Helping smokers quit for good using machine learning
0 min read
The challenge
Most smoking cessation programs send everyone the same messages, and most smokers ignore them. A national cancer charity was building an SMS quit-support service and asked us to design its structure on evidence: which messages, to whom, at what moments.
The solution
We reviewed 15 existing cessation programs and more than 30 academic publications, mapping barriers and drivers to quitting through the COM-B model of behavior change, then surveyed 122 smokers and ex-smokers. We analyzed the data three ways: descriptive statistics on the population, inferential tests across subgroups, and machine learning (random forests, decision trees, and cluster analysis) to find the structure demographics miss.
The decision tree revealed that identity comes first: whether smoking is part of who someone is separates smokers before risk tolerance or social support does. Four groups emerged, from high-identity smokers whose main barrier is smoking company, to low-motivation smokers who distrust the services on offer. For each group we built a dedicated SMS conversation flow, with message themes matched to that group's barriers and message timing tied to each user's declared high-risk moments of the day.
The impact
Four evidence-backed smoker profiles now anchor the service's design, and each carries a complete conversation tree the charity can deploy directly. Because the profiles describe barriers and drivers rather than one channel, they transfer beyond SMS: the charity can reuse them as the behavioral basis for its other cessation programs and services.
