AI replies in Sheng, Swahili, and English on WhatsApp

A buyer opens with "Sasa, hiyo cap iko?", switches to English for "how much total with delivery to Kilimani?", and throws in "si poa" when confirming. One message, three registers. Replies that sound like a US support ticket — "Dear valued customer, thank you for your inquiry" — kill trust in Kenyan WhatsApp. Timi matches how the buyer writes while staying anchored to your prices and delivery rules.

For the full AI overview see AI on WhatsApp in Kenya. For when to automate vs hand off see when to use AI vs reply yourself.

Match the buyer's language

Timi reads how the customer writes and responds in kind:

• English-only buyer → clear English with KES prices and delivery zones
• Swahili greeting → Swahili reply with same facts, not awkward literal translation
• Mixed "Corner sofa ni how much na delivery?" → mixed reply that sounds local, not forced into one language

You paste catalog facts once in shop context; Timi adapts tone per message, not per settings toggle.

Side-by-side: weak vs strong replies

• Bei gani ya size 42? — Weak (generic AI): "Please specify product for pricing."; Strong (Timi + your context): "Size 42 sneakers KES 4,200. CBD delivery KES 150. Which colour?"
• Do you deliver to Thika? — Weak (generic AI): "We serve selected areas."; Strong (Timi + your context): "Thika delivery KES 350, 1–2 days. Reply with estate for exact fee."
• Na ile red iko? — Weak (generic AI): "Item availability unknown."; Strong (Timi + your context): "Red restocked yesterday — same price KES 3,500. Want to reserve?"
• Corner sofa ni how much na delivery Ruaka? — Weak (generic AI): Formal English paragraph, wrong tone; Strong (Timi + your context): "Corner sofa from KES 78k depending on fabric. Ruaka delivery from KES 3,500. Unaweza share ukubwa wa room?"

Strong replies share three traits: number, next step, local tone.

Why generic AI fails in Kenya

Tools trained on US/UK support tickets:

• Misread Sheng as typos and "correct" them
• Quote wrong currency or shipping regions
• Over-formal "Dear valued customer" openings
• Pull web data about products you do not sell

Timi is scoped to your products and your delivery rules — not the open internet. See Timi vs Meta Business AI for built-in Meta AI comparison.

Set your voice in business context

Paste FAQs the way you talk to customers at the stall:

"Yeah tunadeliver Westlands and Kilimani same day if you order before 2pm. Till 123456, name SHOP NAME."

That becomes the base for auto-replies. Avoid:

• Long policy walls pasted as one block
• English-only context when 70% of buyers mix Swahili
• Prices missing — forces vague replies

Update context after every drop when SKUs or fees change — language fit does not fix stale numbers.

Code-switching in one thread

Common pattern:

1. Buyer greets in Swahili
2. Asks product detail in English
3. Confirms payment in Sheng shorthand

Timi follows the latest message tone while keeping facts consistent. You do not need three separate catalogs — one accurate context, adaptive replies.

When matching language beats forcing English

• Buyer writes full Swahili: Reply Swahili with KES amounts
• Buyer writes full English: Reply English
• Mixed urban chat: Match mix — do not force formal English
• B2B corporate buyer: Slightly formal English OK
• Complaint or refund: You take over — see handoff

Forcing English on Swahili-first buyers feels cold. Forcing Swahili on diaspora English-only buyers feels confusing. Follow the buyer.

Common mistakes

• Corporate FAQ paste in context — Timi sounds foreign
• No prices in context — AI fills gaps with vagueness
• English-only context for Nairobi fashion drops
• Ignoring handoff on payment questions — tone matters less than closing
• Expecting Timi to invent shop-specific Sheng not in your FAQs — paste slang you actually use

FAQ

Can I force English only? Timi follows the buyer. Most Kenyan sellers prefer matching language.

Does Sheng change every month? Timi handles common Kenyan chat patterns. Update FAQs if you use shop-specific slang.

Translate catalog to Swahili automatically? Paste key products in both languages in context if you serve both explicitly — do not rely on guesswork.

Wrong language on one reply? Fix context tone; pause chat and reply manually; refine FAQ examples.

Works for Luo/Kikuyu mix? Timi optimizes English/Swahili/Sheng urban mix — paste local phrases you use in context.

Compare to Meta Business AI? See /compare/meta-business-ai — catalog control and handoff differ.