CODED × KJO · Day 2

×

From Data
to Insight

Day two of three. Two rungs of the ladder. A number you can stand behind, and something to put beside it.

Six hours · thirteen blocks

Arabella Hotel

Bring your change log

Press → to begin

Yesterday, in three things

You finished holding clean data.

1

Two dashboards from one file, where the only difference was the prompt.

2

That board, where fifteen prompts produced four different answers to one question.

3

The thing in front of you now: a cleaned Work Order Set, with a change log beside it.

That is the bottom rung, and you earned it.

Nothing else on this slide

Industrial Branch handled 257 service requests this year.

Is that good or bad?

One number, out of a clean file, with nothing beside it.

Name the felt problem

A number on its own is not information.

Yesterday you spent an afternoon making a file you can trust. That was real work and it was worth it. This morning I gave you one number out of a clean file, and not one of you could tell me whether it was good news.

Everything you just asked for is the same move: put something beside it.

What you’ll walk out able to do

By the end of today, here’s what changes.

Point a tool at a file instead of pasting the file into it, and know why that changes the answer.

Ask can this file answer that question before you open anything.

Put something beside a number, and name which of the four comparisons you chose.

Say what a finding means, and what would have to be true for it to be wrong.

The shape of the gap

Four rungs. We climb two today.

A label for the two days, not a theory. Name it, hang the two days on it, move.

Where the week sits on it

Today is the middle two.

AM

Information. Getting a number you can actually stand behind, out of 2,400 rows rather than 40.

PM

Knowledge. Putting something beside it.

Which rung was yesterday’s answer board on?

Topic 03 · Story

When you paste more than fits, the earlier part falls out of view.

Yesterday I told you a model holds a limited amount of text in view at once. That limit has a name: the context window.

The tool does not stop. It answers using what it can still see.

Demo · same file, same question, two routes in

How many rows are in this file?

Pasted into the chat

···

File attached, grounded

···

The term & the link

This is grounding.

GROUNDING You point the tool at the file, instead of pasting the file into the tool.

1 It sees every row.

2 It knows which column holds a date and which holds text.

3 It can count.

Yesterday: it predicts, it does not look up and it does not calculate…

→

grounding is where that stops being true.

A grounded tool runs real operations on a real table.

Three routes, and a license decision made above this room

Which tool, and when.

	Copilot Chat Track B	Copilot in Excel Track A	Claude
Where it sits	Your account and its security boundary	Inside the workbook	A separate browser session
Best at	Files and Microsoft tools	Building what you saw at nine o’clock yesterday	Long reasoning, cleaning logic, explanations
Long pasted data	Weaker	Reads the file, not a paste	Handles more, more reliably
Needs a license	No	Yes, and not everyone has it	Depends on your account

Block 03 · your turn

Ground the full Work Order Set.

The file is work-orders-full.xlsx. Attach it, do not paste it. For most of you this is the first time you have attached a file rather than pasted one.

Workbook · page 2

Ask it: how many rows are in this file?
Ask it: which rows did you exclude?
Write both answers on page 2.

15:00

Timing flexes to the room. The clock is a guide, not a gate.

Close the block

Grounding does not replace part 5. It makes part 5 answerable.

Yesterday your prompt had a part that asked the tool to state its row count. Which part number?

Yesterday the tool could only estimate its row count. Today it can report it.

Topic 04 · Story

Yesterday you asked what is in this file.

Six profile questions, answered once, about the dataset as a whole. This is a different question, and it changes from question to question.

Can this file answer the specific question I am about to ask?

The check, and it is two lines long

Two questions, ninety seconds.

1 Which columns does this question need? Name every one of them.

2 How many rows have all of them filled in? Not the file’s row count. The count for these columns.

Demo · one row at a time

Run the check on four questions.

The questionColumns it needsRows with all of them

Two lines per question. Which columns, and how many rows have all of them.

Say it slowly

Clean is not the same as complete.

CloseDate is filled in eighty-six percent of rows. Every other column in this file is one hundred percent.

,

Nothing about this file looks broken.Nothing you did to it yesterday was wrong. You cleaned it properly.

,

Clean means what is in it is right.It does not mean everything you need is there. And no amount of cleaning fixes a column that was never filled in.

Demo · their questions, not yours

What do you actually want to know about maintenance at your own site?

Take eight or ten. If you cannot name the columns, the question is not ready, and that is a finding, not a failure.

Block 05 · your turn

Run the check on your own questions.

Two rules, and they hold all day: the check goes in your workbook before you open the file, and every answer arrives with the row count the tool reports.

Workbook · page 3

Run the two-line check on your questions, in the workbook, before you open anything.
Then ground the file and answer the ones that clear.
Row count on every answer.

15:00

Timing flexes to the room. The clock is a guide, not a gate.

Fast finishers: find a question this file can answer that nobody has asked yet. Then find one it cannot.

Unlock · confirm every answer out loud

Every question that cleared the check came back right.

0

work orders, May 2024 to June 2025

0

Ahmadi, busiest site. Burgan 564, Sabriyah 400

0

Ahmadi unplanned downtime hours, highest

Ninety seconds each. That is what the check costs, written down, before you open the file.

Topic 05 · The manager has a fifth question

Average time to close a work order, by site.

Block 06 · your turn

Run the manager’s fifth question.

Four minutes. Then three of you read your answer out.

Workbook · page 3b

Average time to close a work order, by site.
Write what the tool gives you on page 3b.

4:00

Timing flexes to the room. The clock is a guide, not a gate.

Three answers, and they agree

What is the finding here?

Site	Average days to close
Ahmadi Refinery	3.17
Burgan Gathering Center	3.30
Sabriyah Station	3.38
Mina Al-Zour Terminal	3.42
Wafra Station	3.42
North Field Booster	3.70

Two prompts, in this order. They are the whole block

Which columns does that question need?

Site · RaisedDate · CloseDate.

How many rows have all three?

2,079 of 2,419. It has been on screen since half past ten.

The line the block exists for

You knew. And on the fifth question, in a rhythm, not one of you ran it.

You have had that check in your hand since half past ten. It takes ninety seconds. Every one of you agreed it was obvious.

That is not a knowledge problem. A better model does not fix this one.

The reveal · time to close needs a close date

340 work orders do not have one.

You found those blanks yesterday, on page 7. You wrote the fill rate on your check card an hour ago. You knew twice.

Name it, after five seconds of silence

Confidence is not evidence.

Ask for the row count on every question, not on the hard ones, because you cannot tell in advance which one is the hard one.

Nothing went wrong with the tool. Every number in that first table is arithmetically correct. It answered the question it was asked, using the rows it could use, and said nothing about the rest.

This was not a hallucination. It was a silence.

Block 06 · your turn

Rewrite it so a wrong answer becomes visible.

Same tool. Same file. Same question. The job is to make the trap show itself before it catches you.

Workbook · page 4

Rewrite the fifth question on page 4.
Two of you will read yours out.

4:00

Timing flexes to the room. The clock is a guide, not a gate.

The working version, run it live

Three extra lines, and the trap is visible.

What is the average time to close a work order, by site?
Time to close means the days between RaisedDate and CloseDate.

Before you answer:
- report how many work orders have no CloseDate
- report that count by site, as a percentage of each site's rows

Then give me two tables:
Table A: the average using only work orders that have a CloseDate.
Table B: the average counting work orders with no CloseDate as still open,
measured from RaisedDate to the latest date in the file.

Tell me which sites change rank between the two tables.

Same tool. Same file. Same question. That is yesterday’s line again. The prompt is the product.

Pause

Three questions before we change rung.

That number is now trustworthy. It is still not useful, and the next twenty-five minutes are about why.

What is still unclear?

Where else do you skip a check under time pressure?

What would you push back on?

Topic 06 · Story · their own list, three hours later

You did not need me for that.

This morning I gave you one number and asked whether it was good. Here is what you asked me for, in your own words, at ten past nine.

You already knew what was missing. What you have not had is a name for it, and a habit around it.

The term

A number becomes information when you put something beside it.

1

The same thing last periodIs this getting better or worse?A short period hides a trend. A long one hides a turn.

2

The targetAre we where we said we would be?If nobody set a target, say so. Do not invent one.

3

Another groupWho is doing this better, and what do they do differently?The groups have to be alike enough to compare.

4

A benchmark outsideIs this normal for work like ours?Where did it come from, and who paid for it.

And a fifth, always available

Compare against nothing, and say so.

Defensible

“257 requests, and I have no baseline yet.”

Not defensible

“257 requests, which is strong.”

One of those you can defend in a meeting.

,

The comparison you choose is a claim about what matters.Against last year says time matters. Against the branch next door says fairness across branches matters. Both are legitimate, and they lead to different decisions.

,

Comparing like with unlike is the most common failure.Industrial handles 257 and Central handles 365. Central is bigger. Raw counts say nothing until you decide whether you are comparing volume or rate.

Mini check · your call

Yesterday’s answer was Ahmadi Refinery, 49.4 days. Which comparison would you put beside it?

Practice · 45 minutes · the longest of the day

One number. Four neighbors. Four different findings, and every one true.

Worked example

Ahmadi Refinery, 49.4 days to close.

Last periodBe careful here. Work orders raised recently have had less time to age, so a naive split makes the recent half look faster at every site.

TargetNo target exists, and that is worth writing down.

Another groupTwenty days worse than the next slowest site.

BenchmarkThe maintenance standard most plants quote is 30 days, illustrative; say where it came from or mark it unknown.

Block 09 · your turn

One number. Four versions of it.

Take any number from the Work Order Set that you can defend. Then produce four versions, one for each comparison. All four will be true.

Workbook · page 5

Write all four down on page 5.
Then put your best one on the Comparison card, with its row count.
Stuck? Which of the four have you not tried yet?

25:00

Timing flexes to the room. The clock is a guide, not a gate.

Fast finishers: find a comparison that reverses your finding. If you cannot, say why not.

Unlock · one comparison from each team

Four comparisons, one file. Which would change a decision?

1

The comparison restates the number.“Downtime was 2,472 hours, which is a lot of hours.”, Push: compared with what?

2

The comparison is true and useless.“Ahmadi is worse than North Field Booster.” North Field is a third the size. Push: are those two alike enough to compare?

Choosing the comparison is choosing what matters. That is not a technical decision, and the tool will not make it for you.

Topic 07 · Story · count the hands out loud

Yesterday you wrote yes or no about one row.

The work order costing forty times the next highest. Hands up who wrote: yes, delete it.

The reveal

0KWD

WO-2087 Ahmadi Refinery · 72 hours · a three-day compressor failure

The most expensive single event in fourteen months of data. Everybody who deleted it produced a cleaner chart and removed the most important row in the file.

If the room is sharp: WO-2087 is also one of the twenty duplicate IDs, and the two rows carrying it are different work orders. Removing duplicates by ID could have deleted it without anybody choosing to.

The term

An outlier is far outside the range. A signal means something.

OUTLIER OR SIGNAL The data cannot tell you which one you are holding. The business can.

1 Is it a data error, or a real event?

2 If it is real, is it a one-off or the first of a pattern?

3 Who in the business would know?

Why the instinct runs the wrong way

The cleaner chart is not the more correct chart. It is the more comfortable one.

Keeping a coding error

Inflates an average for a year, and somebody usually spots it.

Deleting a real event

Quietly breaks a budget twelve months later, and nobody traces it back.

The asymmetry is why the default is keep, and ask.

Topic 08 · Integration · nothing new after the next twenty minutes

Ahmadi Refinery takes 49.4 days to close a work order, twenty days worse than the next slowest site.

That is a true comparison, with a row count behind it. What does it mean?

Meaning is a different question

Not what it is. What changes if it is true.

So what

What does this cost or save, and roughly how much?

Who

Who does something different because of it?

What would break it

What would have to be true for this to be wrong?

Worked example, read it aloud

So whatAhmadi takes 49 days against 28 elsewhere. It has 138 work orders sitting open right now, twice as many as the next site, and forty percent of all the open work in the company.

WhoThe person who does something different is the Ahmadi maintenance planner, not the analyst.

Breaks ifThose blank close dates are a reporting habit rather than open work, which I have not checked.

Notice what that last sentence did. It did not weaken the finding. It told your reader exactly where to push, which is the only way they can trust the rest of it.

Block 12 · your turn

Finish your Comparison card.

Everything today lands on this one object. Grounding, the row count, the comparison, the outlier.

Workbook · page 7

The number, and what it is compared against.
The row count.
What it means, and what would have to be true for it to be wrong.

20:00

Timing flexes to the room. The clock is a guide, not a gate.

Stuck on meaning? Who does something different tomorrow because of this? Fast finishers: write the version your manager would push back on, and answer the pushback.

Close · point at the ladder, then at the card in their hands

Two rungs today.

2

Information. The tool was right four times, then dropped fourteen percent of the file on the fifth without mentioning it. You now write the row count down without being asked.

3

Knowledge. One number, four neighbors, four true findings, and the card in your hand, which says what yours means and what would break it.

Yesterday’s line still holds. The prompt is the product.

The hook into Day 3

Tomorrow is the top rung.

1 Data

2 Information

3 Knowledge

4 Wisdom

Wisdom, which sounds grand and is not. It is one practical question. You have a finding on that card, and it is a good one, tomorrow somebody has to understand it without you standing next to them.

I am going to open tomorrow with a dashboard. Every number on it is correct. And it will not answer the one question the person looking at it actually has.

From Datato Insight

You finished holding clean data.

Industrial Branch handled 257 service requests this year.

A number on its own is not information.

By the end of today, here’s what changes.

Four rungs. We climb two today.

Today is the middle two.

When you paste more than fits, the earlier part falls out of view.

How many rows are in this file?

This is grounding.

Which tool, and when.

Ground the full Work Order Set.

Grounding does not replace part 5. It makes part 5 answerable.

Yesterday you asked what is in this file.

Two questions, ninety seconds.

Run the check on four questions.

Clean is not the same as complete.

What do you actually want to know about maintenance at your own site?

Run the check on your own questions.

Every question that cleared the check came back right.

Average time to close a work order, by site.

Run the manager’s fifth question.

What is the finding here?

You knew. And on the fifth question, in a rhythm, not one of you ran it.

340 work orders do not have one.

Confidence is not evidence.

Rewrite it so a wrong answer becomes visible.

Three extra lines, and the trap is visible.

Three questions before we change rung.

You did not need me for that.

A number becomes information when you put something beside it.

Compare against nothing, and say so.

One number. Four neighbors. Four different findings, and every one true.

One number. Four versions of it.

Four comparisons, one file. Which would change a decision?

Yesterday you wrote yes or no about one row.

An outlier is far outside the range. A signal means something.

The cleaner chart is not the more correct chart. It is the more comfortable one.

Ahmadi Refinery takes 49.4 days to close a work order, twenty days worse than the next slowest site.

Not what it is. What changes if it is true.

Finish your Comparison card.

Two rungs today.

Tomorrow is the top rung.

Presenter & pointer controls

From Data
to Insight