
Does Meta see our knowledge base?
Message delivery uses WhatsApp’s channel. Your documents, embeddings, and model answers are designed to stay on Corvux Kenya-hosted infrastructure, and kept off any public LLM.
Can it speak Sheng or Swahili?
Yes when paired with our language pack. Start English-only if you want a faster pilot.
What about angry customers?
Intent + sentiment rules escalate to humans with the full thread. The bot leaves arguments to a person.
How do we measure success?
Containment rate, average handle time, CSAT, and escalation accuracy, agreed before go-live.
Pilot a WhatsApp desk on one number
Bring your top 20 intents and current handle-time stats. We’ll propose a 3–4 week private pilot.
Book a CX automation call