Use K-means when segments are the goal
In the Airbnb dashboard, K-means was useful because the product question was explicitly about discovering interpretable host segments across price, engagement and portfolio size.
Use DBSCAN when outliers matter
In the wellbeing project, the important question was whether some student profiles deviated from cohort norms. DBSCAN is useful because it can label sparse points as noise instead of forcing every point into a cluster.
Scaling changes the result
Distance-based clustering is sensitive to feature scale. Standardisation and careful feature selection should happen before interpreting clusters.
Clusters are not causes
A cluster describes similarity under the chosen features. Product copy should avoid turning those descriptive patterns into unsupported causal claims.