Introduction

Hello,
I’m Mia Sato, AI Researcher at GDX Co., Ltd.

Recently, I have been supporting the development of a system that automatically generates regional marketing reports for client companies.

As I worked on the project, one issue became especially clear: many companies have data, but they spend an enormous amount of time organizing, cleaning, and aggregating it.

In particular, because the data submitted by each region often comes in slightly different formats, teams frequently have to correct Excel files, redo aggregations, and check data manually each time.

I believe this is a challenge that many companies are facing today.

Ideally, teams should be spending more time on analysis and strategic planning. In reality, however, much of their time is taken up by preparing the data before they can even begin the actual analysis.

Seeing this situation firsthand made me strongly feel that AI and automation could help improve this process.

In this article, I would like to share how GDX approached the introduction of automation, including the thinking behind the project and the steps we took.

If you have ever thought, “Our company may be facing a similar issue,” or “We want to introduce AI, but don’t know where to start,” I hope this article will be helpful.

Before Creating Reports, the Team Had to Fix the Data Every Time

In this client’s case, sales and inventory data were submitted by sales companies around the world. Based on this data, the product planning team and the integrated sales team created weekly reports, monthly reports, and summary reports.

They needed to review sales by region, check store-level trends, compare product-level performance, and analyze changes in e-commerce.

All of these tasks are important for decision-making.

However, when we looked closely at the actual data, we found that the formats submitted by each region were slightly different.

In one region, the sales column was labeled “Sales.”
In another region, it was labeled “Revenue.”
An item that had been in column C the previous month had moved to column D.
Some product codes were submitted blank.
Some cells contained NULL values, causing the process to stop halfway.

These issues were happening almost every week.

As a result, before creating a report, the person in charge first had to open the Excel file and check whether the columns had shifted, whether the required fields were included, whether formulas were broken, and whether VLOOKUP references were working correctly.

Depending on the situation, they also had to copy and paste data, transfer it into another Excel file, or correct formulas manually.

Each task may seem small on its own, but when repeated every week or every month, it becomes a significant burden. The more manual work involved, the higher the risk of errors.

One particularly serious issue was that it became difficult to trace the source of the data.

To compare data across regions or analyze it by store or product, the team had to transfer the data into separate Excel files and re-aggregate it. As this process was repeated, it became harder to identify which original source file the final figures came from.

In some cases, decisions could not be made without asking a small number of experienced employees who understood the data well.

Time that should have been spent analyzing sales factors, comparing regional performance, or examining e-commerce trends was instead being spent fixing Excel files and checking formula errors.

The Real Issue Was Not Excel Work, but Inconsistent Data Definitions

At first glance, it seemed like the problem could be solved simply by reducing Excel work.

However, as we investigated further, we found that the issue was deeper than that.

It was not only that column names differed. The meaning of each data field also varied slightly by region.

For example, even if a field appeared to represent “sales,” its definition could differ. Was the amount tax-inclusive or tax-exclusive? Did it include returns or not? Was it the amount before or after discounts?

Inventory data also changes meaning depending on whether it refers to warehouse inventory, store inventory, or inventory that includes allocated stock.

If AI is used to analyze data while these definitions remain inconsistent, the output may appear to compare regions properly, but in reality, it may be comparing data based on different conditions.

AI is useful, but it cannot automatically and accurately understand every company-specific rule or every assumption that is only implicitly shared among people on the ground.

That is why, before asking AI to process the data, it was necessary to align the meaning of the data among people first.

What We Organized Before Handing the Work Over to AI

In response to this situation, we did not try to hand everything over to AI from the beginning.

The first step was not deciding what AI should do. Instead, it was organizing the data and the business workflow.

We examined questions such as:

Which regions submit which types of data?
What formats are used?
Which fields have the same meaning, and which fields are defined differently?
Where do people open Excel files and make corrections?
Where does manual data transfer occur?
Where are mistakes most likely to happen?

After mapping out this workflow, we separated the areas that could be automated from the areas that still required human review.

In this project, we mainly worked on the following three areas.

1. Making Data with the Same Meaning Usable as the Same Data

The first step was to organize the definitions of each data field.

Terms such as “Sales,” “Revenue,” “Net Sales,” and “Sales Amount” may look similar, but they do not always mean the same thing.

We therefore checked differences such as tax-inclusive versus tax-exclusive amounts, how returns were handled, and whether figures were before or after discounts. Based on this, we defined standard fields to be used by the headquarters team.

We also clarified which fields should be mandatory, what units should be used, which code systems should be applied, and how NULL values or blank fields should be handled.

This process may seem simple and unglamorous, but it is extremely important when using AI. If this foundation remains unclear, even advanced AI tools will not produce outputs that people on the ground can trust.

2. Reducing Manual Weekly Excel Checks as Much as Possible

Next, we worked on automating the Excel checking tasks that had been repeated every week.

Previously, the person in charge manually opened each file and checked column names, column shifts, NULL values, and missing product or store codes.

At GDX, we built a workflow that checks, transforms, and integrates files as soon as they are submitted by each region.

For example, whether a column is labeled “Sales” or “Revenue,” the system can treat it as the same sales field. Even if the column position changes, the system can still read the necessary fields. If a product code is missing, the system can identify which region and which file contains the issue.

This means that the person in charge no longer has to check every file from scratch.

Of course, final confirmation is still performed by people. However, the areas that require human review are significantly narrowed down.

As a result, the time spent on routine weekly work was reduced, making it easier for the team to focus on sales factor analysis and regional comparison.

3. Gradually Improving the Submission Format on the Regional Side

By creating a system on the headquarters side to clean and standardize data, reports submitted in different formats by different regions can be converted into a more common structure.

However, this alone has its limits.

If major conversions and corrections are required every time, operational workload will remain. Therefore, in the long term, it is also necessary to gradually improve the format used by the teams submitting the data.

For example, required fields such as product code, store code, target period, sales amount, sales quantity, and product category should be clearly defined. Fields that are prone to input errors can be changed to selection-based inputs. NULL values and missing entries can be checked before submission. Rules can also be established for updating master data when new stores or products are added.

That said, it is not necessary to demand a perfect unified format from all regions from the very beginning.

A more practical approach is to first automate the process on the headquarters side, identify where discrepancies tend to occur and which fields are often missing, and then gradually improve formats and input rules where needed.

We believe this approach is easier for local teams to accept and more sustainable in actual operations.

Conclusion: How GDX Can Support You

Through this project, I was reminded once again that AI adoption is not simply about introducing a tool.

What matters even more is how clearly the meaning of the data and the business workflow are organized before AI is used.

AI can be a powerful tool for organizing regional changes, identifying product-level trends, and creating draft comments for reports.

However, for AI to work effectively, the data it reads must first be reliable.

The purpose of automation is not to eliminate human work. It is to allow people to spend more time on the analysis and judgment that truly require human expertise.

At GDX, we provide end-to-end support, from reviewing current business workflows and organizing format differences across regions and subsidiaries, to designing data transformation rules, generating reports with AI, and improving submission formats.

If your company is facing challenges such as spending too much time creating weekly or monthly reports, dealing with different report formats across regions or subsidiaries, or wanting to use AI but not knowing where to start, please feel free to contact GDX.

Part of this article was created with the support of ChatGPT and then edited and revised by the author. The content reflects the author’s personal views and does not represent the official views or statements of GDX Co., Ltd. The information is for reference purposes only; please refer to official announcements and primary sources for confirmed information.