Practising AI in Emerging markets
Building AI solutions requires a number of basic steps, and foundational infrastructure to deploy effectively. It is helpful to get a guide on how to acquire these foundational pieces in your country, such as power, data, internet, powerful computers.
Steps to Deploy AI

The Path to Successfully Deploying an Ai Solution in Africa
Define Problem
Need to properly select problems that can be tackled with data/ AI using either ML or DL.
After you've picked the problem, you select between machine learning and deep learning. This post does a nice job of explaining the difference between ML & DL, while below we summarize when you might choose to use either.

AI vs ML vs DL in Africa
Find Data
Need to find data that is available, cheap and in proper format. You'll have to capture, buy or build millions to hundreds of millions of examples of the behavior you desire.
After you've picked the problem, you select between machine learning and deep learning. This post does a nice job of explaining the difference between ML & DL, while below we summarize when you might choose to use either.
Open Data Initiatives
Search for data on large independent repositories of open data on the African continent
Data from Foundations
A number of development-focused organizations gather data and make them available for public use
Self-generated
Many startups are getting creative about collecting data they need themselves. Get Inspired by some examples below
Acquire Knowledge
Once again, your local AI communities are your best bet for acquiring knowledge on how to leverage AI. Below, we will list some useful online resources that members of our alliance have recommended.
Access Compute
Need the right hardware and software that greatly impact time to solution and quality of product to be any competitive. Broadly speaking, you will either buy these to setup in-house, or rent from the Public cloud.

Directly from your website, access powerful-enough GPUs and compute for free. Perfect for students learning Ai who don't want to spend much money

You are sure of how much load you have, have expertise to set it up and ability to provide power & security - save costs by building your own cluster

This is the modern way. Easy to setup and grow or increase as you want, don't need much special skills. Gets expertise after a long time though
AI involves working with very large amounts of data. It is very important to stay up-to-date with the best choices for your scenario so you are not wasting resources, and are staying competitive
Purchase & Running Cost
Choosing the wrong option of hardware or software can have you spending 5 times what you are supposed to spend. This can run you out of business very quickly.
Time to solution
This is felt both by the developers working and the business people paying. Imagine spending two months training an AI model, when you could have spent 2 days on similar task if you were better informed.
Quality of solution
Some software and hardware solutions push the edge of what is possible with your AI use-case. Being uninformed can allow your competition build solutions that are miles ahead of yours.
Today there are many hardware approaches to tackling AI challenges, and accompanying software that is compatible with your choice. Below are the top options so far and links to popular vendor websites.
CPU – Intel Product Page, AMD Product Page
GPU – Nvidia Product Page
FPGA – Intel Product Page
TPU – Google Product Page
Broadly speaking you can either access these compute options on the cloud or with your own compute cluster you buy for your space. You can also physically visit centers that rent out compute per hour – these are beginning to pop up in several locations around Africa.
Cloud computing is the new way of computing, and great for these reasons
- You access tools only when you need them. This way you save a lot of money especially if you need just for short time, or want to run tests
- You don’t want to spend your time and expertise setting things up from scratch
- You want easier access to tools and consulting services that make it a lot easier for you to get started
Global cloud options accessible in Africa are AWS, Azure, GCP, Alibaba
On-prem computing is the older way of computing, but still has its merits
- You know exactly how much computing you want to do so can predict what you need
- You will utilize what you buy more than 70%
- You have some restrictions around where your data can live
- You have some very strong reason for wanting to setup all the infrastructure by yourself and provide the physical security and power to keep it going.
In Africa you can purchase compute this way by working with small and large distributors like Lenovo, Dell, IBM and consulting shops like LigaData, Absolut Data, TechnoTree, Tavia Technologies, BlueChip Technologies
Google Codelabs are a great alternative for students learning about AI, looking for a cheap access to very fast compute they can use through the night.
Happy learning
Build Model
There are hundreds of ML & DL model types and architectures. You can build from scratch for absolute control, or with tools that dramatically ease the process

Deploy Solution
Complete solutions need a willing partner to purchase or distribute it to reach millions. Many times connection to the right large partners is elusive.
Overtime at A4AI, we will work to create mechanisms for top AI project innovators to work with larger organizations to deploy, so their solutions realize the visions they are built for.
