Get Guardian on GoogleAfrican countries are actively developing the digital economy, with artificial intelligence increasingly becoming an integral component of everyday digital services. Nigeria offers a particularly illustrative example, as it is Africa’s most populous country. According to the Our Life with AI 2025 study by Google and Ipsos, 84% of internet users surveyed in Nigeria had used generative AI in the previous 12 months. By comparison, the global average in the study was 62%.
The growth of digital services is creating new opportunities for developers. According to DataReportal, Nigeria had approximately 109 million internet users as of October 2025, with internet penetration reaching 45.5% of the population. This represents a substantial potential audience for mobile products, although access to digital services remains unevenly distributed.
I faced these precise challenges when founding Eda AI, an AI-powered nutrition and meal-tracking application. Before launching Eda AI, my background was rooted in marketing and online education; my venture, Pulatov School, attracted over 10,000 paying clients. That experience provided deep insight into what motivates consumers to engage with a product and ultimately commit to a purchase.
When designing Eda AI, I drew on both my professional background and my personal experiences as a user. I knew firsthand how tedious calorie counting could be. Manual data entry is time-consuming and steadily erodes user motivation. We set out to streamline this workflow by allowing users to receive instant estimates of calories and macronutrients simply by taking a photo of their meal. Unlike online courses, nutrition involves daily decision-making, which creates a natural dynamic for repeat app engagement. However, frequency of use alone does not guarantee long-term user retention.
Localisation as a Competitive Advantage
Before expanding a product into any given market, developers must rigorously analyse the habits and expectations of the local audience, studying how consumers currently solve their problems and which tools they rely on.
This localised approach is especially critical for health and nutrition applications. Nigeria and Uzbekistan place great importance on these issues. Moreover, both countries have their own distinct culinary traditions. That is why an app designed to help people manage their weight needs to take into account not only universal nutritional principles but also the dishes people eat every day.
Global calorie-tracking platforms reliably recognise common international foods but often struggle with regional cuisines. The variety of recipes, ingredients, and cooking methods makes it more challenging for algorithms to correctly identify the composition and calorie content of such dishes. This is indirectly supported by a study on digital dietary assessment in West Africa, published in the journal Nutrients. The study developed a food composition database for Nigeria covering 924 food items and dishes, with data collected on up to 54 nutrients for each.
Living and working in Uzbekistan, I have come to similar conclusions. There are dozens of different varieties of plov here. Two visually identical servings can differ dramatically in their proportions of rice, meat, and oil. An image alone cannot definitively reveal an exact recipe or ingredient breakdown. To resolve this, we progressively built a proprietary dataset that enabled our model to distinguish regional dishes with higher precision. Additionally, we introduced manual editing capabilities so users could adjust entries when they knew the exact ingredients used.
During its first year, Eda AI reached approximately 2.9 million downloads — a figure representing nearly 10% of Uzbekistan’s active internet audience. However, download volume reflects only initial consumer interest. To evaluate whether a product delivers enduring value, developers must track retention: the percentage of users who continue using the app over defined intervals post-install. Through continuous iteration, we increased our six-month retention rate to 32%.
The Cost of Consistent Value
While retention metrics indicate whether a user continues to interact with a product, they do not answer whether that user is willing to pay for it.
A user may download an application, test its primary feature, and never return. To prevent drop-off, a service must integrate seamlessly into daily routines. Simultaneously, subscription pricing must balance audience purchasing power with product unit economics: an inflated price point suppresses conversion rates, while underpricing fails to cover operational overhead.
AI applications carry a distinct cost structure that must be factored into financial modelling: model inference introduces variable unit costs. The more frequently users trigger compute-heavy models, the higher the infrastructure expenses climb. On the other hand, modern AI tools enable lean engineering teams to build and maintain sophisticated applications without massive headcounts, though net margins ultimately depend on product architecture, API utilisation volume, and infrastructure design.
Customer acquisition represents another major expenditure line. We secured our initial 20,000 installs through short video ads demonstrating the core user journey alongside my personal experience using the product. Today, our distribution operates at a significantly larger scale: we partner with over 400 content creators while running paid campaigns across Meta Ads and Google Ads. Throughout this expansion, we have adhered to a foundational principle: the simpler it is to explain an application’s core functionality, the easier it becomes to validate creative messaging and scale high-performing acquisition channels.
The Future of Consumer AI Applications
When scaling a product, founders must look beyond immediate metrics and anticipate how user expectations will evolve over time. While underlying technologies rapidly advance, core human desires remain unchanged: consumers want to save time and money, extract greater utility, and solve everyday tasks with minimal friction. Therefore, I view the future of consumer AI apps not merely through the lens of incremental model capabilities but through the radical reduction of manual steps required to achieve an outcome.
Eda AI has already replaced manual calorie logging with a single photo scan. In my view, the next evolutionary leap will be full process automation. Looking ahead, wearable hardware, including smart glasses, will eventually capture relevant data ambiently, allowing applications to function seamlessly in the background. I refer to these as “invisible products”: solutions where human interaction is minimized, yet the desired output is consistently delivered. This represents a forward-looking vision for technological trajectory rather than an active Eda AI feature today.
However, full automation introduces a new set of challenges: ensuring decision accuracy, safeguarding user privacy, and maintaining user agency over ambient data collection. As direct interaction with software diminishes, explicit user consent and operational transparency become paramount. Companies will also face the burden of clearly communicating the value proposition to justify subscription pricing for invisible workflows.
Ultimately, the long-term winners in the consumer AI space will not be determined solely by raw model metrics or hardware form factors. Victory will belong to the teams that eliminate unnecessary user effort, build unwavering trust, and pair seamless user experiences with sound, sustainable business models.
Ulugbek Pulatov is an Entrepreneur, founder, and CEO of consumer AI app studio Thinking Apps (Founder of Eda AI)
