The context
Membership organizations plan against demographic change they can feel but rarely quantify. Leadership needed a forecast rigorous enough to build strategy on, plus levers they could test themselves rather than a static chart.
The method
A cohort-replacement forecast designed, built, and quality-gated by an autonomous overnight multi-agent AI run. Two candidate models were built and judged in an independent bake-off; the winner backtested to under 1% error on the full sample and was validated against a blind holdout. The deliverables include an interactive scenario simulator that lets leadership test recruitment, retention, and pricing levers directly.
What it demonstrates
Quantitative strategy work run by AI can be held to the same standards as work run by a research team: competing models, backtests, and independent judging before anything reached a decision-maker. The overnight-run pattern, a defined contract with gated quality checks, generalizes to any analysis too large for a single working session.