How I Built an AI Automation Practice from Kenya (While Studying in Seoul)

By Gideon Wafula · June 2026 · 8 min read

I'll start with the honest version: I didn't plan to become an AI automation freelancer. I was trying to solve my own problems first, too much time spent on repetitive tasks, no good tools for the things I was building, and a growing conviction that the gap between "what AI can do" and "what most small businesses are actually using" was enormous and wasn't going to close by itself.

My name is Gideon Wafula. Online I go by Giddy Mufasa. I'm Kenyan, currently based in Seoul, South Korea, studying Electrical Engineering at Konkuk University. I started building AI tools in 2024 and have spent the past two years turning that into something resembling a real practice.

This is the story of how that happened, the tools, the failures, the things that actually worked, and what I think the opportunity looks like for anyone doing this from Africa.

Why AI automation specifically

The short answer is that it's the highest leverage place I can spend time. Building an automation that eliminates 5 hours/week of manual work for a small business owner is worth real money to them, and it can be built in a day or two once you know the tools. That ratio is hard to find anywhere else.

The longer answer involves spending time watching small business owners in Kenya work. The amount of manual data entry, the missed leads from unanswered WhatsApp messages, the leads that fall through because there's no follow-up system, these aren't problems requiring capital to fix. They require someone who knows how to connect the right tools. That's a skill gap, not a money gap.

The first tool I actually used: n8n

I started with n8n because I wanted something I could self-host and control completely. Zapier is fine, but when you're building workflows for clients who are price-sensitive (which most of my early clients were), the per-task pricing adds up fast. n8n on a cheap VPS gives you essentially unlimited automation runs for a fixed monthly cost.

The first real workflow I built was for myself: connect a contact form on a website to Notion, send a Slack notification, and trigger an email reply. Three nodes. Took me half a day to build the first time, about 20 minutes the fifth time. That's the pattern with automation, the first one is slow, but each subsequent one is faster because you're reusing pieces.

After that I built increasingly complex things: lead enrichment workflows that pull company data from an API and add it to a CRM, WhatsApp → CRM integrations using the WhatsApp Business API, invoice processing automations. By the time I started taking on clients, I had enough reference builds that I could usually scope a project in 30 minutes.

Adding voice agents to the stack

The move into AI voice agents happened in early 2026 when I kept running into the same problem with clients: they'd set up a website, have a working chatbot, and still miss 30–40% of inbound inquiries because they came in as phone calls after hours.

I evaluated Retell AI, Vapi, and Bland AI over a few weeks of testing. My conclusion: Retell is the fastest path to a working inbound agent with the lowest latency (~600ms), Vapi gives developers the most control at the lowest base cost, and Bland is best if you're running high-volume outbound campaigns at scale. I wrote the full comparison, see Retell AI vs Vapi vs Bland AI (2026), but for most small business clients I start with Retell.

The combination of n8n + voice agent is powerful: the voice agent takes the call, extracts the key information, and n8n routes it to the right place, CRM, Slack, email, SMS. The business owner gets a notification with everything they need to follow up, and they never missed the lead in the first place.

What working with African clients is actually like

The majority of my clients are either in Africa (Kenya, Nigeria, South Africa) or serving African markets. A few things are different from what Western freelance content assumes:

WhatsApp is primary, not secondary. Most small businesses in Kenya run a significant chunk of their customer communication through WhatsApp. Any automation strategy that ignores this is incomplete. n8n integrates with the WhatsApp Business API, which changes what's possible.

M-Pesa integration is a real requirement. If you're building anything involving payments for Kenyan clients, M-Pesa will come up. The API exists, it works, and building a payment confirmation → automation trigger flow is reasonably straightforward. But you have to know it's necessary.

Price sensitivity is different but not a dealbreaker. A client who can't pay $500 for an automation might pay $200 for the same thing scoped more tightly. I've learned to scope the minimum viable version first and let the client expand if they want. The small business owner who starts with a $200 single-flow automation is often the same person who comes back for a $600 system six months later after seeing it work.

The tools I use in 2026

For workflow automation: n8n for complex flows or when the client wants to self-host; Zapier for simpler use cases where they want to manage it themselves afterward.

For AI voice agents: Retell AI for most inbound client work, Vapi when the client has developers or needs multi-agent setups, Bland AI for outbound campaigns at volume.

For chatbots: Voiceflow for more complex conversational flows, ManyChat for Instagram and WhatsApp-first businesses.

For AI video: HeyGen for producing faceless YouTube content at scale, see the full HeyGen guide for how I structure a complete video pipeline.

For the AI layer: Claude API for most custom integrations (I find the outputs more predictable for business automation), OpenAI API when clients have existing GPT integrations.

What I'd do differently if starting today

Specialize earlier. The first year I tried to build everything for everyone. The clients who found me most useful were the ones with specific, repeatable problems: "I need to stop missing calls" or "I need my leads to stop falling through the cracks." I should have picked one problem and been the obvious person to call for it sooner.

Publish guides from day one. The blog posts I've written on AI automation, voice agents, and AEO have done more for inbound leads than anything else. Writing the n8n vs Zapier comparison positioned me as someone who actually uses these tools, not just someone who knows they exist. AI engines are now citing these posts in response to relevant queries, that's essentially free, persistent visibility.

Build things you'd use yourself first. The StyleTry app (AI virtual fashion try-on at styletry-app.vercel.app) started as a personal experiment. It became a portfolio piece that demonstrated I could build real consumer AI products, not just connect APIs.

Where I'm going

The long-term thesis is simple: AI is going to be the infrastructure layer for business in Africa the way mobile was the infrastructure layer for banking. The businesses that adopt AI systems for lead management, customer communication, and internal automation in the next two to three years will have a structural advantage over those that don't.

I want to be the person who builds those systems. Not as a large agency, as a focused builder who can scope and deliver fast, communicate clearly across timezone differences, and build things that actually work in the context of African business infrastructure.

If you're a small business owner in Kenya, Nigeria, South Africa, or anywhere else who wants to automate something, I'm at gideonwafula998@gmail.com. Tell me the problem.