Machine learning identifies patterns in data to support predictions or decisions, adapting as conditions change rather than relying only on fixed rules. In finance, it can draw on varied data to inform high-stakes decisions shaped by demands for explainability, fairness, and compliance. section 1.1
A cost-aware system can use traditional machine learning as a first-pass filter, reserving more expensive LLM analysis for cases that warrant it. section 14.2.1
Retention screening
A churn score screens 100,000 users and flags about 7,000 as high-risk candidates for deeper LLM-based analysis. The ML filter runs in milliseconds of CPU time and reduces LLM calls by about 93%. section 14.2.1
Figure 14.2 The cost-aware retention intelligence pipeline. Traditional ML (Churn Score model) filters 100,000 users down to approximately 7,000 high-risk candidates. Only these candidates proceed to the LLM-based retention agent for nuanced analysis and strategy generation. This architecture reduces LLM costs by 93% while preserving reasoning quality where it matters most.
Shows the filter-first pipeline: traditional ML narrows the population before LLM-based analysis. section 14.2.1
