Messy data is usually a process problem before it is a technical problem. Start by listing the fields that matter to the decision and the rules each field should follow.
Standardize names, dates, categories, and units before trying to analyze the data. Keep the original values in a raw layer so that every correction can be traced and reviewed.
Then measure what changed. A short data-quality report showing missingness, duplicates, and invalid values gives stakeholders a shared view of progress and helps prevent the same issues from returning.