Platform · Explainable AI
Why should a risk committee believe it?
Every score comes with an explanation. SHAP-based attribution and published model cards show exactly which factors drive a result — the transparency institutional risk committees require before capital moves.
- 1,000
- draw Monte Carlo robustness tests
- 100%
- of inputs provenance-flagged
Transparent decision making
What Explainable AI is
Kilwa's Explainable AI layer is the transparency surface behind every published score. It returns SHAP-based attribution showing which indicators drove a result and in which direction, alongside model cards documenting scope, training data, known weaknesses and the conditions that would invalidate the score.
An unexplained score is not usable in an institutional process, however accurate it is. Kilwa publishes the decomposition alongside the number: which indicators contributed, in which direction, and by how much.
Model cards document what each model does, what it was trained on, where it is weak, and what would invalidate it. Inputs carry verified or estimate flags. Robustness checks — re-weighting and 1,000-draw Monte Carlo tests — are published in the appendices of the flagship reports rather than held back.
Kilwa runs two kinds of score and says which is which. Report scores are weighted composites of house assessments — structured risk rankings. ISI and METI are modelled. Neither carries a published out-of-sample track record, so neither is described as a validated predictive model: using predictive methods and having demonstrated predictive accuracy are separate claims.
Capabilities
SHAP attribution
Per-indicator contribution and direction for every score we publish.
Model cards
Scope, training data, known weaknesses and invalidation conditions, documented.
Verified / estimate flags
Every input is labelled by provenance. Estimates are never presented as measurements.
Published robustness tests
Re-weighting and 1,000-draw Monte Carlo tests in the report appendices.
Seen in the research
The Four Percent Trap
A Kilwa reading of the World Bank's April 2026 outlook, extended to all 54 markets, scored for distance from takeoff, and repriced for the war premium.
Read the analysis
Get started
See Explainable AI running across our covered markets.
We'll walk your team through the module on the markets you follow.