Is your data ready for an AI analyst? A 12-point checklist
By SchemaChat (Data BI LLC) · Published 30 October 2026 · Last reviewed 30 October 2026 · 5 min read
Most wrong answers from AI analytics tools start with the data, not the AI. This checklist gives twelve quick checks for headers, stacked sections, blanks, dates, duplicates and units, with the fix for each. Ten minutes on these saves hours of doubt later.
Whether you use SchemaChat or any other tool, the same data problems produce the same wrong answers. Go through these before you connect or upload anything important.
| # | Check | Why it matters | Fix |
|---|---|---|---|
| 1 | One header row, at the top | A second header halfway down shifts columns for every row after it | Remove it, or split the file into two tables |
| 2 | No pasted lists, notes or totals below the data | Mostly-empty rows inflate missing-value rates and counts | Move them to another sheet |
| 3 | No subtotal or “Total” rows | They double-count when summed | Delete or tag them |
| 4 | One row per record | Several records in one row, or one record over several rows, breaks counts | Reshape before analysis |
| 5 | A single date format | Mixed formats are read as different dates or as text | Convert to ISO (2026-10-30) |
| 6 | Dates stored as dates, not text labels like “Jun-21” | Text labels sort alphabetically and can't be parsed by many engines | Add a proper date column |
| 7 | Blank means blank | “N/A”, “-”, “none” and empty cells are counted differently by different tools | Pick one convention for missing values |
| 8 | Numbers are numbers | A currency symbol, a comma or a stray letter turns a column into text | Strip symbols; keep units in the header |
| 9 | Consistent units and currency | Mixing dollars and euros, or kilos and pounds, in one column | Convert or add a unit column |
| 10 | Consistent spellings and case | “Free”, “FREE” and “free” are three categories to many tools | Standardise case; trim stray spaces |
| 11 | Duplicates understood | Repeated rows inflate totals; a repeat may also be legitimate | Decide what a duplicate means and mark them |
| 12 | IDs are unique and names are descriptive | Non-unique names merge different customers | Keep an ID column and use it for joins |
A five-minute test
- Sort by each date column: do blanks or odd values collect at one end?
- Count rows where most cells are empty.
- Search for your column names in the middle of the data.
- Compare the total of a numeric column with a figure you already trust.
Fix the source, keep the original
Save an untouched copy of the export before you change anything. Fix problems in the system or process that produces the file where you can — a stacked export will be stacked again next week. When you can't, write down the fixes you applied so the next person repeats them the same way.
What SchemaChat does for you, and doesn't
On upload SchemaChat reports the structure it found, re-aligns a repeated header, sets aside a long pasted list, normalises dates, and counts blanks and common placeholders as missing values. It cannot know your business rules (what counts as a duplicate, which currency), so checks 9, 11 and 12 remain yours.