Before AI helps you file,
read the map first

Every year after the books close, every Thai legal entity must submit its financial statements to the Department of Business Development through DBD e-Filing. On the surface it looks like filling in a form, but there are many specific rules hiding underneath: the file format, the filing order, an accounting standard that just changed, and balance checks the system enforces.

Faced with repetitive work like this, many people want to add AI right away. But the lesson from actually building accounting tools is that a good tool starts from understanding the work fully, not from the AI. So this piece lays out the whole filing process first, then uses it as a worked example of where AI genuinely helps.

It goes step by step: what DBD e-Filing is, who must file, and what, then the real process on both tracks, then what changed from 2024 onward, and finally how to analyze the work to place where AI stands.

Part 1What DBD e-Filing is, who files, and what

DBD e-Filing is the Department of Business Development’s online system for filing financial statements and the shareholder list (Bor.Or.Jor.5). The statements must be prepared in XBRL using DBD XBRL in Excel (currently V.2.0): fill them in Excel offline, then convert to a file to upload. This is the only channel for the statements; there is no web form.

There are five types of entities required to keep accounts, each with its own line-item form.

• Form 1 registered partnership
• Form 2 limited company
• Form 3 public company
• Form 4 entity established under foreign law
• Form 5 joint venture under the Revenue Code

This piece focuses on the limited company (Form 2). The filing package is: the statements (.zip XBRL file), the notes (attached as PDF or filled into sheet 510000 of the XBRL file), the auditor’s report page (PDF), the Bor.Or.Jor.5, and the Sor.Bor.Chor.3 form. Attachments must be portrait, black-and-white PDF.

On timing, take a 31 December year-end as an example: the shareholder meeting is around April, the Bor.Or.Jor.5 is filed within 14 days of the meeting, and the statements within one month of the meeting. A common mistake: the Bor.Or.Jor.5 cannot be filed together with the statements. The DBD requires it separately, and it usually comes first. Late-filing fines follow a published schedule, with a two-year limitation period.

Part 2The real process: two tracks, three steps

Filing splits into two tracks: the Bor.Or.Jor.5 and the financial statements. Each runs the same three steps: prepare the data, submit, then check the result. The statements go through DBD XBRL in Excel only. The real sequence looks like this.

What to know at each step.

1. Download the template per entity. Log in with the 13-digit registration number, then answer a wizard to set the correct statement layout: the accounting period, standard (NPAEs or TFRS), notes format, income-statement type, cash-flow statement, and whether there are retrospective adjustments. Choose wrong here and the whole file is wrong. Unzipping gives an Excel file and a Java Builder that must stay in the same folder and must not be renamed, or the XBRL conversion will not work.
2. Open the file, Unblock, and Enable Content so the macros run, then fill every sheet except the check sheets. There are specific rules, for example sub-lines total no more than 8 per main item; if a line does not apply, leave it blank or type “none”, and never delete rows.
3. Convert to XBRL. The system validates during conversion; if something is wrong you must fix it first, and if it passes you get a .zip file.
4. Submit. Pick the accounting period, fill in the auditor and the accountant details, attach the statements, notes, and auditor’s report page, confirm, then print the submission slip.
5. Check the result. Print the Sor.Bor.Chor.3 form; the status moves from “processing” to “received”. If a defect is found, you get an email, then you fix it, send a letter of explanation, and wait for the officer to review.

The system enforces several balance checks, for example: total assets must equal liabilities plus equity, paid-up capital must not exceed registered capital, the opening and closing cash in the cash-flow statement must match the balance sheet, and the profit in an indirect cash-flow statement must match the income statement. If they do not match, the file will not convert.

Part 3What is new since the 2024 filing

Statements filed from 2024 onward look different, because two things changed at once.

1. The standard changed. TFRS for NPAEs (the financial reporting standard for non-publicly-accountable entities), 2022 revision, effective 2023.
2. The line-item layout changed. The DBD announcement on required statement line items, B.E. 2566 (2023), repealed the old ones and applies from the 2024 filing year.

What is visible: statement names changed, several income-statement formats are now allowed, new lines were added, and the notes must cite the 2023 announcement. The practical effect: keep enough prior-year data, because you may need to restate the comparatives. This kind of change is exactly why “get this year’s process right” matters more than following last year’s habit.

Part 4Task analysis: placing where AI stands

The question people usually ask wrong is “can AI file the statements for me?”, because “file the statements” is not one task but many different sub-tasks joined together. Before answering whether AI can do it, you have to break the work into pieces first. This is called task analysis, and it is what we actually use when deciding which tool to build. In plain terms: take the process you now understand from the parts above, and look at it piece by piece to see where AI should help.

Step 1: break the work into a tree

Take the filing process and split it into sub-tasks, instead of seeing one lump called “file the statements”. Seen one at a time, it is: pull the trial balance, map accounts into the line-item layout, build the Form 2 statements, draft the notes, pass the balance checks, key it into XBRL in Excel, then submit. Just splitting it out shows that the tasks are not alike: some have a fixed answer, some need a person to decide.

Step 2: ask three questions of each sub-task

1. What data does it take as input, and is that data clearly structured?
2. Is the decision fixed right-or-wrong (one answer, checkable) or judgment (depends on context and law)?
3. How bad is a mistake, and is it reversible? Filing wrong brings both fines and a defect status, so the cost of error is high.

Step 3: group by the answers, then assign who does it

This table says four things.

• For the fixed right-or-wrong parts, like the balance checks and the going-concern flag, do not let AI guess. Pull them out into a rule-based checker that returns the same result every time and is auditable. Same principle as the post don’t let the LLM do the math: separate the part that computes exactly from the part that needs judgment.
• For the pattern-matching parts, like extracting figures, arranging lines, and drafting notes, let AI do it and have a person sign off. This is where AI genuinely saves a lot of time.
• The account-to-line-item mapping has a trap: each entity’s chart of accounts differs, so you cannot write the rules in advance from a guess. You have to see that entity’s real chart of accounts first. Work that connects to the real world like this differs from work that is pure logic, and this is where people slip when they rush AI in.
• For the judgment and legal parts, keep a person’s hand on it, not because AI is dumb but because the liability rests with a person.

Without breaking the work down first, you get a tool that does everything so-so instead of doing the right parts really well. Task analysis is the same discipline as judging whether to build a tool at all before starting: build from understanding the real work piece by piece, not from a guess.

In short, and where to start

Understand the process fully first, then split the work into two kinds, “repetitive and pattern-matchable” versus “fixed right-or-wrong or judgment”, and only then place where AI stands. A first thing you can do today: draw your own team’s filing process on one page, then mark which lines are repetitive and which have a fixed rule. That alone shows where to put AI first.

If you want your accounting team to actually do the parts AI helps with, on their own work, we run a Claude Cowork training for accounting teams, at productize.life/services.