# Session 1 Handout: Prepare the Data

<p class="meta">FAST-11 AI-Assisted Analytics and Business Intelligence · 7formations · Instructor: Wadood Chaudhary · All examples fictional</p>

## Why this session matters

Every analysis inherits the problems in its data. AI features make cleanup much faster: they spot duplicates, mixed date formats, and inconsistent names in seconds. They will also happily "fix" a blank by guessing or delete a real outlier. This session is about using AI to find problems while you make the fixes yourself, with repeatable, logged steps.

## Key concepts

### 1. Data rules come first

| Never in class tools | OK in class |
| --- | --- |
| Customer personal information (name plus SSN, license, account or card number; G.L. c. 93H) | Fictional course datasets |
| PHI, even with names removed | De-identified, fictional scheduling files |
| Confidential financials, bids, pricing | Public data (for example, Census tables) |

At work: check the file's sensitivity label and your employer's approved tools before loading anything.

### 2. Cleaning with AI help

- Ask the AI to **find** problems: duplicates, mixed date formats, name variants, inconsistent capitalization, blanks, outliers.
- Ask for **formulas or query steps**, not a cleaned file.
- **Apply fixes yourself** in Excel, Power Query, or Tableau Prep.
- **Never guess** a blank or delete an outlier without checking the source.
- Spot-check rows before and after.

### 3. Document every transformation

| Step | What changed | Rows | Reason | Method |
| --- | --- | --- | --- | --- |
| 2 | Machine names standardized to VF-4 | 2 | Variants split totals | Replace values |

Power Query records each change as an **Applied Step** that reruns on refresh, which is a log you get for free. Keep your W1 log next to the dataset. **Test:** could a coworker repeat your cleanup from the log alone?

## Worked example (fictional)

Bayside's 20-row scrap excerpt had a duplicate (same event in two date formats), machine variants ("VF4", "vf 4"), a blank quantity, and a quantity of 60 that was a typo for 6. Cleaned total: **$1,365.50**. The raw file would have reported **$2,430.50**, overstating scrap by $1,065. The typo alone added $993.60.

## Checklist

- [ ] Data classified; approved tool confirmed.
- [ ] Raw file kept untouched.
- [ ] AI used to find problems; fixes applied with formulas or steps.
- [ ] Blanks and outliers checked at the source, not guessed.
- [ ] Every step logged with rows affected.
- [ ] Totals compared before and after.

## Do and don't

| Do | Don't |
| --- | --- |
| Ask for steps you can rerun | Accept a mystery "cleaned file" |
| Confirm duplicates by inspection | Delete every repeated part number |
| Flag blanks for the source owner | Fill blanks with averages |
| Log rows removed and values replaced | Clean silently |