OSEMN: Obtain, Scrub, Explore, Model, iNterpret

Think about cooking for a kenduri. You buy the ingredients at the pasar. You wash the vegetables and throw out the bruised ones. You taste the sambal as it cooks. You cook the dishes. Then you plate them and serve. Skip the washing and the whole meal suffers, however good the recipe.

Working with data is the same. Many weak reports start with good intentions and dirty data: a name left in, the same activity counted twice, one state spelled three ways. Then a confident chart sits on top of it, and nobody can tell it is wrong.

OSEMN (say “awesome”) is a five-step path that stops this. Hilary Mason and Chris Wiggins described it in 2010. The letters stand for Obtain, Scrub, Explore, Model, iNterpret. Its job is simple: turn raw records into information you can trust.

✦

1

The idea in 30 seconds

Obtain: collect with permission

Know every source and that you may use it.

Watch the 30-second explainer

The idea in 30 seconds0:00 / 0:31

1. Obtain: collect with permission Know every source and that you may use it.
2. Scrub: clean before you count Remove personal data and duplicates. Fix spellings.
3. Explore: look around AI suggests questions and charts.
4. Model: find a simple pattern A grouping or a trend. No advanced maths.
5. iNterpret: you say what it means Check it, then let AI draft the summary.
6. Clean data in, trusted information out.

Key takeaways

If you remember nothing else, remember these.

• Obtain: know where each record came from, and that you are allowed to use it.
• Scrub: remove names and MyKad numbers yourself, then fix duplicates and spellings.
• Explore: let AI suggest questions and charts, then check the numbers.
• Model: find a simple grouping or trend. Recount one part by hand.
• iNterpret: you decide what it means. AI only drafts the words.

Cook it before you serve it

The five stages run in order, like the kenduri. Each one has a job AI can help with and a check that stays with you. Tap each stage below to see both.

The order matters. If you explore before you scrub, you will find “patterns” that are really typing mistakes. If you interpret before you model, you will see what you hoped to see.

The example: six months of KRT activity records

Everything below is fictional. Puan B is an officer at the made-up Pejabat Perpaduan Daerah Seri Contoh. Her head of department asks a plain question: “What kinds of KRT activities did we run in the last six months, and is anything changing?”

She has a spreadsheet of activity records sent in by several KRT, including KRT Taman Seri Contoh and KRT Kampung Contoh. Some rows were typed by hand, some were copied from forms. Let’s take it through OSEMN.

O for Obtain: know where it came from

Obtain means gathering the data and knowing two things about every piece: where it came from, and whether you are allowed to use it for this purpose. If you cannot answer both, leave that piece out.

You are a data assistant for a district unity office.

My head of department asked: "What kinds of KRT activities did we run in the last six months, and is anything changing?" I have activity records from several KRT in a spreadsheet.

List the fields I need to answer this question, and the fields I do not need. Then draft a short, polite note to KRT secretaries asking for any missing activity records, without names or MyKad numbers.

Mark any field you think is useful but that may hold personal data.

Check it: for each source, write one line: where it came from and who allowed you to use it. Follow your department’s own data and security rules. If a source is not cleared, it does not go in.

S for Scrub: clean before you count

Scrub is the washing step. It is the least exciting and the most important. Here is a small slice of Puan B’s spreadsheet before and after. Tap between them.

Dirty dataScrubbed

ID, State, KRT, Activity, Date, Participants
R01, Selangor, KRT Taman Seri Contoh, Gotong-royong, 6 Jul, 40
R02, Selangor, KRT Taman Seri Contoh, Ceramah, 13 Jul, 25
R03, Pulau Pinang, KRT Kampung Contoh, Gotong-royong, 20 Jul, (blank: ask the KRT)
R04, Pulau Pinang, KRT Kampung Contoh, Sukan, 27 Jul, 60
R05, Pulau Pinang, KRT Kampung Contoh, Ceramah, 3 Aug, 30

What you typically get back

Two states, five activities, each counted once, and no names or MyKad numbers. The blank is flagged for a person to follow up, not filled with a guess.

All names, numbers and places in the table are made up. The MyKad numbers are deliberately impossible.

Scrubbing has two parts, and the order matters.

1. You remove personal data first, in your own spreadsheet. Delete the name and MyKad columns, and anything else that identifies a person, before anything goes near AI. Give each row a plain ID instead.
2. Then AI can help with the rest: spotting duplicates, odd spellings and blank cells.

You are a careful data checker.

Below is a table of KRT activity records. Personal data has already been removed. Each row has an ID.
<table>
(paste the table here)
</table>

Do not change the table. Instead, give me three lists: (1) rows that look like duplicates, with their IDs; (2) every spelling of each state, with one standard spelling to use; (3) blank cells, with row ID and column.

Only list what is in the table. If you are unsure whether two rows are duplicates, say so.

Check it: count the rows before and after. If you started with 120 rows and removed 6 duplicates, you should have 114. Open three of the “duplicates” yourself: two gotong-royong events on the same day in the same KRT might be real.

E for Explore: let AI suggest questions

Now the data is clean, you look around. Explore is about asking good questions before you look for answers. AI is useful here, because it can quickly suggest angles you had not thought of.

You are a data analyst helping a district unity officer.

Here is my cleaned table of KRT activity records for six months. Columns: ID, State, KRT, Activity, Date, Participants. My head of department asked what kinds of activities we ran and whether anything is changing.

Suggest five questions this table can answer, and for each one, a simple chart (bar, line or table) and the columns it uses.

Mark any question the table cannot fully answer, and say what data would be missing.

Check it: if you make a chart, pick two bars and check their numbers against the table. A chart that looks right can still be built on the wrong column.

M for Model: find a simple pattern

In data work, a modelIn OSEMN, a simple pattern you find in the data, such as a grouping or a trend. It does not need advanced maths. sounds technical. Here it means something simple: a grouping or a trend that answers the question. Count activities by type. Count them by month. See whether a line goes up, down or stays flat. Tap each part of the prompt below.

Dissect a prompt

Model: group and show the trend

ExploreTest yourself

Tap a coloured part of the prompt to see what it does.

You are a data analyst who explains things simply. Here is my cleaned table of KRT activities for six months. Make two tables: the number of activities by type, and the number of activities by month. Then describe the monthly trend in one sentence. Show how you counted each total, and say if any month has too few records to compare fairly.

Check it: pick one group, for example “Ceramah in July”, and count it yourself with a filter in your spreadsheet. If your count and the AI’s count differ, trust neither until you know why.

N for iNterpret: you say what it means

The last stage turns numbers into meaning. This is the officer’s job. Only you know that July had fewer activities because of the school holidays, or that a new KRT joined in September.

Write your own notes first: three to five bullet points on what the tables mean. Then let AI turn your notes into a clean summary.

You are a writer for a district unity office. You write plain, official Bahasa Melayu.

Here are my two tables and my own notes on what they mean.

Draft a summary of no more than 150 words for my head of department. Use only the numbers in my tables and the meaning in my notes.

Mark with [check] any sentence that goes beyond my notes. Do not suggest causes I did not write.

Check it: trace every number in the summary back to a table. Then read it once more for claims about why. A trend shows that something changed. It does not show what caused it.

Common mistakes

• Pasting before scrubbing. Personal data must come out in your own spreadsheet first.
• Letting AI “fix” the data silently. Ask for a list of proposed changes, then make them yourself.
• Reading causes into trends. “Activities fell in July” is a finding. “Because residents lost interest” is a guess.
• Skipping Explore. Jumping straight to a chart answers the first question you thought of, not the best one.

OSEMN often feeds the Observe stage of the OODA loop. The next lesson, DRAG, sorts all of this work into four kinds you can hand to AI.

Check your understanding

Check yourself

OSEMN in five questions

1/5

Pick an answer, then tap Check answer. You can also use the arrow keys to move between answers.

Question 1 of 5: What do the letters OSEMN stand for?

AObtain, Scrub, Explore, Model, iNterpretBOrganise, Sort, Edit, Measure, NotifyCObserve, Summarise, Estimate, Map, Note

Check answer

Nice work!

Lesson complete

You made it to the end. Take the quiz above if you have not yet, for bonus XP.

• 0 XP
• 0 day streak