CODED × KJO · Participant workbook
Day 1. Working With AI on Real Data
What you leave with today
1A five-part prompt you wrote yourself that you can use on Sunday.
2One cleaned dataset with a written record of every change you made.
3That record, which is the thing that survives this room.
Words used in this workbook
Copilot Chat: the Microsoft chat tool you reach through your browser or Windows.
Copilot in Excel: Copilot working inside the workbook. It needs a separate license.
Claude: the Anthropic chat tool.
Work Order Set: the shared maintenance dataset you use today.
Token: the piece of text a language model reads and writes, roughly a short word or part of one.
Context window: how much text a model can hold in view at one time.
Data profile: a short description of what a dataset contains, written before you analyze it.
Five-part data prompt: the prompt structure on page 4.
Verification move: a specific action that tests whether an answer is correct.
Page 1
Block 1: Two prompts, one file
You watched two dashboards built from the same file, by the same tool, in the same five minutes.
Exercise 1.1.What was different? Write down four things.
1
2
3
4
Prompt A, in full:
Analyze this maintenance data and create a dashboard.
Prompt B is on the prompt sheet in front of you. Keep it. You use it again on pages 3 and 5.
The line that holds these three days together:
The prompt is the product.
The tool was identical. The file was identical. The time was identical.
Your note: which single line in Prompt B do you think did the most work?
Page 2
Block 2: The Site Downtime question
The file:
work-orders-40.xlsx. Forty maintenance work orders.The question: Which site had the worst unplanned downtime last quarter?
Exercise 2.1.Paste the file into Copilot Chat or Claude. Ask the question the way you would normally ask it. You have 12 minutes.
Write the prompt you used.
Write your answer.
| Site | Downtime hours |
|---|---|
Which tool did you use?
Exercise 2.2.After the answer board.
How many different sites appeared on the board?
Answer this before your facilitator speaks again: how would you show that your answer is the right one, without asking the tool again?
Page 3
Block 4: What an AI is, and how it works
The one mechanism
A language model predicts the next piece of text, over and over.
It predicts. It does not look up, and it does not calculate.
Four consequences
1. It is not a search engine. It holds no database you can query. It produces the text that most plausibly follows your question.
2. It is not a calculator, unless it uses one. Asked to add pasted numbers, it predicts what the answer looks like. When it writes and runs code, or works on an Excel table directly, it computes instead.
3. Confidence is a writing style, not a signal. The tone is identical whether the answer is right or wrong. You cannot tell them apart by reading.
4. It only sees what you give it, and only so much at once. No memory between sessions. It had never seen your file until you attached it. The context window limits how much it holds in view, and text beyond it falls away, often with no warning.
Exercise 4.1.During the two demonstrations.
Demonstration 1, the same prompt twice. What differed between the two answers?
Demonstration 2, the pasted total against the formula. Which one was right, and could you tell by looking?
Five things that follow
1Tell it what the work is for. It cannot infer your intention.
2Describe your data. It has never seen it.
3Define your terms. It will define them for you if you do not.
4Make it compute rather than predict, whenever a number matters.
5Ask it to show its working. You cannot judge the answer by reading it.
Page 4
Block 5: The five-part data prompt
| Part | What it does | Example |
|---|---|---|
| 1. Role | Tells the tool what work it is doing | "You are a maintenance data analyst." |
| 2. Data description | Says what the data is and what one row means | "Each row is one maintenance work order. There are 40 rows." |
| 3. Task | Asks the question, with your terms defined | "Unplanned means Emergency or Urgent. Last quarter means April to June 2025." |
| 4. Output shape | Says what you want back | "A table of sites with total downtime hours, highest first." |
| 5. Verification ask | Makes the tool check itself | "State the row count you used. List any rows you excluded and say why." |
Find each part in Prompt B. Write the line that carries it.
| Part | The line in Prompt B |
|---|---|
| 1. Role | |
| 2. Data description | |
| 3. Task | |
| 4. Output shape | |
| 5. Verification ask |
Prompt A had one of these. Prompt B had all five. Nothing else was different.
Three things worth knowing
Part 2 does more work than it looks like. One row here is one work order. Not one asset, not one failure, not one day. Skip part 2 and the tool infers from the column names, reasonably, and you will not see the difference in the answer.
Part 4 is where interactivity lives. The slicers existed because a line asked for them. If you want to click something, say so.
Part 5 is the one you will drop when you are busy. It moves the check inside the answer, so it happens whether or not you remember.
Page 5
Block 6: The question, done properly
Exercise 6.1.Rebuild your prompt with all five parts. Run it on the same 40-row file.
Role:
Data:
Task:
Output:
Verification:
Your new answer.
| Site | Downtime hours | Row count the tool reported |
|---|---|---|
What changed between your first answer and this one? One sentence.
Exercise 6.2.The three pitfalls. Check the ones your new prompt caught.
How many did you catch?
Exercise 6.3.Answer honestly. Nobody reads this but you.
Which of the five parts are you most likely to skip when you are busy?
Part number:Because:
Page 6
Blocks 8 and 9: Why dirty data breaks an answer, and the profile
Four defensible readings of the same forty rows
Every one of these is a reasonable thing to assume. Each one changes the winner.
| The setting | What somebody assumed | The answer it gives |
|---|---|---|
WO-1183 counted twice | "I did not check for duplicates" | Burgan wins, 108.5 |
| The negative row read as +6 | "I assumed the sign was a typing error" | Sabriyah wins, 97.0 |
| March included | "Last quarter meant the last three months" | Sabriyah wins, 112.0 |
| Emergency only | "Unplanned means emergency to me" | Burgan wins, 52.0 |
| All four handled | Ahmadi wins, 96.0, from 24 rows |
Which one is wrong?
None of them. Only one of them is what you meant.
You cannot write part 2 of a prompt about a file you have not looked at.
The six profile questions
A data profile is a short description of what a dataset contains, written before you analyze it.
1How many rows and how many columns?
2What does one row represent?
3What is the date range?
4Which columns have blank values, and how many?
5Are there duplicate records?
6Which text columns have the same thing written more than one way?
Look at Prompt B again. Its whole second paragraph is a profile. That is why the dashboard cleaned the dates before it built anything.
Worked profile prompt
You are a data analyst. I am giving you a maintenance work order file. Each row is one work order. Before any analysis, profile this file. Answer these six questions: 1. How many rows and columns? 2. What does one row represent? 3. What is the date range, and which column holds it? 4. Which columns have blank values, and how many in each? 5. Are there duplicate records? Say whether you matched on ID or on the whole row. 6. Which text columns contain the same value written more than one way? List the variants. Do not analyze anything yet. Report the profile only.
Page 7
Block 11: Profile the full Work Order Set
The file:
work-orders-full.xlsx. About 2,400 rows.Exercise 11.1.Run the six profile questions.
| # | Question | What you found |
|---|---|---|
| 1 | Rows and columns | |
| 2 | What one row represents | |
| 3 | Date range | |
| 4 | Blank values, and where | |
| 5 | Duplicate records | |
| 6 | Text written more than one way |
The blank close dates. Are those work orders still open, or is the data incomplete?
Your answer:
How would you find out? The tool cannot tell you this one.
Page 8
Block 12: Your change log
Exercise 11.2.Clean the file. Record every change here.
In six weeks, nobody will remember why rows are missing. This page will.
| # | What you changed | Why | Rows affected |
|---|---|---|---|
| 1 | |||
| 2 | |||
| 3 | |||
| 4 | |||
| 5 | |||
| 6 |
One decision on this page that you are not certain about. Write it here, and write who you would ask.
Decision:
I would ask:
Exercise 12.1.The cost outlier.
One work order costs about 40 times the next highest.
Do you delete that row?
Why?
After the discussion. Write the question you would ask Operations about that job.
Keep this page. You return to it tomorrow, when you decide whether an outlier is noise or a signal.
Page 9
Day 1 key takeaways
Write these in your own words. The ones you write down are the ones you keep.
1
2
3
What today built
The four things worth carrying to your desk
The prompt is the product. The tool is the same for everybody. The prompt is not.
Use all five parts. Part 5 is the one that gets skipped.
Profile before you analyze. Six questions, about three minutes.
Keep a change log. Your memory will not hold it.
What you are holding at the end of today
A file you can defend, and a record of why.
That has a name. It is called Data, and it is the bottom rung of a ladder with four rungs on it. You climb one tomorrow.
Before tomorrow
You do not bring your own files to this program. Every exercise runs on the practice datasets, including the capstone on Day 3. That is a design decision, not a limitation.
Tomorrow: pasting stops working. And you finish the day holding something you cannot get from a clean file alone.