The lesson does not ask AI to calculate payroll immediately. It first creates a long-lived project structure: source payroll files remain untouched; mappings, cleaning logs, exceptions, and the standardized database live in an intermediate layer; management reports live in a separate output folder.
The six steps are workspace setup, source inspection, field mapping, cleaning and normalization, analysis, and audit. Uncertain fields, blank amounts, duplicates, and mismatched totals must be recorded rather than guessed.
Six-step workflow
Create README and rule files, a payroll field guide, a job-title dictionary, intermediate processing, and output folders. Inspect departments, file counts, fields, naming differences, and possible issues before producing final statistics.
Map inconsistent field names to a standard model, build a new standardized database without editing originals, then analyze roles, pay components, and department labour cost. Finally reconcile all totals, spot-check ten employees, and confirm that department totals equal the sum of role totals.
Create a maintainable payroll-analysis workspace while keeping the original payroll folder read-only. Create a README, analysis rules, payroll-field guide, job-title dictionary, intermediate-processing folder, and output-report folder. Stop after setup. Next, inspect files and field differences; create a mapping; normalize data into a new database; log every missing, duplicate, or inconsistent value; produce analysis reports; and audit total reconciliation, ten sampled employees, and department-versus-role totals.