Poor definitions, stale data, duplicates, and weak lineage. Bias and sampling errors skew models. Fix the basics: assign data owners, define business terms, set validation rules, standardize IDs, and implement MDM where needed. Monitor quality with automated tests and scorecards; route issues like tickets.

Start with decisions, not dashboards. Define the questions that move revenue, cost, and risk. Map required data, build governed pipelines, and add a semantic layer so business terms match tables. Deliver role-based views and alerts wired into daily workflows. Close the loop with experiments and action logs.