<p>Financial planning for individuals jointly determines consumption and investment to achieve multiple prioritized goals over the life cycle and is commonly formulated as a multi-stage stochastic asset–liability management problem. Solving such problems repeatedly for many clients is computationally demanding, which limits the scalability of personalized financial planning services. We introduce Deep Financial Planning (DFP), a decision framework that treats the individual goal-based planning problem as a parametric stochastic program and approximates its optimal policy with deep neural networks. In DFP, we first solve a representative set of multi-stage stochastic goal programming problems under diverse parameter configurations and then use the resulting optimal asset and goal allocation policies as training data for the network. We provide theoretical results that establish conditions under which the optimal policy of the underlying parametric problem can be approximated arbitrarily well by a neural network. Numerical experiments on individual financial planning scenarios show that DFP delivers near-optimal policies in unconstrained and moderately constrained settings with substantial reductions in response time relative to directly solving the stochasticprograms, enabling real-time what-if analysis for end investors. We further demonstrate that transfer learning can enhance the efficiency and accuracy of DFP, reinforcing the practical viability of DFP as a scalable tool for lifelong personalized financial planning.</p>