About us
Thirty-five years of research, turned into a platform.
The Prosperity methodology was not designed in an office. It is backed by more than 35 years of research and implementation in the regime of extreme multidimensional poverty, walking alongside people and households on their way to prosperity. That work is documented in a book, and that book is the basis of everything the platform does.
The book
The methodology is published and documented: the dimensions, the thresholds and the weights come from there, not from an improvised model.
35 years in the field
Sustained research and implementation in the regime of extreme multidimensional poverty, with real households and real field teams.
An engine that learns
The plan does not come from fixed rules. The system proposes routes, observes what happened in similar households and adjusts what it recommends. Every plan taken up improves the next one.
Built to last
We take that body of knowledge and turn it into a product: evaluation, plan and instruments operating at scale, traceable back to the answer every number came from.
The engine
A system that plans over horizons of years.
The engine optimises one thing: how much of what the household is missing gets closed. It searches for the sequence of instruments that most reduces its gap across the four dimensions, over horizons of years and subject to what the family can sustain month by month. It does not classify households: it builds each one a way out.
Sequential planning
The system does not pick a product: it explores combinations of insurance, assistance, training and credit ordered in time, and evaluates the whole trajectory. A cheap instrument today can unlock a large one in month 18.
Multidimensional reward
The learning signal is the weighted closing of the gap in each dimension, measured in later reviews. What the engine pursues is prosperity; what the family can sustain each month is the limit it respects, never the goal.
Counterfactual evaluation
Before recommending a new route we test it against historical cohorts of comparable households, with off-policy methods. No strategy reaches the field on intuition.
The model proposes, the rules bound it
A closed catalogue of instruments, hard limits on capacity to pay, and traceability of every number back to the answer it came from. Every recommendation is auditable and reversible.
Today that engine runs on reinforcement learning agents. Tomorrow it will run on whatever works better. What does not change is the principle: the system learns from results measured in households, not from opinions.
Bring a real case and we assess it with you.
Request a demo