In Part I, we introduced the African AI opportunity and the Moat Framework: six competitive edges that define how defensibility is created. But in practice, winning companies rarely rely on a single edge. They combine multiple edges into strategic positions that strengthen over time, where each layer reinforces the next.

Welcome to Part II. This is the African AI playbook.

Across the continent, we observe founders assembling repeatable combinations of edges that produce distinctive competitive signatures. Some own the entire value chain from data to workflow. Others fuse physical infrastructure with intelligent software. Some embed expert humans inside AI systems, while others target micro-niches too small for incumbent businesses or traditional business models to serve.

The Moat Framework

Taken together, the six edges define where durable advantage in African AI emerges.

1. The Distribution edge provides privileged access to customers others cannot reach.

2. The Workflow edge embeds products deep inside operations, making switching costly.

3. The Business Model edge restructures unit economics in ways incumbents cannot match.

4. The Data edge compounds proprietary datasets that improve with usage.

5. The Technology edge creates defensible intellectual property born from local constraints.

6. The Economies of Scale edge drives structural cost advantages as adoption grows.

AI Strategies for Africa

Individually, each edge can create meaningful defensibility. Combined, they form powerful moats that compound over time and define the strategic foundations of winning African AI companies. We call these moat combinations “strategies”. Each strategy activates a different combination of the six edges. We use the MoSCow prioritisation framework to map the relationship:

• Must have (M) means the edge is essential to the strategy’s defensibility; without it, the strategy fails.

• Should have (S) means the edge materially strengthens the position but is not a prerequisite.

• Could have (C) means the edge is a bonus but not structurally important.

The zigzag silhouette that you see below, gives you the competitive signature of each strategy at a glance. Full-Stack runs almost entirely along the Must-have line because it requires strength across nearly every dimension. Micro-Niche zigzags sharply because it depends heavily on Distribution and Business Model but not on the others. When evaluating a company, overlay its actual edge profile against the strategy it claims to be pursuing. The gaps between the claimed strategy and the actual edge activation shows the path to success.

1. Full-Stack: Own the entire chain from data collection to client operations.

Full-Stack companies control the customer relationship (Distribution), how data is collected and used to train the model (Data), how the model is built (Technology), and how it embeds in client operations (Workflow). Each layer feeds the next. This is the most defensible strategy because it activates almost all the edges, but it is also the hardest to execute and the most capital-intensive to build.

South Africa’s Orca Fraud approaches this pattern. The company provides AI-driven fraud orchestration for banks and fintechs across emerging markets. It controls distribution through direct integration with African payment platforms, embeds deeply into client fraud operations (Workflow), generates proprietary signals on fraud typologies specific to African payment rails that compounds with every transaction (Data), and builds custom ML models trained on patterns global providers have never seen (Technology). Each new client strengthens every layer simultaneously.

2. Phygital: Atoms generate bits, bits optimise atoms.

Phygital companies combine physical infrastructure (warehouses, vehicles, field agents) with intelligent software layers that make those physical assets dramatically more valuable. Each physical touchpoint generates proprietary data that feeds the digital layer, and the digital layer optimises the value of each physical touchpoint. The moat is difficult to replicate: software competitors can copy code, but not warehouses, fleets, or agent networks.

Ghana’s Gigmile operates a lease-to-own vehicle financing platform for gig workers across 13 cities in Nigeria and Ghana, with OEM partnerships (Yamaha, TVS, Bajaj, Hero) supplying the physical fleet. Over 10,000 vehicles have been deployed, each one simultaneously a productive asset and a data collection device: rider behaviour, route patterns, earnings, and repayment consistency build credit profiles for workers who have never had a bank account. This behavioural data feeds the platform’s risk scoring and automation engine, which determines who gets financed, at what terms, and when to intervene on delinquency.

3. Expert-in-the-Loop

Human judgement amplified by AI. As AI becomes ubiquitous across industries, the premium value of human expertise and personalised service increases dramatically, yet fewer people can afford it. Companies that leverage hybrid AI + human models, combining Africa’s cost-competitive skilled labour with advanced technology, win by offering white-glove human service deeply embedded in client workflows alongside intelligent automation. This can take the form of reinforcement learning with human feedback, forward-deployed engineers, or managed services. The efficiency gains from an individual consultant equipped with AI tools mean that services can be more easily scaled and productised, and humans are not necessarily as negative a signal as they may have been when automation was less sophisticated. Revenue per employee and gross margins remain the critical metrics to track.

Hakimu, a pan-African AI legal research engine, illustrates the model. The platform aggregates case law using multilingual embeddings and conversational search, but it does not replace lawyers or judges. It makes each one dramatically more productive. The AI surfaces relevant precedent across jurisdictions and languages that no individual could search manually. Hakimu leverages unique partnerships to inform an evaluation method that improves the models. Neither layer works without the other, and the deeper the platform embeds into workflows, the more knowledge it accumulates, making displacement increasingly painful.

4. Micro-Niche

AI dramatically lowers the cost of building software, making it viable to serve micro-niches historically too small for venture-backed models: smallholder agriculture, informal retail, francophone legal services. The strategy here is about inventing new customer segments: creating a viable market where none existed by restructuring how value is priced and delivered. This requires two edges working together. The Distribution edge gets the product to the customer. The Business Model edge makes the unit economics work. Neither alone is sufficient. But combined, they turn deep vertical focus into a beachhead for expansion.

Kenya’s M-KOPA is the archetype. The company provides solar energy systems and smartphones to households earning as little as $2 a day, financed through micro-payments of roughly $0.50 daily via M-PESA. No upfront cost, no credit history required. AI-powered scoring models built on mobile money usage, repayment patterns, and device behaviour determine creditworthiness for customers the formal financial system has never seen. Over a million households have been connected. The distribution runs through the mobile money rails millions already use. The business model (pay-as-you-go, priced against daily earnings rather than monthly income) is structurally inaccessible to traditional energy utilities or consumer electronics companies, who would need to dismantle their entire pricing architecture to compete.

5. Second Mover Advantage

Enter proven categories with adapted technology and pricing that incumbents cannot match. Most African AI companies active in the platform layer are second movers. The strategy becomes powerful when the second mover builds edges the first mover cannot replicate.

The global market for AI enablement tools (LLMOps platforms, evaluation frameworks, agent orchestration, fine-tuning services) is vast, but the African market alone may not offer the scale required. African innovators at the platform layer are better positioned addressing a global customer base from day one. By developing equivalent or superior products that are more cost-effective or tailored to underserved use cases, second movers learn from global pioneers, avoid early missteps, and build leaner, faster, more focused solutions with differentiated value.

Cape Town’s Cerebrium, founded in 2021, built a serverless AI infrastructure platform that lets engineers deploy multimodal applications without managing infrastructure, at up to forty percent lower cost than traditional cloud. The platform now powers Tavus, Deepgram, and Vapi globally, and an $8.5 million seed round led by Google’s Gradient Ventures, with Y Combinator backing, validated the thesis: infrastructure born from African constraints, competing at the global frontier.

6. Edge Deploy

Push proprietary AI to the point of use through channels competitors cannot access, with no cloud dependency. Edge Deploy companies build on-device or edge-optimised AI that runs on constrained hardware (low-power microcontrollers, solar-powered devices, SMS-based interfaces) and distribute it through physical agent networks or institutional channels that global tech companies will never pursue. The Technology edge means it works without connectivity. The Distribution edge means it reaches users no one else can. Deployed devices can also collect field data back into the system, creating a secondary Data edge over time.

This is the strategy most aligned with Africa’s deepest infrastructure constraints and the one most likely to produce globally exportable IP. An on-device crop disease detection model optimised to run on a $5 microcontroller in rural Kenya has immediate applications in smallholder agriculture worldwide.

Cactus builds software infrastructure that enables AI models to run directly on smartphones, laptops, and other edge devices instead of relying entirely on the cloud. Its hybrid inference engine automatically routes simple requests to on-device models for low latency, privacy, and lower cost, while sending more complex tasks to frontier cloud models when needed. Through a cross-platform SDK and optimized runtime, developers can deploy speech, vision, and language AI locally with sub-150ms latency, offline capability, and significant cost savings.

Another example is Kenya’s Fastagger who’s building another software infrastructure that enables AI models to run directly on low-power smartphones. Its product Auni, launched on Safaricom’s M-PESA Business app, reads mobile money SMS on-device and gives micro-entrepreneurs AI-powered customer intelligence and business coaching for $1 to $5 a month.

Where to Play Determines How to Win

Each strategy maps most naturally to a specific layer of the stack, and founders who align their strategy to the right layer compound their advantages fastest.

Application Layer

At the vertical layer, where companies serve specific industries or customer segments, three strategies dominate.

• Micro-niche strategy works when the vertical targets customers very few can reach.

• Phygital works when the vertical requires physical infrastructure that generates proprietary data.

• Expert-in-the-Loop works when the vertical demands human judgement that AI augments but cannot replace.

At the horizontal layer, where companies build tools and applications that serve multiple industries, Second Mover Advantage is the natural fit: enter proven categories with adapted technology and pricing that incumbents, burdened by legacy infrastructure, cannot match.

Platform Layer

At the platform layer, where companies own a highly differentiated model honed through access to a unique dataset, Full-Stack is the dominant strategy. A proprietary model alone is commoditising rapidly; what converts a platform-layer company into a durable business is the ability to control the entire chain from data collection through to client operations. This means activating the Data edge (proprietary datasets that compound with usage), the Technology edge (differentiated model architecture or training methodology), the Workflow edge (deep embedding into how customers operate), and the Distribution edge (privileged access to the end users whose activity feeds the data flywheel). Each layer reinforces the next: distribution generates usage, usage generates data, data improves the model, and a better model deepens workflow integration. Companies that stop at the model layer, owning intelligence without owning the relationship, leave themselves exposed to faster-moving application-layer competitors who treat the model as interchangeable infrastructure.

Infrastructure Layer

At the infrastructure layer,where companies build the foundational technology that everything else runs on, Africa’s opportunity is bifurcated:

• Large-scale compute and connectivity infrastructure requires significant capital deployment and is being pursued by established players with balance sheets and regulatory relationships that most startups cannot match. We are watching this space closely but it is not where we expect the majority of venture-scale returns for early-stage investors.

• Edge Deploy pushes AI to the point of use through constrained hardware and channels. It is what users need and it is capital efficient. We believe new computing platforms will emerge that are purpose-built for Africa’s connectivity and power realities, and that the intellectual property created in this process will have global applicability.

These strategies do not guarantee the outcome. But misalignment between layer and strategy is one of the most reliable predictors of failure in our view. An infrastructure company pursuing full-stack will drown in complexity. A platform company pursuing Microniche will never reach scale. Choose the layer, then choose the strategy that compounds within it.

Conclusion

In this new AI paradigm, strategy eats models for breakfast. The most successful African AI companies are not necessarily those building state-of-the-art models, but those deliberately choosing positions where multiple edges reinforce one another, making them progressively harder to displace.

Generative AI and transformer architectures get most of the attention, but they are not the whole game in town. Machine learning has dozens of declinations (from recommender systems, anomaly detection, computer vision, time-series forecasting, optimisation, etc) many of which are perhaps better fits for the problems African companies are actually solving than a large language model would be. Small language models also exist. Boring technology, applied well, compounds. What matters is where you sit, what choices you make, what data you accumulate, and which customer relationships you own.

The most common failure pattern we see is founders trying to activate the wrong edge, or too many edges at once. The playbook we presented here forces a hard question: which edges are essential, which are optional, and which are distractions given where you sit in the value chain and your maturity level?

As the cost of building AI falls, the importance of positioning rises. The companies that get this right will define how we live, work, earn, and thrive in Africa and beyond.

Read Part 1: How to Build the Next AI Champions in Africa: Our AI Thesis (Part 1/2)

Share

Artificial intelligence is the next such shift, and its implications for the African continent may be more profound than any that preceded it. McKinsey estimates that AI could unlock up to $200 billion in annual economic value across African economies,¹ and we expect this to grow as more use cases for AI in Africa emerge.

As investors, we have a front-row seat to this transformation. Every day, we see founders building to leverage the AI opportunity, and in an era where a working prototype can be assembled in hours, the ability to distinguish signal from noise becomes the most consequential investing discipline of this decade.

This piece sets out our framework for doing exactly that. Developed through our work at Enza Capital, it is our evolving thesis for how investors and entrepreneurs can make the most of the AI opportunity across African markets. It is rooted in a simple observation: that the constraints Africa faces (limited capital, sparse infrastructure, fragmented data, scarce talent) are not obstacles to be overcome, but the raw material of competitive advantage.

The AI Stack

We divide the AI ecosystem into three broad layers: infrastructure, platform, and application.

• Infrastructure covers energy access, compute, and data. These are the foundational resources, the AWS and Nvidias of the world, upon which everything else depends.

• Platform encompasses the intelligence models and the tooling that enable developers to build on top of them. This is where OpenAI and Anthropic sit: the layer that transforms raw compute into abundant intelligence.

• Application is where AI meets the customer. This layer spans horizontal tools (productivity, communication, analytics), vertical solutions (fintech, agritech, healthtech, edtech), and consumer products (entertainment, social platforms, personal productivity).

We believe value accrues across the entire stack, albeit with different dynamics and risk-reward profiles at each layer.

• At the top of the stack, companies that own the customer relationship across verticals, and that understand the specific workflows of African companies, can deliver solutions orders of magnitude better than those offered by global competitors disconnected from the realities of the continent.

• At the bottom of the stack, companies building Africa-specific infrastructure and models, designing for low bandwidth, intermittent connectivity, edge devices, and the continent’s extraordinary linguistic diversity, create intellectual property that positions them not merely as African solutions but as global infrastructure providers for the resource-efficient AI era that is coming everywhere.

Each layer carries its own structural tensions. The further down the stack, the more capital-intensive the play and the greater the concentration of talent required. The further up, the more competition intensifies. We will explore these dynamics in greater detail in future installments of this series. For now, we turn to the question that matters most.

How African AI Companies Win

If opportunity exists across the stack, then the critical question is: what builds moat? What gives a company a distinct advantage and a durable right to win in the specific layer it occupies? Through extensive analysis of the African AI landscape, we have identified six value creation patterns that define winning strategies. We call these “edges,” and each represents a strategic capability that compounds into a defensible competitive position over time.

Each edge provides both a benefit (superior economics) and a barrier (something that prevents competitors from arbitraging it away). An advantage without a barrier is a feature. An edge with both is a moat.

Below are the six edges we identified:

The Distribution Edge

Thesis: You win because customers can’t be reached any other way.

In many African markets, where infrastructure can be fragmented and trust is earned offline, distribution is not simply a go-to-market strategy. In these contexts distribution is a moat.

The founders who win progress through three reinforcing levels. First, deep domain expertise, built by spending years inside a sector, shapes what to build, how to price it, and which features matter. Second, those relationships translate into design partners: early adopters who enable iteration against real needs rather than assumptions. Third, this momentum compounds into owned distribution channels, where social proof and embedded trust unlock entire verticals and make customers difficult for competitors to reach.

The Workflow Edge

Thesis: You win because ripping you out would break the client’s operations.

Once inside an organisation, the most defensible AI companies become embedded in operational workflows. Over time they accumulate institutional knowledge, making switching increasingly costly. Replacing the software means rebuilding processes, retraining staff, and migrating historical data. As AI compresses development cycles, competitive advantage shifts from execution speed to strategic positioning: knowing what to build, when to deploy it, and how to weave it into the client lifecycle.

The Business Model Edge

Thesis: You win because your unit economics are structurally inaccessible to incumbents.

Companies that adopt novel business models, particularly those charging per value delivered rather than traditional SaaS metrics, can establish powerful competitive advantages. Incumbents often cannot respond due to legacy infrastructure constraints or fear of cannibalising existing revenue streams. In African markets, this edge is amplified where payment flexibility (pay-per-use, micro-transactions, outcome-based pricing) aligns with customer cash flow realities and willingness to pay. The business model itself becomes a moat when it fundamentally restructures unit economics in ways incumbents cannot replicate. Capital efficiency compounds this advantage, especially in capital-constrained markets.

The Data Edge

Thesis: You win because your dataset cannot be purchased, only grown.

Unique datasets (local languages, proprietary transaction streams, and regulated institutional data) can create durable advantages when combined with a feedback loop. Each interaction improves the model, attracts more users, and generates more data. Without this loop, data decays. With this loop, it compounds. Data can be collected through multiple pathways: original creation, exclusive licensing agreements, digitisation of existing analogue resources, or direct capture. Regulatory compliance and data sovereignty requirements can further strengthen this advantage. Critically, a data edge on its own is not enough. The platform must be intelligent and automated enough to recursively deepen the moat over time.

The Technology Edge

Thesis: You win because your technical IP has global applicability born from local necessity.

The ability to build and control proprietary intelligence, whether through novel architectures, model compression, on-device inference, or distributed training systems, remains one of the most durable moats in AI. Across Africa, this edge is being sharpened by constraint. Innovations designed for intermittent connectivity, limited compute, and milliwatt power budgets do not just serve local markets. They solve problems every emerging market will face, and when protected by patents, they create defensible positions that transcend geography entirely. Founders at this edge may build technology the world will eventually license.

The Economies of Scale Edge

Thesis: You win because your cost structure improves with every unit sold, and competitors cannot catch up without matching your volume.

Companies that achieve economies of scale spread fixed costs (model training, infrastructure, regulatory compliance, data acquisition) across an expanding base of users or transactions, driving marginal costs toward zero while competitors still carry the full burden of their fixed investments. Scale also unlocks negotiating leverage with suppliers, regulators, and distribution partners that smaller competitors simply cannot access. As businesses mature, the Economies of Scale edge becomes the critical moat that separates category leaders from everyone else.

Building From Constraint

Africa’s AI future will not be built by replicating Silicon Valley playbooks. It will be built by founders who know their markets well enough to turn constraints into structural advantages. When capital is scarce, you make sharper choices. When infrastructure is unreliable, you build systems that work without it. When customers cannot pay enterprise rates, you invent pricing that aligns with how they actually earn. Scarcity-induced choices improve unit economics and deepen defensibility faster than capital-rich rivals can replicate.

Our conviction, forged through evaluating dozens of AI businesses over the past year, comes down to this: Africa’s constraints are investable moats. The companies that will win are those that combine multiple edges, own the chain from data to customer relationship, and compound their fundamentals with discipline. In an environment where the cost of build decreases rapidly, continuous moating is the strategy.

In a world where AI increasingly demands resource efficiency, cultural specificity, and operational creativity, Africa’s builders have an opportunity to create something the rest of the world will use. The next generation of globally competitive AI companies will not all come from San Francisco alone. Some of them are being built right now, from constraint, on this continent.

Read Part 2: How to Build the Next AI Champions in Africa: The African AI Playbook (Part 2/2) — where we discuss how African AI companies combine edges into winning strategies.

If you are a founder building against the AI opportunity in Africa and you find these insights useful, we would love to hear from you. Reach out via LinkedIn.

In Africa, the conversation has quietly, but decisively, shifted. AI is no longer an abstract promise or a distant frontier. It is showing up in how work gets done, how services are delivered, and how productivity is being unlocked under real constraints.

We are no longer debating whether AI will matter on the continent. That question is settled. The more interesting questions now are how AI will show up, where value will actually accrue, and which constraints will once again turn into advantages.

Over the past two years, Africa’s AI ecosystem has moved from experimentation to early execution, and back to experimentation. Progress has been uneven and often under‑resourced, and it lacks a common definition. Progress has rarely been linear. But, the advances in Africa’s AI ecosystem have also become unavoidably visible if not existential for many companies, industries, and governments.

The list below is a map of vectors: the directions in which we predict AI in Africa will move in 2026, shaped by global technological shifts, emerging‑market economics, and the realities of building on the continent.

Subscribe now

From Frontier Models to Focused Intelligence

Globally, the AI narrative is evolving. The centre of gravity is slowly shifting away from frontier only progress toward smaller, cheaper systems built for truly accessible and reliable deployment.

This matters disproportionately for Africa.

In 2026, we expect to see more African companies building or fine‑tuning intelligence that is deeply embedded in workflows rather than showcased as standalone capability. This includes generative models, but just as importantly spans classical machine learning, optimisation, rules‑based systems, and natural language processing that quietly improve decision‑making, routing, pricing, credit, fraud, service delivery, and operations.

Local languages, accents, sparse or noisy data, constrained compute, and price‑sensitive customers all favour systems that prioritise robustness and relevance over raw capability. Many of the most effective applications will feel incremental rather than transformative, yet we expect these gains to compound meaningfully over time.

The result here will be more contextually relevant intelligence better fit for purpose.

The Real Race Is for Data

As access to models improves, data increasingly determines who can build defensible businesses.

Africa is one of the most data‑rich yet digitally under‑represented regions in the world. Large parts of the economy still operate through paper records, informal processes, voice interactions, local languages, and legacy systems that were never designed to generate machine‑ready data.

In 2026, we expect a growing share of value creation will come from companies focused on the slow, operational work of data creation rather than data extraction: digitising activity at the source, structuring it cleanly, translating and labelling it appropriately, and embedding feedback loops that improve quality over time.

This pattern mirrors earlier waves in India and Southeast Asia, where logistics, payments, and public‑sector digitisation quietly produced some of the most valuable datasets in those markets, such as the Unified Payment Interface (UPI) in India. In Africa, ownership of high‑signal, legally obtained, continuously refreshed data will increasingly underpin long‑term advantage, often more so than model choice itself.

Compute Moves Closer to Home

Another shift gaining momentum is the gradual localisation of compute.

For years, African builders have relied almost entirely on offshore infrastructure, accepting higher latency, FX exposure, and volatile costs as unavoidable. That assumption is beginning to soften.

In 2026, we expect broader use of hybrid compute strategies that combine global hyperscalers with regional infrastructure, edge deployments, and purpose‑built inference capacity closer to end users. This does not eliminate dependence on global platforms, but it changes how teams design systems, price products, and plan scale.

Cassava Technologies is pushing compute localisation in Africa as on-prem and local data centres for training and inference rise on national agendas. Its rollout of Africa’s first “AI factory”, an NVIDIA-powered data-centre platform, starts in South Africa and expands to Egypt, Kenya, Morocco, and Nigeria. The wedge is GPU-as-a-Service: on-demand high-performance compute for startups, enterprises, and governments without owning or operating clusters. Cassava is targeting ~12,000 GPUs across multiple sites over three to four years. Partnerships with Accenture (to build a sovereign AI cloud) and The Rockefeller Foundation (NGO access) suggest demand from both state and commercial buyers. The result is more in-country capacity shifting system design, pricing, and procurement away from “foreign infrastructure by default” toward hybrid and local options.

When Everything Becomes “AI”

As AI becomes embedded in everyday software, the label itself will start to lose precision.

In 2026, many African technology companies will describe themselves as AI companies. Some will be right. Many will simply be software businesses using AI-enabled components, much like earlier generations adopted cloud, APIs, or mobile-first design. Investors, too, will lean into AI narratives, as capital follows the dominant technological story of the moment.

This is not unique to Africa. Globally, “AI company” is already becoming a broad umbrella rather than a clear category. As intelligence becomes cheaper and more accessible, it stops being a product category and becomes infrastructure.

In this environment, the real distinction is not whether a company uses AI, but where the intelligence sits:

• As a feature that improves efficiency

• As the core engine of the product

• Or as the product itself, learning and compounding over time

The most durable African AI companies will fall into the second and third categories. They will treat intelligence as core operating logic, embedded in pricing, underwriting, routing, fraud detection, customer acquisition, or service delivery. Their advantage will come less from model novelty and more from proprietary data, tight workflow integration, and continuous learning loops.

As this shift unfolds, noise will rise, and more companies will claim AI capabilities. At the same time, signal will become more valuable. Customers, talent, and capital will increasingly reward businesses that can show where intelligence creates real, measurable outcomes.

Over time, the most consequential companies may not describe themselves as AI companies at all. They will simply be the best operators in their markets powered by intelligence that is so deeply embedded it no longer needs to be named.

Agentic AI moves from Demo to Deployment

The next phase of AI adoption in Africa may be defined less by new models and more by new agents.

Agentic AI refers to systems that can take actions across multi-step workflows rather than responding to single prompts. Instead of just generating answers, these systems can handle tasks: managing customer conversations on WhatsApp, reconciling mobile money payments, verifying documents, or coordinating logistics steps across fragmented tools.

This model fits naturally with African operating environments, where many businesses still rely on manual coordination, call centres, and spreadsheet-driven processes. These workflows are repetitive, structured, and economically sensitive, making them strong candidates for agent-driven automation.

In 2026, we expect agentic systems to move from pilots into everyday deployment across fintech, logistics, healthcare, and workforce platforms. The first impact will be operational: faster response times, lower support costs, fewer errors, and more consistent service delivery.

Platforms like Lua AI, which allow companies to integrate AI agents directly into existing communication channels and internal tools, illustrate how this shift will unfold. Rather than rebuilding their stacks, African businesses will increasingly layer agents on top of the systems they already use.

For many companies on the continent, the first meaningful impact of AI will not come from building frontier models. It will come from agents quietly handling the everyday work of the business.

Capital Will Increasingly Overlap

Early capital follows insights and narratives, but more scaled capital will come for results.

There is more capital dedicated to venture in Africa than there ever has been in history, however the pullback from global funds has led to a net decrease in investments into Africa. Investment numbers for 2025 are estimated around $2.4B equity, about 54% lower than the peak of $5.2B equity in 2021.

Globally, AI‑focused funds will continue to proliferate, and locally, African AI companies will increasingly serve global customers. This will push Africa-committed funds to rethink what qualifies as an African business, and it will bring both global and Africa-focused capital to the same set of companies.

In 2026, we expect more overlap between these pools of capital. African companies building AI‑enabled products that are globally competitive will increasingly attract non‑Africa‑dedicated investors engaging with the continent through specific companies rather than broad theses.

Builders Move Faster Than Ever

The cost of automating and building intelligent systems continues to fall.

Open‑source models, improved tooling, AI‑assisted development, and modular infrastructure have compressed the distance between idea and deployment. For African teams, who are often more capital‑constrained than global teams, this lowers the threshold for experimentation and speeds up iteration, similar to how cloud opened the path for SaaS businesses.

In 2026, we expect more products to reach production with leaner teams and shorter build cycles. At the same time, differentiation becomes harder as similar capabilities spread quickly across markets.

What increasingly matters is not exclusively how quickly something can be built, but how clearly a team understands where to apply intelligence, how to integrate it into real workflows, and how to deliver value to customers.

We are likely to see many more companies get started as the cost to build will decrease, however the percentage of startups funded in Africa will likely decrease due to increased competition.

Talent Concentrates, Competition Intensifies

As AI adoption accelerates, so does competition for talent.

Top technical talent across Africa is increasingly concentrating in fewer companies — those with the clearest missions, strongest execution paths, and best access to capital. Global remote demand continues to exert upward pressure on compensation, particularly at the senior level.

Likely outcomes in 2026 include continued wage inflation at the top end and founder–talent alignment becoming a core strategic advantage.

Scale Comes Through Combination

Africa’s markets remain fragmented. AI accelerates the logic for collaboration.

In 2026, we expect more mergers and acquisitions, cross‑border partnerships, shared data strategies, and shared distribution models, particularly in data‑heavy and infrastructure‑adjacent sectors.

Scale will increasingly come not just from organic growth, but from combination.

The African Vector, 2026

Africa’s AI story in 2026 will not be defined by a single technology wave, nor by generative models alone.

It will be shaped by many forms of applied intelligence: machine learning, NLP, optimisation, and generative systems, working quietly inside products, organisations, and public systems. Progress will often appear incremental, but its effects will compound.

The most important work will happen close to the customer and close to the data, under real constraints and real economics. Builders who treat context as an input rather than an inconvenience will continue to find room to compete.

The exact destination remains uncertain, however the direction is clearer than ever.

The vector is becoming visible, even if some of the most consequential companies shaping it do not describe themselves as AI companies at all.

At the same time, evidence shows that acute care delivered safely at home can reduce costs by at least a third while maintaining or improving clinical outcomes. Until recently, however, there was no way to do this safely, at scale, and within regulatory guardrails.

From pure software to AI under Constraint

Value-creating AI applications are not found solely in the pure technical software plays so frequently cited. At Enza Capital, we believe durable opportunities also lie in regulated, physical industries where constraints force architectural choices that deepen defensibility faster than capital-rich rivals can replicate.

As we set out in The Constraint Advantage, Africa’s biggest constraints (scarce capital and talent, fragmented data, and tough operating environments) can be a source of defensibility when founders design with them in mind.

Quro Medical: hospital-level care, at home

Quro Medical, founded by Dr Vuyane Mhlomi and Zikho Pali, is a powerful example. Quro goes beyond telehealth to extend the reach of hospital-quality care into patients’ homes, using wireless patient monitoring and smart software to track vital signs in real time. The service includes in-home visits by nurses or clinical associates, 24/7 access to a multi-disciplinary team of clinicians, and home-based diagnostic and medication management.

Four years ago, Enza Capital led Quro’s seed round, backing Africa’s pioneering technology-driven virtual hospital (Hospital-at-Home) provider. We believed that South Africa’s massive unmet demand for high-quality, affordable outpatient care was a problem that technology could solve. But only with founders who understood both the clinical and regulatory terrain. The Quro team had to prove that its model could work at scale, that medical schemes would pay for it, and that patients would trust receiving hospital-level care at home.

Share

AI as a force multiplier, not a feature

Within this model, AI (specifically machine learning) plays a critical role in effective patient monitoring and treatment.

• A patient’s physiological data can be used to recognise subtle patterns that precede heart failure. Clinicians are alerted early and can intervene before a crisis, reducing the likelihood of escalation.

• A patient’s symptoms can be analysed against Quro Medical’s aggregated, anonymised patient dataset, providing recommendations on suitable patients for Hospital-at-Home programmes in line with doctors’ guidance.

The net effect is improved efficiency in hospital operations, with inpatient beds prioritised for the most urgent cases, lower patient costs through shorter hospital stays where warranted, and more successful treatment outcomes since deterioration can be identified early.

As Sir John Lazar, co-founder of Enza Capital and former Board Member of Quro Medical, puts it:

“Africa’s healthcare systems reveal the difference between technology as a feature and technology as leverage. Quro understood this early. They built within constraints rather than wishing them away, and used AI to extend scarce clinical capacity safely. That’s why their model scaled, and why strategic partners took notice. Quro proved that when you respect constraints and apply intelligence well, you can change an entire sector. That’s the kind of entrepreneurship Africa needs, and has.”

Navigating the trust gap to strategic exit

To deliver this safely, Quro has worked closely with regulators to enable effective monitoring and analysis of patients through the data they collect. In a market where clinical talent is scarce and concentrated in urban centres, Quro made a deliberate architectural choice: build AI that multiplies practitioner reach rather than assumes abundant supply. This scarcity-induced design created a solution that overstretched doctors actively adopted, and that well-funded competitors would struggle to replicate quickly.

Their proven impact has earned them ecosystem trust and partnerships with some of the largest medical scheme providers and private hospital networks in the country. Most recently the Netcare Group (South Africa’s largest private hospital group) acquired a strategic stake in Quro Medical, providing an exit for us and other early investors.

The Netcare acquisition validates more than Quro’s vision. It validates a thesis: AI in regulated traditionally physical industries is investable, scalable, and strategically valuable.

Reflecting on the partnership, Dr. Vuyane Mhlomi noted:

“What I’ve always appreciated about Enza is how founder-friendly and grounded they are. They recognised the constraints we faced and never pressured us to prioritise speed over clinical integrity. That trust allowed us to turn those same constraints into a source of defensibility. Netcare’s investment is a strong validation of that journey and of the support Enza offered us from the beginning.”

One outcome, repeatable strategy

This is precisely the outcome that we look for when we back pioneering technology/tech-enabled businesses: that their solutions move from the cutting edge to the mainstream. However, it also marks the point at which we are no longer at our most effective as an early-stage investor. We know that Quro has so much more in store, but Netcare is the right investor and the right platform for Quro’s next phase of development.

Quro represents the kind of opportunity we will continue pursuing: founders who convert African market constraints into structural advantages, in sectors where AI is not a feature but a force multiplier. We are actively backing the next generation of builders at this intersection.

We are incredibly proud of what Vuyane, Zikho, and the Quro team have achieved and grateful to have been part of their journey. They had the courage to challenge industry orthodoxy, the resilience to push through scepticism, and the capability to prove a contrarian thesis. We know that Quro’s mission to build Africa’s largest virtual hospital continues, and we will keep cheering for them and celebrating their success.

“Scarcity-induced choices that improve unit economics or deepen defensibility faster than capital-rich rivals can replicate.”

In our previous post, we outlined the challenging decisions that African founders face on where to play and how to win. Through a short series, we now focus on Where You Build, How You Operate, and How You Distribute. This note addresses Where You Build using a taxonomy constructed from working with a wide range of founders across Africa.

Taxonomy: Where You Build

Systems of Intelligence (SOI), according to Geoffrey Moore, and later, Jerry Chen (Greylock – here & here), represents the decision layer that learns from enterprise data and improves work. The SOIs are connected to “data chests” roughly categorized as:

• Systems of Record: data stores and process backbones e.g., CRM, HRMS or ERP.

• Systems of Engagement: user facing channels that often link to systems of record e.g., email, Slack, WhatsApp, Tiktok.

AI systems are Systems of Intelligence, learning from Systems of Record and acting through Systems of Engagement. We operationalize that idea in our B2B AI Operating Stack, where Foundational Infrastructure provides the core power, Platform & Enablers connect the data or model to the user, and Applications are the Systems of Intelligence in action.

B2B AI Operating Stack

Exhibit: Operating Stack - Layers that Operationalize Systems of Intelligence

Share The African Vector

We map African B2B AI companies building systems of intelligence into a practical hierarchy.

1. Foundational Infrastructure: This is the base layer of the stack, providing the core power for all systems built on top. It includes the foundation models and compute resources that are the building blocks of AI-native products. In this note we focus on foundation models. We bracket compute and associated physical infrastructure for a later post. Notable African foundation model providers are highlighted below; it should be mentioned that these players have productized their models and delivered them straight to the end user. Examples include:

• Widebot building an Arabic dialect LLM powering customer communication chatbots for businesses and governments.

• Lelapa AI building an LLM for African languages starting with isiZulu transcription services.

• Intron building an LLM for transcription and voice agents across multiple languages and accents e.g., Pidgin English and Swahili.

2. Platform and Enablers: This layer supports the application workflows above it. These companies provide the essential tooling to build, evaluate, and securely deploy performant AI solutions, often as embedded tools or white-label services. The primary archetypes we observe in this category are:

• LLMOps / Data & AI infrastructure e.g., Guepard, Salus Cloud: Tools to prepare and govern data, adapt and evaluate models, and serve them cost effectively.

• Engagement Infrastructure e.g., HelloDuty, Lua: Platforms that aggregate data and access across customer engagement channels to deploy an optimal AI solution where users live.

3. Applications: This is the layer closest to the user and the brand they interact with most directly. We segment this by the end user:

• Internal to the Org (Employees): Solutions designed to make teams faster, more precise and more effective.

• Ops Intelligence: Provide deep insight to improve decision-making and task competency. These tools monitor systems, identify gaps, and suggest improvements. Examples include Pastel for monitoring financial fraud and Leta powering delivery and fleet management to improve logistics outcomes.

• Agent Copilots: Help employees in task execution and ensuring quality. For example, Pidima, which helps manufacturers prepare compliance documentation, and Rescue, building an internal QA tool to accurately audit emergency calls for protocol adherence to improve call handling and dispatch procedures.

• External to the Org (Customers): Solutions that drive behavioral change to enhance customer engagement, retention, and revenue; for example:

  • Flowcart: Deploys AI agents that identify abandoned shopping carts and re-engage customers to complete their purchase increasing conversions.

  • Mindjoy: Provides AI powered STEM tutoring for K-12 and university students to improve learning outcomes.

Each layer carries different capital and technical demands. The most intensive sit at the foundational layer with lengthy and costly R&D cycles that extend your time to market and validation of customer demand (cash earned). As you move up the stack you can test willingness to pay faster, and you should expect higher competitive pressure.

What will separate winners in the next 3 – 5 years is an operating rhythm of punctuated gains, capitalizing on technological step-changes, with consolidation of unit economics before the next technology inflection. We explore this in the next part of the series.

Subscribe now

At Enza Capital, we believe the future of AI will not be built solely in Silicon Valley or Shenzhen, but in Nairobi, Cairo, Lagos, and beyond - not because they’re following global trends, but because they’re solving problems that the rest of the world has ignored. In the last post, we spoke of entrepreneurs who are building brilliant innovations under constraints - ones that work in the messy realities of language, regulation and culture.

Conversational AI is a perfect example. Everyone talks about it as if it’s already global. It isn’t. Most Africans still speak languages that AI platforms barely understand. Arabic, spoken by 250M+ people across North Africa, has been nearly invisible in the global AI revolution. The result has been solutions that misinterpret intent, models that collapse across dialects, and customer interactions that feel clumsy. In sectors like banking, telecoms, and government services, where nuance is everything and mistranslation erodes trust, this isn’t a minor inconvenience; it’s a systemic barrier to digitization. This gap is exactly where the next wave of AI infrastructure needs to be built and it’s why we’ve been actively backing founders solving for it. That search led us to Widebot, and it is why we are thrilled to have co-led Widebot’s $3m Pre-Series A Round.

WideBot is not simply a collection of customer experience tools - it is building the foundational infrastructure for Arabic AI. Its Enterprise and Hulul platforms generate massive volumes of conversational data across 20 dialects, 10+ channels, and thousands of use cases, creating one of the richest Arabic conversational datasets of our generation; while AQL, its proprietary Arabic LLM, transforms that data into increasingly sophisticated generative AI capabilities. Together, these offerings position WideBot not as another customer experience vendor, but as the foundational layer for digital communication and customer intelligence for enterprises, government agencies, and SMEs across the MENA region.

WideBot’s traction speaks for itself. Today, Widebot serves more than 400+ enterprises and has powered over a billion conversations across 20 Arabic dialects. These interactions are the engine of Widebot’s moat. Every conversation feeds into AQL, which in turn powers more accurate and sophisticated products for both enterprises and SMEs. The result is a self-reinforcing flywheel: every conversation makes AQL smarter, every smarter model delivers more value; more value locks in more enterprises, and every new enterprise strengthens Widebot’s defensibility.

Four dynamics point to Widebot’s success and what we’re learning works for AI solutions in Africa:

1. Build a Solution that Speaks the Market: Intelligence that isn’t culturally fluent won’t scale in emerging markets. Widebot is not just “localizing” existing models; they’re creating the foundational layer through AQL that is trained on 20+ dialects across the region. This isn’t a derivative GPT-wrapper bet. It’s an early attempt to lay the foundation for localized AI infrastructure in MENA. In doing so, Widebot is not only unlocking customer experience in the region but also laying the groundwork for entirely new categories of Arabic-first enterprise-grade solutions in finance, commerce, and public services.

2. Sell Ownership, Not Just Access: On-prem turn regulatory barriers into a competitive advantage. In sectors like banking, telecom, and government, which face strict compliance requirements that make reliance on foreign cloud-hosted GPT-like models untenable. Widebot’s on-prem approach ensures sensitive data - be it citizen records or financial transactions - remains fully under its client’s local control. This approach not only accelerates adoption among highly regulated and deep-pocketed actors but positions Widebot as a trusted partner where clients aren’t just passive consumers of foreign technology, but as owners of its infrastructure.

3. Own the Interface: Design for Human Context, Not Just Query: We believe winning in AI will not be about answering queries, but building sophisticated interactions that feel human and build trust. That is why Widebot is rapidly evolving beyond traditional text-based interactions, integrating voice, video, and real-time contextual awareness. Their systems understand tone, context, and sentiment, making voice conversations more dynamic and personalized. Perhaps even more compelling is its avatar offering, which converges AR with VR, offering lifelike experiences that mimic human communication. These combined have shifted Widebot’s role from simple query handling to sophisticated, human-like interaction management.

4. Move Up the Stack to Become Mission-Critical: The most valuable AI companies won’t stop at the interface; they will climb the stack and control more downstream workstreams. Widebot is executing this playbook: using voice agents as an entry point, integrating deeply into adjacent mission-critical workflow tools, such as CRMs and ERPs. The result? WideBot is evolving from a front-end interface into an intelligent (and far stickier) layer, turning it from a vendor into infrastructure.

We’re excited to partner with co-founders Mohamed Nabil, Seif Ahmed, Mohamed Mostafa, and the exceptional WideBot team as they scale the platform. They have been obsessed with solving Arabic conversational AI since 2016 - long before ChatGPT made AI mainstream. Their combination of deep technical expertise and a decade-long focus on Arabic conversational AI sets them apart in a market global players have previously overlooked. Just as importantly, they’re investing in the ecosystem: WideBot launched an in-house training program for fresh graduates, tackling the region’s AI talent gap head-on and cultivating the next generation of AI leaders.

For us, Widebot is another story demonstrating that AI built in Africa can be resilient, scalable, and globally relevant. We can’t wait to see how WideBot continues to redefine how enterprises and governments across MENA engage their customers - and the broader impact this will have on trust, inclusion, and the digital economy in the region. That is why we’re betting on them.

Subscribe now

While building AI solutions in Africa, you will have undoubtedly heard about the limitations: poor infrastructure, fragmented data, constrained spending power, and heterogeneous regulations. However, these very constraints could lead to fundamentally different technical and business architectures that create competitive advantages.

At Enza, our work is driven by identifying the non-obvious patterns that separate leading and enduring companies from the rest in this market. We posit that the most successful startups can thrive both despite African constraints and because of them; this is especially true in the “AI era”. The founders of these businesses while optimizing for inherent scarcity could yield solutions in ways that matter to customers and potentially outperform abundance-optimized alternatives. This is the core rationale behind The Constraint Advantage Thesis.

The Constraint Advantage: Scarcity-induced choices that improve unit economics or deepen defensibility faster than capital-rich rivals can replicate. We propose an evolving thesis that the AI startup constraint advantage operates through the following mechanisms that can turn limitations into strategic assets:

• Capital scarcity leads to discipline.

• Infrastructure limitations drive technical innovation.

• Data fragmentation provides a temporary moat.

• Talent scarcity drives creative team building.

Capital Scarcity Leads to Discipline

African venture funding fell from roughly USD 6.4 billion in 2022 to approximately USD 3.2 billion in 2024. Fewer dollars mean founders must prove unit economics early. Limited runway influences design architectures and forces prudent, difficult decisions. For example, given the high cost of compute and the preferential access afforded to well-funded entities in primary markets, an African founder might opt to procure older model high-performance GPUs from the resale market. This can result in cost savings of 10% to 30% compared to new units. Alternatively, high-performance CPUs can be integrated into workflows to satisfy system component requirements. Such strategic choices contribute to lower CAPEX and burn rates, enabling smaller funding rounds to achieve proportionally similar milestones while keeping founders “equity rich”.

Infrastructure Limitations Drive Technical Innovation

An example best illustrates this; building with a focus on edge infrastructure using heterogeneous compute devices, such as Raspberry Pi 5s, can result in a 90% accuracy trade-off compared to 98% accuracy on cloud-based solutions. However, these edge solutions offer lower operational costs and reduced dependency on connectivity. Depending on the use case, your users may prefer more reliable locally processed solutions over higher-accuracy cloud solutions with greater latency and intermittent failure episodes. The reliability offered today potentially compounds into a loyal customer base while technology advances in model compression and edge hardware narrow the accuracy compromise.

Data Fragmentation Provides a Temporary Moat

Data is a scarce resource in Africa; often unstructured, digitally inaccessible (paper–based), and concentrated within organizations such as telcos, banks, and government entities. This fragmentation can create temporary competitive windows. For example, a transcription-focused large language model builder collaborating with telcos and their BPO solutions provider could gain exclusive access to high-quality data, which may take late entrants another 12 to 24 months to obtain due to lengthy enterprise sales cycles.

The temporary moat provides an opportunity to compete or eventually collaborate with global incumbents whose priorities may not yet include Africa. This positioning is increasingly fortified by the trend towards data sovereignty. The emerging regulatory and strategic imperatives driven by privacy laws like South Africa's POPIA, national security concerns, and preference by large enterprises to control their own data, mandate the processing and storage of African data within the continent, or at least under the governance of local laws. Crucially, this time-limited defensibility is not the end-game. It is a critical opening to quickly iterate and build indispensable solutions reflective of deep understanding of customer needs.

Talent Scarcity Drives Creative Team Building

The market for senior AI practitioners is undeniably tight, however, a distinction must be made. The knowledge base for traditional machine learning is well-embedded, forming the technical backbone of the continent's fintech sector e.g., consumer and SME credit scoring. The primary challenge and opportunity now lie in generative AI.

The constraint of a smaller senior talent pool is forcing the most creative founders to build their teams differently. They accelerate promising junior talent or amplify the capacity of their senior talent by augmenting them with AI-powered coding tools. They are also taking lessons from further-developed engineering markets such as strategically importing senior expertise to mentor local teams e.g., an AI chip designer drawing senior engineering talent from India to train up junior designers. This creativity extends beyond technical roles, as the best founders source business talent by attracting voracious learners from adjacent industries who are drawn to a powerful mission.

II. Market Selection: Where to Play

Where should you build in this constrained environment? Which sectors offer the greatest chances of success? We dedicate significant time to exploring these questions and assessing whether founders are targeting the “right” markets. While it is impossible to predict every opportunity, there are both obvious and less obvious (non-exhaustive) factors to consider when evaluating a market for AI tools.

• Willingness to pay (WTP): Do customers view your solution as a necessary expense or a nice-to-have feature?

• Data availability: Can you cost-effectively acquire exclusive training data before it becomes commoditized?

• Regulations: Are regulatory barriers significant, and how challenging are they to overcome compared to your competitors?

• Domain expertise: With generative AI democratizing application development, deep problem-specific knowledge could become a natural way to defend against competitors who lack a thorough understanding of the issue.

• Addressable market: Is the market large enough to allow for growth without encountering significant competition too early?

WTP Criticality

The relative importance of these factors varies depending on the use case the founder is targeting. We highlight willingness to pay (WTP) which has considerable influence on your business survivability early on. A capital rich startup may have several months to figure this out, depending on investors’ patience. In a resource-constrained environment, it is crucial to quickly assess how much cash users would fork out for your solution and chart a path to determining your business margins.

Fortunately, AI itself provides a powerful lever for rapid affordable market entry and WTP discovery. For example, a startup enhancing employee skills is leveraging large language models and coding assistants to generate assessment content for cohorts of ~200 users at approximately USD 150. This approach significantly reduces the time, coordination effort, and cost than when solely relying on human experts.

As a founder, you may develop WTP assumptions through proxies. For example, an initial indicator of WTP for a data analytics solution provider might include examining how much companies spend on data or analytics as a proportion of their operating expenditure. This may be a useful initial hypothesis; but reliable insights and data are predominantly gained through experience with a live solution in the market, testing your initial assumptions with customers, and iterating based on their feedback.

Nuances of the Factors

The sensitivities and subtleties of these factors further complicate market selection. For instance, you may experience an addressable market misperception: a market with 10,000 businesses and an annual IT budget of USD 20,000 (for offshore solutions) may imply an addressable market of USD 200 million. However, if these businesses are unwilling to pay more than USD 10,000 annually for your locally developed solution, the true addressable market is only up to USD 100 million.

You may also encounter a WTP paradox: markets where customers readily invest in adjacent solutions provide faster validation paths compared to greenfield opportunities that require behavioral change. For example, it may be easier to market a computer-vision add-on for existing security cameras that identifies “unknown persons after hours”. This is simpler than developing and selling a comprehensive computer vision platform that tracks personal protective equipment compliance, production line anomalies, near‑miss detection, and key performance indicator dashboards linked to enterprise resource planning and health safety & environment systems. However, the former may raise critical questions about sustainability. How long will this opportunity last? Given the ease of validation, how many competitors pursuing the same opportunity should you anticipate? All the while, your business has limited cash to explore various paths, reach key milestones, and continue advancing toward your vision.

Staged Learning Investments

In a resource-constrained environment, it is essential to strategically stage learning investments. Set clear goals tied to tangible metrics that are critical to business survival; prioritize generating recurring cash inflow while carefully managing cash outflow. Structure your journey as a series of experiments linked to clear returns on invested capital and where possible sequence your irreversible bets last as you maximize opportunities for validation.

A prime example is a B2B voice AI provider that has architected a flexible platform using third party voice models. This allows them to: develop in-house core agents based on fine-tuned open source models for accent synthesis, leverage premium closed-source APIs for high-stakes tasks, and allow enterprise customers to 'bring their own' self-hosted open-source models. By instantiating agents per call across parallelized vector databases and containers, a substantial proportion of fixed costs are transformed into variable costs. They are discovering the most performant and economically sustainable model before scaling.

III. Business Architecture: How to Play

How do you set up the business to compete in the market that you have chosen? What drives those decisions? For a B2B AI company, we will focus on three foundational design choices:

• Workflow integration level

• Foundational model type

• Resource allocation

These are not independent decisions; they are an interconnected set of choices that form your company’s operational DNA. Your approach to workflow integration is a direct response to infrastructure limits and the need to acquire fragmented data. Your choice of foundational model is dictated by the trade-offs between capital and talent scarcity. And your resource allocation is the expression of the discipline forced by that capital scarcity.

Workflow Integration Level

How deeply you integrate determines both customer adoption speed and competitive defensibility. We propose three stylized integration levels that you could optimize for:

• Overlay: A stand-alone solution that operates independently of the customer's core workflow e.g., SeamlessHR which is a separate cloud based HR and payroll software system.

• Sidecar: A solution that runs parallel to the customer's core system, supporting functionality or providing reports triggered by specific events. The user may not actively interact with it unless prompted e.g., TransUnion API marketplace which offers a suite of credit, identity and risk‑scoring APIs that integrate alongside existing lending or onboarding systems.

• Inline: A solution deeply embedded into the customer's workflow, playing a critical role in real-time processes e.g., Paystack whose payment APIs are embedded directly into merchants’ checkout flows.

Practically, these stylized modes exist on a spectrum and are useful to frame the impact to both the sales cycle and defensibility. The sales cycle reflects how quickly you can validate customers' willingness to pay (WTP) and start generating revenue. Defensibility ensures long-term resilience against emerging competitors. An overlay solution may sell faster due to its ease of testing and usage; however, disengagement is also more likely. Sidecar solutions, while harder to abandon, often require a longer purchase decision timeline. Inline solutions face the greatest adoption friction, but they also offer the strongest defensibility, as abandoning them would disrupt key workflows. Moreover, defensibility can be further enhanced by leveraging frequent and high-quality data access to refine your solution; inline, potentially providing the best access constrained by data regulations. Viewed through an AI lens, this is not just a product choice; it serves as your data acquisition strategy.

The choice between these integration levels often depends on factors such as customer trust, resource availability, and market structure. An overlay approach might be ideal for seeding your company due to its simplicity and validation potential or the endgame depending on customer friction relative to retained value. A sidecar solution might provide a more defensible pragmatic middle ground to serve customer needs effectively in the medium to long term. The proposed framework of stylized integration levels blurs in reality and likely requires ongoing reassessment as you grow following your workflow design — influenced by domain knowledge. You remain a moving target as you refine your market edge. Your decisions on the integration level will be informed by customer switching costs, trust levels, and or sales cycle durations in relation to your runway.

Foundational Model Type

The foundational model type decision belies a critical strategic commitment influencing your unit economics and defensibility.

The “closed” route

You leverage state of the art capabilities of the labs’ models accessed via APIs. Your time to market and revenue is fast; however, you give up self-direction and face potentially high variable costs as you scale. This is especially true if your identified use case requires the latest model for all workflows with significant caching and context windows per user interaction. You are building on rented space with a risk of disintermediation if the labs decide to cost-effectively serve your target customer. Your defensibility likely emerges from deeply embedding into your customer’s workflow (i.e., inline) – better than anyone else. The model becomes a utility, and your value comes from understanding your customer’s needs (rapidly translated into an amazing user experience) and the earned trust you wrap around the underlying model. And as you become indispensable, you gain pricing power.

The “open” route

You leverage open-source models to build a solution radically optimized for local constraints. You take a bet on turning a proprietary data advantage into a technical one. You have more sovereignty and need strong technical talent to finetune models and conduct optimizations where applicable. This implies a significant time and financial investment creating significant pressure on your runway. You must validate quickly and start bringing in cash.

While you prototype and validate, the cost may be low with optimized architectures. However, as usage grows, you may face ballooning setup and maintenance costs if you gradually become infrastructure for others to rent. In this state, your users define when your tool triggers in their product workflow. You may achieve a technical edge as these users expand your distribution and provide enriched context specific data. But you are in a race to offer the strongest capabilities to power intermediate users’ applications and could have limited pricing power. Sitting close to the end customer, enables you to gain more of the customer’s share of wallet; this doesn’t necessarily imply a reduction in performance demands and the associated compute resources.

The foundational model type decision is not binary and static. Hybrid approaches can and are being employed in the market. For example, finetuning an “open” model to better interpret African accents and language structures while integrating with a “closed” model to undertake text to speech and speech to text translations for a voice AI agent. And while you may start prototyping a use-case using a “closed” model, you may transition to an “open” model for finer-grained control.

We underscore again a crucial principle of defensibility in the AI era: a moat is not a static achievement you build once. You must continually invent and or innovate to earn the right to serve your customers. Cultivate a fast shipping culture — part of your DNA to build evolving moats in the “AI era”. Your choice on the foundational model will ultimately depend on — how much value you can gain from your customers, the cost of serving them given your underlying architecture and your level of integration into their systems.

Resource Allocation

We end with brief commentary on your execution. We are not well positioned to know the best way to spend the limited resources you have. We are best suited to evaluate if how you are spending these resources matches what you say you are prioritizing and whether it will enable you to get to your next goal. Plainly, this is not about accounting but rather sequencing, appropriate to your cash runway, stated objectives, and validated traction. Your resource allocation choices are tested across the following fronts:

1. People: Who you hire and when determines your capabilities. If you take the closed route, you may iterate and set up a working product faster than if you took the open route. It could be justifiable to have a closely matched proportion of sales personnel to your product team after customer validation — getting the product into as many paying hands as possible.

2. Product Validation vs Distribution: Your sequenced spending on product versus distribution is a test of discipline. For instance, if your goal is to achieve product market fit, but you are spending time negotiating with resellers, this may imply a contradiction to the stated objective of deeply validating the product with a target core customer base. That is not to suggest that you cannot further validate the product perhaps with new customer groups while expanding distribution, but there is an appropriate time (signaled by your traction) at which both can concurrently occur.

3. Discovery vs Deepening: The focus on deepening your current market versus discovering a new one defines your growth trajectory. You may identify a new entry market and observe an opportunity to deepen in your current market. You need to decide on how much you spend (time and cash) on either of the markets. A strong founder may continue building on the existing market while tying entry into the new market on a critical amount of customer dollars earned based on a validated GTM strategy before swinging for the fences in this new market.

Resource allocation demands discipline and decisive action. You have to carefully calibrate to ensure business survivability and success.

IV. Build to Last

We reiterate the need for continuous, rapid evolution in the AI era taking deliberate action cognizant of resource limitations. Founders face a multitude of decisions to build sustainable businesses. These choices lie on a spectrum and are not black or white. However, we believe painstaking consideration of the factors that influence and shape your business will enable you to make the optimal choices to “build to last”. African startups are not excluded from this achievement because of the constraints they encounter; rather, it is by embracing these constraints that they will foster the very innovation needed to win.

As always, keep building.

Enza Capital invests in teams using technology to solve large and meaningful problems across fintech, logistics & mobility, human capital, energy & climate, and enterprise infrastructure in Africa. This publication distills our on-the-ground learnings into practical tools.

We publish when we have something worth your time. For the most value, please subscribe to get new posts in your inbox. Please reply or comment with examples from your work — we want to learn from you — and share posts with operators who will benefit.

We are glad you are here.
The Enza Capital Team

Subscribe now

Share The African Vector