The rapid integration of generative artificial intelligence (GenAI) is transforming personal financial advisory services. While Large Language Models (LLMs) have demonstrated unprecedented strength in complex knowledge synthesis and macroeconomic sentiment analysis, their application is often constrained by high computational costs, latency, and significant data privacy issues. This paper investigates the new potential of small language models (SLMs) as an efficient, domain-specific alternative in the financial industry. While there is growth in GenAI adoption in India, there is a research gap in terms of the practical efficacy of different model architectures in nuanced financial scenarios (e.g., mitigating look-ahead bias and managing sensitive client data). This study employs a conceptual and comparative framework to distinguish the functional roles of LLMs and SLMs. It measures the performance of ten key financial advisory dimensions using industry data and current academic research. The results show that LLMs are better at macro-level tasks such as estate planning and insurance analysis. SLMs perform better for micro-level operations like cash management and debt management. The study concludes by suggesting a hybrid design that combines the cognitive power of LLMs and the localised precision of SLMs. This dual-model approach will be recognised as the best way to deliver scalable, secure, personalised digital wealth management in emerging economies.