At most one email a week. Unfollow anytime with one click.
Transcript
The full narration of this video. The current sentence is highlighted while the video plays; click any sentence to jump there.
Show transcript
Introduction
Hi, welcome back to the data analysis session. Sunset Coffee's October order data is now clean. The question now is: what are those numbers telling us? We'll learn to tell numbers from insights, write the right questions, have Claude do the math, sanity-check the findings, and turn them into actions.
Step 1: Numbers, insights, and actions
Many people think data analysis means producing lots of numbers and tables. But numbers alone don't change anything. What we're looking for is insight: a new understanding from the numbers that makes us act differently. For example, not just knowing total sales, but knowing who contributes the most. Here's an example. The number: October sales of thirty-eight point four six million rupiah from four hundred three orders. That's a fact, but it doesn't give direction yet. The insight: it turns out customers who reorder bring in most of the revenue. That's a new understanding. The action: Laras creates a loyalty card program for regular customers. A good insight can always be carried forward into an action like this. So every time you see a number, ask two things: what does it mean, and so what should I do? Insights can be found in any kind of work. For Laras, for example: weekend buyers prefer large sizes. That means liter-size coffee stock needs topping up every Friday. For Dewi: the three biggest customers place half of Maju Bersama Co.'s orders. That means the relationship with these three customers needs special care. For Mr. Hendra: grades drop for students who are often absent on Mondays. That means students who fall behind need a summary of the material. For Nadia: her client's short videos get three times more comments than photos. That means the content schedule can shift toward video. All these examples are fictional, but the pattern is the same: number, meaning, then action.
Step 2: The right questions
Claude can calculate anything, so it's the questions that decide the quality of the analysis. There are five kinds of questions that work for any data. First, how much: total revenue, number of orders, and average value per order. Second, which is biggest: the best-selling product by quantity and by revenue. The two can be different. Third, when: the busiest days and weeks. Fourth, who: new customers or customers who order repeatedly. Fifth, what changed compared with last month. This question often brings out the most interesting insights. The pattern is the same for office work. Dewi can ask how many stationery orders there were in total, which items sold best, when the peak was, which customers were biggest, and what changed. Laras puts those questions into a prompt. The context is the cleaned October orders file. The task is four clear questions, not just please analyze this data. The format is a short table per question, then three insights in everyday language. The limits matter: show how you calculated, and keep calculated results separate from guesses about causes. Calculations can be checked; guesses need testing. With these limits, you know which parts are safe to use.
Step 3: Claude does the math
Laras uploads the clean file along with the analysis prompt. Claude runs the analysis on all four hundred three rows, four products, over thirty-one days. The first result is a table by product. Notice: by quantity, the iced coffee cup sells best, with six hundred twenty cups. But by revenue, the one-liter iced coffee is the biggest. Claude also confirms the total matches the cleaning result. This is sample data. If you're unsure about a number, ask Claude to show how it calculated it, for example quantity sold times the official price. That way you can check the logic instead of just trusting the result. Laras asks for the key numbers and three insights in one summary. The right panel shows four number cards: revenue, number of orders, an average value per order of about ninety-five thousand, and the share of revenue from repeat customers, sixty-one percent. Below them are three insights in everyday language. Notice the third insight: the jump in hampers at the end of the month is written as a guess. Claude keeps calculated results apart from guesses about causes, as the prompt's limits asked.
Step 4: Digging deeper and sanity-checking
The first insight triggers a new question. Laras asks: which products do repeat customers usually buy? Claude runs the analysis again, this time by customer code. This is why the customer codes were useful: the analysis still works without real names. The result: sixty-four of one hundred sixty-eight customers ordered more than once, and their favorite is the one-liter iced coffee. New customers mostly buy cups. Laras is then tempted to conclude: does the one-liter product make people loyal? Claude answers honestly: not necessarily. It could be the other way around, and people who are already loyal are the ones willing to buy the large size. To find the cause, you need an experiment. That conversation teaches an important rule. First, a pattern isn't necessarily cause and effect. Two things that move together don't always affect each other. Second, a small amount of data can be a coincidence. One month isn't enough to call something a trend. Compare several months if you have them. Third, there are outside factors: weather, school holidays, payday, or a competitor's promo. Ask Claude which outside factors might play a role, then judge for yourself. Fourth, check one number by hand. Laras calculates the one-liter iced coffee: one hundred fifty times eighty-five thousand makes twelve point seven five million. It matches Claude's table. These four checks don't need statistical expertise. You just need a careful attitude: before believing a pattern, ask whether there's another explanation.
Step 5: From insight to action
Now we turn the insight into action. Start with the insight that has the most impact: loyal customers like the one-liter coffee. Design a small experiment. Laras gives a discount voucher for the one-liter coffee to every first-time cup buyer, for two weeks. Set the measure of success from the start: what percentage of voucher recipients order again within two weeks. Then review it the same way with the November data. Because the standard rules and analysis questions are already saved, the comparison will be fair. Laras asks Claude to help design the experiment: what discount is safe, and how to record the results. Claude suggests a ten-thousand-rupiah discount for fourteen days, with the voucher code recorded in a new column of the orders file. Notice that Claude asks back about the cost per liter, because only Laras knows her business's margin. As a guide, here are the marks of a good insight. First, it's backed by numbers you can check. Second, it can be written in one simple sentence that both Dimas and customers would understand. Third, it can be carried forward into action. Fourth, it's clear what's fact and what's a guess. And what isn't an insight: a sentence that just repeats a number with no meaning, such as October revenue was thirty-eight million. That's a number, not an insight yet.
Done!
We've told numbers, insights, and actions apart, written five analysis questions, had Claude calculate and dig deeper, sanity-checked the findings, and designed an experiment. Insights are easier to understand when they're drawn. In the next episode we make charts that can be read in five seconds. See you there!
Practice exercises
Try these on your own after watching to lock in what you learned.
Write five analysis questions for your data: how much, which, when, who, and what changed
Ask Claude to answer them and show how it calculated
Pick one insight, then design a small two-week experiment to test it
Ask the class material
Type a question or keyword; we find the part of this episode that explains it and jump the video there.
My notes
Notes are saved with the current video time and are only visible to you.
Questions & answers
Stuck on a step? Ask here; the instructor and other learners can answer.
No questions yet.
Your instructor
Cerdas Inovasi Digital Team
Software developers, BIM & 3D engineers, and AI practitioners
53 video classes1,290 episodes
We build software, BIM models, games and AI automations for clients every day, and turn that hands-on work into step-by-step classes: every step is shown on screen and comes with a PDF guide you can keep.
