The AI industry has spent the last three years obsessing over agents that do things — write emails, close tickets, generate code. That is the lower-leverage opportunity. The higher-leverage opportunity, and the one almost no one in Africa is pursuing, is AI agents that model reality. Agents that simulate how populations think, react, and decide before the decisions go live.
This is the founding thesis of Kento: that simulating South Africa's 62 million people through thousands of calibrated AI agents will unlock a class of predictive intelligence that no survey panel, loyalty programme, or traditional research methodology can match. Not because those methods are bad — but because simulation operates on a fundamentally different value curve.
"Simulation turns a 10-year market cycle into a 10-hour thought experiment. It compresses iteration time from weeks to hours — and the compound effect of that speed is not linear."
The Science Behind It
This is not speculation. In a landmark study, researchers at Stanford University built a generative agent architecture that simulates the attitudes and behaviours of 1,052 real individuals. Each agent was constructed by combining a detailed qualitative interview transcript with a large language model. The results were striking.
The generative agents replicated participants' responses on the General Social Survey 85% as accurately as the participants replicated their own answers two weeks later. On personality prediction and behavioural economics games — dictator game, trust game, prisoner's dilemma — the agents performed comparably to real humans. The finding held across demographics, income groups, and political identities.
Separately, EY independently validated a similar approach using a commercial simulation platform, comparing simulated responses against a 3,600-person traditional survey. They reported 90% correlation — and completed the study in a single day instead of weeks.
The scientific question — can AI agents accurately simulate human populations — has been answered. The question Kento is built to answer is: can it work for South Africa?
Why South Africa Is Different
Every synthetic research platform operating today — the unicorns and the challengers alike — was built on Western population data. Primarily American. Primarily English-speaking. Primarily built to model consumers who live in stable infrastructure, speak one dominant language, and sit within a handful of recognisable socioeconomic buckets.
South Africa is none of those things.
Eleven official languages. The Living Standards Measure framework as the standard market segmentation tool — absent from every global platform. Township, suburban, and rural behavioural splits that produce fundamentally different consumer journeys. Informal retail channels that account for a significant share of everyday commerce. Coalition-era ward-level politics following the 2024 elections. Infrastructure instability — load-shedding, water, transport — that reshapes daily decision-making in ways no stable-infrastructure model can replicate. Data cost sensitivity that determines which products people access and which they do not.
These are not edge cases. These are the fundamental axes along which South African consumer, citizen, and voter behaviour operates. A foreign platform cannot import its way into understanding them. The calibration required is local, deep, and continuous.
"This complexity is not a liability — it is the moat. A global competitor entering South Africa would need years of local data work, cultural calibration, and institutional relationships to match what Kento is building from day one."
The Category Has Already Been Validated
We are not asking the market to believe in a new idea. The global market has already decided. In under two years, the synthetic population simulation category went from academic curiosity to venture-backed juggernaut. The leading platform in the space achieved a one-billion-dollar headline valuation by December 2025, with enterprise partnerships across the world's largest consulting and advertising firms. Investors were not pricing current revenue — they were pricing a category that is about to reshape the $140 billion market research industry.
Kento does not need to convince anyone that simulation works. That work has been done. What Kento needs to demonstrate is that it works here — for the specific, complex, multilingual, multi-ethnic reality of South Africa. That is a narrower thesis and a more achievable one.
What We Are Building
Kento is not one product — it is a population simulation platform with three specialised applications, each purpose-built for a different decision-making context.
Isimo — Business
Isimo is built for the corporate market. Brand managers, product strategists, and market researchers who need to test hypotheses before committing budget. A typical use case: a brand preparing to launch a product targeting LSM 5–7 consumers in Gauteng can assemble thousands of synthetic agents matching that exact profile, test packaging variants, price points, and messaging — in hours, not weeks, at a fraction of traditional research cost. The result is not a replacement for human research. It is a precision filter that ensures the questions asked of real people are the right ones.
Imbizo — Government
Imbizo serves government departments and policy consultancies. The challenge in public policy research is reach — traditional consultation processes systematically miss rural populations, informal settlement residents, and non-English speakers. Imbizo simulates responses across all nine provinces, all eleven language groups, and the full LSM spectrum, including the populations that surveys routinely fail to capture. A department preparing to roll out a national programme can test public communication strategies, anticipate resistance points, and refine messaging before a single rand of implementation budget is spent.
Inkundla — Politics
Inkundla is built for the coalition era. South Africa's 2024 elections produced a new political landscape — one where no party holds a majority and ward-level dynamics determine coalition stability, floor-crossing risk, and local election outcomes. Traditional polling cannot deliver the granularity or speed that this environment demands. Inkundla simulates ward-by-ward voter sentiment, models turnout under different scenarios, and answers counterfactual questions that polling data cannot: what happens to a party's support in KwaZulu-Natal if a coalition collapses? What is the message that moves the undecided voter in a specific municipality? These are questions that have never been answerable before.
Who This Is For
Kento is built for every decision-maker who has ever needed to know what would happen before it happened.
For corporate clients: the FMCG brand manager testing a product launch, the bank designing a micro-credit product for informal traders, the telco modelling the demand impact of a pricing change.
For government and public sector: the department that needs to understand how citizens will respond to policy, not after implementation, but before.
For political and campaign organisations: strategists who need ward-level intelligence, message testing, and coalition scenario modelling on a timeline that polling cannot match.
For research and consulting firms: the consultancies that hold government contracts, run public research programmes, and advise on strategy — and who can now offer their clients a layer of predictive intelligence that supplements, and in many cases accelerates, their existing methodologies.
"Rewards data tells you what people did. Simulation tells you what people would do under conditions that have not yet occurred."
A Note on What Simulation Is Not
Simulation is not a replacement for human voices. It is not a claim that AI understands people better than people understand themselves. It is a tool for exploring futures — a structured way of asking "what if" at a scale and speed that was previously impossible.
The strongest objection to synthetic research is a fair one: loyalty programmes and transaction data already tell businesses what people do. Why pay for simulated behaviour when you can observe real behaviour?
The answer is the distinction between descriptive and predictive intelligence. Observed data tells you what has happened. Simulation tells you what would happen under conditions that have not yet occurred. A retailer with transaction data knows their customers bought more chicken last month. Kento tells them what would happen to purchasing behaviour if chicken prices increased by 15% during load-shedding season while a competitor launched a cheaper alternative in the same LSM bracket. That question has no answer in any dataset, because the scenario has not happened yet. Simulation is the only methodology that can address it.
The Moral Case
Policy decisions in South Africa affect 62 million people. Product launches shape daily lives of consumers who navigate economic precarity, load-shedding, and infrastructure failure. Political dynamics determine service delivery for communities that depend on it most.
If simulation technology can meaningfully improve the quality of these decisions — if a government department can anticipate how its citizens will respond to a policy before the policy lands, if a company can avoid launching a product that misses its market entirely, if a political party can understand the real concerns of its voters before an election rather than after — then the compound impact of better-informed decisions, multiplied across millions of people and thousands of decisions, is enormous.
Kento does not claim to replace human judgment. It claims to inform it. To give decision-makers across corporate boardrooms, government departments, and campaign rooms access to a layer of predictive intelligence that was previously available only to the largest organisations in the world's wealthiest markets.
South Africa deserves to be simulated — accurately, respectfully, and in service of better decisions for its people.