Decisions Lab recreated a real restaurant food-waste intervention funded by the Hong Kong Government and matched about 95% of the real-world weekday effect in simulation. Our model produced a 40% reduction in weekday food waste, versus 41.9% in the live pilot.
Success Metrics
- 95% match to the real-world weekday result
- 40% simulated weekday food-waste reduction
- 41.9% weekday reduction in the live pilot- Hong Kong Government-funded benchmark study
Dr. Daisy Lee's study tested whether simple nudges in restaurants could reduce plate waste. The live intervention gave diners options like less rice, garnish opt-outs, and clearer menu information, then measured the effect on waste. The official report found a 41.9% drop in weighted weekday plate waste.
We set out to test a narrower question: can this behavioral outcome be reproduced in simulation closely enough to be useful?
What we modeled
The simulation recreated the same core decision system:
- Diners with different preferences and habits
- Restaurant staff and operators
- Menu and ordering flow
- Nudges such as portion-size reduction and garnish opt-outs
- Resulting food waste, measured against the same outcome logic
This mattered because the result was not just about opinions. It depended on how people actually behaved inside a real environment.
The result
Our simulation produced a 40% reduction in weekday food waste. The live pilot produced 41.9%. That means the simulated effect matched about 95% of the real-world result.
That is the core proof point in this case.
It shows that our system can model human decisions and downstream behavior with a high degree of fidelity against a real benchmark, not just generate plausible-sounding responses.
Why this matters
If a simulation engine can get close to a real behavioral outcome in a government-funded field study, that increases confidence in using the same modeling approach in other decision contexts.
That does not mean every future simulation will have the same accuracy. But it does show that our modeling can move beyond surface-level opinion and approximate how people respond when real choices, constraints, and incentives are involved.