Your internal data only records the interactions you have already had. What is far better documented — and far more stable — is human behaviour itself: how age conditions income, income conditions price sensitivity, digital habits condition response to a format.
The constraints you know arrive aggregated and disconnected. MCMC methods disaggregate them into coherent individuals, Bayesian networks encode the causal chains between their attributes, and VAEs plus self-organising maps turn the population into a navigable geography where similar profiles sit close together — which is how you find demand that no traditional metric has surfaced yet.