Excel weekly dashboards at scale

PURPOSE

Designs refreshable Excel dashboards (Power Query + structured tables + validation + pivot reporting).

WHEN TO USE

• TRIGGERS:
  • Build me a Power Query pipeline for this file so it refreshes weekly with no manual steps.
  • Turn this into a structured table with validation lists and clean data entry rules.
  • Create a pivot-driven weekly dashboard with slicers for year and ISO week.
  • Fix this Excel model so refresh does not break when new columns appear.
  • Design a reusable KPI pack that updates from a folder of CSVs.
• DO NOT USE WHEN…
  • You need advanced forecasting/valuation modeling (this skill is for repeatable reporting pipelines).
  • You need a BI tool build (Power BI/Tableau) rather than Excel.
  • You need web scraping as the primary ingestion method.

INPUTS

• REQUIRED:
  • Source data file(s): CSV, XLSX, DOCX-exported tables, or PDF-exported tables (provided by user).
  • Definition of ‘week’ (ISO week preferred) and the KPI fields required.
• OPTIONAL:
  • Data dictionary / column definitions.
  • Known “bad data” patterns to validate (e.g., blank PayNumber, invalid dates).
  • Existing workbook to refactor.
• EXAMPLES:
  • Folder of weekly CSV exports: exports/2026-W02/*.csv
  • Single XLSX dump with changing columns month to month

OUTPUTS

• If asked for plan only (default): a step-by-step build plan + Power Query steps + sheet layout + validation rules.
• If explicitly asked to generate artifacts:
  • workbook_spec.md (workbook structure and named tables)
  • power_query_steps.pq (M code template)
  • refresh-checklist.md (from assets/) Success = refresh works after adding a new week’s files without manual edits, and validation catches bad rows.

WORKFLOW

1. Identify source type(s) (CSV/XLSX/DOCX/PDF-export) and the stable business keys (e.g., PayNumber).
2. Define the canonical table schema:
   • required columns, types, allowed values, and “unknown” handling.
3. Design ingestion with Power Query:
   • Prefer Folder ingest + combine, with defensive “missing column” handling.
   • Normalize column names (trim, case, collapse spaces).
4. Design cleansing & validation:
   • Create a Data_Staging query (raw-normalized) and Data_Clean query (validated).
   • Add validation columns (e.g., IsValidPayNumber, IsValidDate, IssueReason).
5. Build reporting layer:
   • Pivot table(s) off Data_Clean
   • Slicers: Year, ISOWeek; plus operational dimensions
6. Add a “Refresh Status” sheet:
   • last refresh timestamp, row counts, query error flags, latest week present
7. STOP AND ASK THE USER if:
   • required KPIs/columns are unspecified,
   • the source files don’t include any stable key,
   • week definition/timezone rules are unclear,
   • PDF/DOCX tables are not reliably extractable without a provided export.

OUTPUT FORMAT

When producing a plan, use this template:

WORKBOOK PLAN
- Sheets:
  - Data_Staging (query output)
  - Data_Clean (query output + validation flags)
  - Dashboard (pivots/charts)
  - Refresh_Status (counts + health checks)
- Canonical Schema:
  - \x3CColumn>: \x3CType> | Required? | Validation
- Power Query:
  - Query 1: Ingest_\x3Cname> (Folder/File)
  - Query 2: Clean_\x3Cname>
  - Key transforms: \x3Cbullets>
- Validation rules:
  - \x3Crule> -> \x3Caction>
- Pivot design:
  - Rows/Columns/Values
  - Slicers

If asked for artifacts, also output:

• assets/power-query-folder-ingest-template.pq (adapted)
• assets/refresh-checklist.md

SAFETY & EDGE CASES

• Read-only by default: provide a plan + snippets unless the user explicitly requests file generation.
• Never delete or overwrite user files; propose new filenames for outputs.
• Prefer “no silent failure”: include row-count checks and visible error flags.
• For PDF/DOCX sources, require user-provided exported tables (CSV/XLSX) or clearly mark extraction risk.

EXAMPLES

• Input: “Folder of weekly CSVs with PayNumber/Name/Date.”
  Output: Folder-ingest PQ template + schema + Refresh Status checks + pivot dashboard plan.

• Input: “Refresh breaks when new columns appear.”
  Output: Defensive missing-column logic + column normalization + typed schema plan.