In a live customer scenario, an enterprise reports that their application became slower and more expensive after switching to a new model configuration. You are given time-series samples for request rate, prompt/completion tokens, cache hit rate, retries, timeout rate, and per-endpoint latency/cost before and after the change. Write a program that identifies the most likely root causes, ranks them by impact, and returns both a technical diagnosis and a customer-safe response plan with next actions. Your output should include a brief explanation tailored for executives that avoids jargon, plus a deeper explanation for the customer’s engineering team. Expected solution outline: compute before/after deltas for key metrics; derive impact signals such as token growth, retry amplification, cache degradation, and endpoint latency regression; score and rank candidate causes; map top causes to remediation actions like prompt trimming, batching, caching, timeout tuning, or model mix changes; return two explanation layers—executive and engineering—demonstrating role-play readiness and the ability to translate technical findings into business terms.