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, and welcome back. We now have a tidy master sales table. But management doesn't read two thousand rows of data; they read reports. In this episode we ask Claude Code to build an automatic monthly report: a summary of numbers, tables, a chart, and short notes you can trust.
Step 1: Design the report content first
A common mistake is asking for a report right away without knowing who will read it. Dewi asks the director first. The first answer: total sales and how they compare with last month. Second, a branch ranking from highest to lowest, so it's clear which branches need help and which can serve as examples. Third, the top five products by rupiah value, not just by number of items. Fourth, one bar chart of sales per branch. One clear chart is more useful than five busy ones. Fifth, follow-up notes of three sentences. The condition is strict: every number in the notes must come from the data, not from estimates. Dewi saves those five requirements in a text file, then asks Claude to build a report script that compares September with August. Claude reads the director's notes first. Keeping requirements in a file makes the request reusable every month. Then it reads the master table: two thousand three hundred twenty-one rows, six columns. Remember, the content of files Claude reads is sent along for processing. This sales data is safe because it has no customer personal data; sensitive data should be masked first. Before writing anything, Claude shares its plan: one Excel file with three sheets and a chart. The follow-up notes are written from the script's numbers. Dewi approves and adds one small request: rupiah formatted with thousands separators, so it's easy for the director to read.
Step 2: Claude writes the report script
Claude writes a report script of about one hundred eighteen lines. It holds the calculations Dewi usually does with pivot tables and formulas for hours. We don't need to understand every line; what matters is that the result can be checked. A permission box appears. Read the command: the script reads September and August data, then creates one new report file. No old files are overwritten, so Dewi chooses Yes. The result comes out in a few seconds. Total sales of one point two eight billion rupiah, up six point two percent from August. In the past, this number was ready only after half a day of formulas and pivot tables. Claude sums up the main findings. The highest branch is Bandung, the lowest is the newly opened Cirebon, and the best-selling product is A4 paper. Dewi opens the Branches sheet. Bandung ranks first, up eight percent from August. Bekasi is second, up slightly by two point six percent. This change column is calculated automatically from the August data, so Dewi doesn't need to open two files side by side. Bogor is third but down four point three percent. Numbers like this are the ones worth discussing in a meeting. The middle rows are shortened on screen, but the full file has fifteen branches. Cirebon ranks last. The script doesn't calculate a percentage because there's no August data yet, and labels it as a new branch. That's more honest than showing a misleading number.
Step 3: Charts and a one-page summary
It turns out the director often reads reports on a phone, not a laptop. Opening an Excel file on a phone is a hassle. Dewi asks for a one-page version in PDF format. Claude explains one important thing: the PDF uses the same numbers as the Excel file, calculated by the same script. That way the two documents can never show different numbers. To make charts, Claude needs one more library. The permission box is clear: install a charting library. Dewi approves. Claude adds the chart and PDF parts to the existing script instead of creating a new one. The fewer scripts you have, the easier they are to maintain. One run now produces two files: the full Excel report and a one-page PDF summary. Here are the four numbers at the top of the summary. First, total sales of one point two eight billion rupiah. Second, an increase of six point two percent compared with August. The director sees the direction right away. Third, the top branch: Bandung. This branch name is taken straight from the ranking in the Branches sheet, not typed by hand. Fourth, the best-selling product: A4 paper. Below these four numbers are a bar chart per branch and the follow-up notes. It all fits on one phone screen.
Step 4: Report notes you can trust
Now the riskiest part: the notes written as sentences. Dewi asks for three follow-up sentences, and every number must come with its source. The first sentence mentions the sales increase, complete with where the number sits in the Summary sheet. The second sentence explains Bogor's drop and its main cause: A4 paper fell by one hundred eighty reams. The third sentence is about Cirebon, the new branch, in its first month. These three sentences are short, but every number can be traced to a specific cell. Dewi doesn't take it on faith. Dewi opens Excel and works out the Bogor number by hand. It matches. Checking one or two numbers like this takes only two minutes, but it protects your reputation. Claude then builds the notes into the script, so every month the sentences are written from the latest calculations. The director has a taste in wording: the word decreased instead of fell, and billions rounded to two decimals. Dewi asks for the template to be changed. Look at the diff. Red lines are the old text, green lines their replacements. Fell becomes decreased, and the total is written in billions with two decimals. Because the template is what changed, this style applies to every month after this. The report is rebuilt in the new style. There's no need to edit sentences one by one every month. That's the benefit of automation: you fix something once, and it keeps working.
Step 5: One command at the start of every month
Let's look at the full chain, from the previous episode until today. It all starts with branch files arriving in a new month's folder. The merge script standardizes the columns from every branch. The total check compares the numbers. If there's a difference, the chain stops here and doesn't produce a wrong report. The report script builds the tables, chart, and notes. The final step still belongs to a person: Dewi checks and sends. Automation prepares; people decide. To keep things practical, Dewi asks for the three scripts to be combined into one main script that stops and explains if any step fails. Claude writes a main script that runs the three steps in order. One run: merge, reconcile the totals, build the report. It's all done in fourteen seconds. Claude adds a reminder that the report isn't sent automatically. That's on purpose. The screens in this video are simulations, so your Claude Code may look slightly different, but the principle is the same: you are the final checker.
Done!
We built an automatic monthly report: designed around the reader's needs, calculated by a script, with notes sourced from the data, and still checked by a person before it's sent. In the next episode we turn to CSV data from different sources, such as Sunset Coffee's marketplace and order form, and convert and merge it. See you there!
Practice exercises
Try these on your own after watching to lock in what you learned.
Write down the five numbers your report's readers need most
Ask Claude Code to build a report script from one of your data tables, complete with one chart
Check three numbers in the report notes by recalculating them by hand in Excel
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
52 video classes1,276 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.
