Build or Buy? Implementing AI Solutions for East African Organisations

The approach to implementation of artificial intelligence (AI) solutions can vary significantly with specific use-cases. One of the critical decisions for organisations to consider is whether to develop bespoke AI solutions or integrate off-the-shelf products into their IT estate. This decision is key for large-scale digital transformation projects requiring substantial investment, as the limited resources within the East African region leave little room for failure. 

The decision between developing custom AI solutions and utilising pre-existing ones depends on several factors, including infrastructure constraints, access to expertise, costs, and concerns around security and data sovereignty. A key consideration is the level of customisation required for the specific use case. In the healthcare sector, for example, tailoring open-source AI models to address East Africa's unique needs—such as regional linguistics, mental and emotional health, and genetic factors would be required and would necessitate the integration of expert local knowledge to ensure the models are both impactful and relevant. Although the customisation of open-source models would be suggested for this industry, this level of fine-tuning requires significant operational capacity, particularly in terms of computational resources. 

Collaboration within the region can address some of these challenges. Governments, local AI communities, and universities within innovation hubs can work together on key research projects in industries involving sensitive data that requires fine tuning. This approach not only helps comply with regional data protection regulations but also mitigates the risk of biased models negatively affecting the societies they serve. While off-the-shelf models, such as Microsoft Copilot Studio, are faster to deploy and demand less expertise, they may conflict with regional data sovereignty laws like Kenya’s Data Protection Act. 

If a use case involves fine-tuning an open-source model or training it with sensitive data, such as in healthcare, significant computing resources and software development (DevOps) expertise are required. Organisations in East Africa face the challenge of deciding whether to invest in on-premise infrastructure, such as AI clusters, or opt for cloud-based solutions like Amazon Web Services (AWS), Azure, or Google Cloud. On-premise infrastructure provides full data control, which is crucial for ensuring compliance with data sovereignty regulations and protecting sensitive information. However, the costs of equipment – such as mid-range GPUs like NVIDIA A100 or H100 alternatives (e.g. NVIDIA RTX 6000 Ada) – along with power requirements, can be prohibitively high. This is especially true in regions where electricity reliability is a concern. To address this, organisations should consider solar-powered on-premise data centres, which can offer a more sustainable and reliable energy source for maintaining operations. 

Cloud-based solutions, on the other hand, provide flexible scalability, making them suitable for use cases that do not involve extremely sensitive data, such as agriculture. These solutions can help manage compute costs effectively through load balancing and auto-scaling, particularly when workloads are intermittent. However, cloud adoption may raise compliance challenges, such as the risk of data breaches or violating regional data protection laws like Kenya's Data Protection Act or Rwanda's Data Protection and Privacy Law. Latency issues can also arise due to the geographical distance from major cloud data centres, which may impact real-time applications. Despite these challenges, cloud solutions remain a feasible option for many applications, provided organisations are cautious about vendor lock-in risks and carefully evaluate compliance and latency trade-offs. 

Vendor lock-in occurs when organisations become overly reliant on a single cloud provider’s services, making it costly and technically challenging to switch providers. Migrating to another provider often requires re-engineering applications, retraining AI models, and managing data transfer fees, which can strain budgets and resources. Additionally, many international providers store data outside the continent, potentially violating local regulations. For organisations seeking greater compliance or more accessible localised services, these challenges become even more pronounced. The technical complexity, combined with costs and the need to maintain operational continuity, makes vendor lock-in a significant risk for East African organisations. 

Adopting hybrid cloud strategies with tools like Kubernetes can help mitigate these risks. Kubernetes is an open-source platform that orchestrates containers – self-contained software packages that can run consistently across different environments – allowing organisations to remain cloud-agnostic. Kubernetes can support East African organisations in running their AI workloads across multiple providers or combine cloud services with on-premise infrastructure. This approach can offer greater flexibility, enabling sensitive data to remain on local servers while computationally intensive AI tasks are executed on cloud-based platforms. Additionally, Kubernetes supports resource scaling and cost optimisation, crucial for budget-conscious organisations. By using hybrid clouds, businesses can reduce dependency on a single vendor, maintain compliance with local regulations, and optimise operating costs. 

Deploying AI solutions involves significant technical complexity, requiring a skilled team to ensure the smooth operation and integration of these systems. In the context of East African organisations, the decision between building bespoke AI solutions in-house or using off-the-shelf tools requires careful consideration of the required expertise. Building in-house demands a multidisciplinary team, including data scientists to evaluate models and datasets, machine learning (ML) engineers to scale and optimise AI models, and software architects to ensure seamless integration of models into production environments. Additionally, DevOps engineers play a crucial role in automating deployments, managing infrastructure, and ensuring system reliability. However, with only 5% of Africa’s talent reportedly having access to computational power for research and innovation, the scarcity of local expertise presents a significant challenge for the region. 

For organisations aiming to develop bespoke AI solutions, substantial capital may be needed to train local talent or hire experts from outside the region, an investment that may only be justified for use cases requiring highly sensitive data, such as healthcare. These scenarios demand full control over data and systems to comply with strict privacy regulations and safeguard patient information. Conversely, organisations with less critical requirements may benefit from adopting off-the-shelf tools like Microsoft Co-pilot Studio. These pre-trained solutions are easier to deploy and require minimal training, thereby reducing the reliance on in-house technical expertise. By utilising such tools, organisations can minimise costs and implementation time while integrating these solutions into their existing workflows. 

Time to market is a critical factor when launching new products, as the speed at which an organisation can develop and deploy a solution often determines its competitiveness and impact. Building bespoke AI solutions involves substantial time and effort, particularly in data collection and labelling, which are foundational to creating effective models. In East Africa, collecting high-quality datasets is often challenging due to fragmented or incomplete data sources, limited digital infrastructure, and the need to account for diverse languages, dialects, and cultural contexts. Additionally, labelling data to train AI models is a labour-intensive process requiring specialised domain knowledge on top of skills related to data acquisition, cleaning and preprocessing. Complex use cases like healthcare diagnostics or agricultural optimisation may require experts in crop diseases or rare human conditions to conduct quality assurance of the datasets. 

Once data is prepared, model training and testing further extend development timelines. Training models on large datasets requires significant computational resources and fine-tuning them to achieve acceptable levels of accuracy often involves iterative cycles of refinement. Testing these models to ensure they perform reliably under real-world conditions adds another layer of complexity and time. For organisations with urgent AI use cases that pose minimal risks regarding compliance and sensitivity, procuring off-the-shelf solutions offers a more pragmatic approach. These pre-built tools can be implemented quickly, enabling organisations to address immediate needs while gradually developing the expertise and infrastructure required for building bespoke AI solutions in the future. 

In summary, the decision to build versus buy AI solutions for organisations within East Africa largely depends on the specific use case. However, for most scenarios, off-the-shelf products with some degree of customisation are the more practical choice. This is primarily due to the significant time, expertise, and capital required for bespoke solutions. Bespoke AI systems are best suited for use cases involving highly sensitive data, such as healthcare, where strict data control is essential, or for industries like agriculture that demand region-specific adaptations, such as accounting for unique climate conditions. Similarly, applications that require a deep understanding of local linguistic and cultural nuances may also require custom-built solutions to ensure accuracy and inclusivity. However, for most organisations, the combination of cost-efficiency and quick deployment makes off-the-shelf solutions, supported by cloud services, the preferable option. 

For infrastructure, cloud platforms like AWS are generally more viable than on-premise setups, which demand substantial investment and ongoing expertise for maintenance. On-premise solutions, while offering greater control, often exceed the budgets and technical capacity of many organisations in the region. Looking ahead, the development of more regional data centres across East Africa could significantly reduce reliance on international cloud providers, addressing key issues such as latency and compliance with local data sovereignty laws. Such advancements would not only enhance the feasibility of deploying AI solutions but also support the region’s broader digital transformation by fostering a more accessible and resilient technology ecosystem.