Bold Group
The fix needed two AI tools, not an AI system.
A public office registering trademarks, receiving about 250 applications a day and approving about 150. An application took 12 months on average to register, and every desk carried a backlog of its own.
The situation
Every desk had a backlog, so the obvious answer was to automate every desk.
An application could arrive six ways: a paper form, a service counter, a provincial office, the post, an international filing, or online. Fees came in three ways, all matched by hand. About 250 applications arrived a day, and about 150 were approved.
On average a mark took 12 months to register, not counting the 60-day public notice. With every paper in order, the work itself could take about three hours. The rest was waiting: a 4-month backlog at examination, 8 months at the registrars, and 15 months for international filings.
The turn
The queue was full of files nobody could finish.
Every desk is slow, so every desk needs automating.
What has to be true of a filing before anyone should touch it?
The decision: stop incomplete files at the door, and point AI only at the two searches people cannot do quickly.
Before
With the teams at each desk, we drew the route one application takes from filing to certificate, and marked every point where it is sent back. There were five.
- Service counter. Papers incomplete or wrong: back to the applicant. Under a signed undertaking, an incomplete file could still take a number and enter the system.
- Examiner. A translation, signature, power of attorney or the stated use of the mark is missing or unclear: back to the applicant.
- Registrar. A correction order. The reply comes back as a PDF that cannot be searched for the corrected point, so the whole file is read again.
- Records. The order is printed, partly by an outside printer and then re-checked by staff, and sent. Each reply is checked against the application and every earlier fix.
- Publication. The draft notice needs correcting: back to the registrar.
Every return means a wait for the applicant’s reply, and the reply lands back on a desk with a backlog of 4 to 15 months. Files coming round again join the same queues as new ones.
12 months on average to register, excluding the 60-day public notice. With every paper in order, about 3 hours of work.
After
Corrections and oppositions still go back to the applicant, but only through e-filing, where the file and every change to it sit in one system. Design target: no more than 7 days, excluding the 60-day public notice.
Simplified from the process map drawn with the agency’s teams. Not to scale.
How we worked it
How the design was built.
- 01
Map every desk, and every way back
With the teams at each desk, draw the route one application takes from filing to certificate, and mark every point where it is sent back. Count the months it waits, not the minutes anyone works on it.
- 02
Find the assumption behind the checks
Three root causes: the process assumed applicants could not understand the law or the procedure; it let incomplete data into the system; and the same checks were repeated from desk to desk.
- 03
Move the checks to the applicant’s side
One e-filing channel replaces the other five. The applicant crops the image to size, states how the mark reads and what it means in other languages, and names the classes precisely enough for a registrar to accept, while the form checks each entry. Fees go through PromptPay, confirmed by the bank in real time. An applicant can still take a number before everything is in, but the file stays on their side until it is complete.
- 04
Give AI only the narrow jobs
Records and examination become one team. In the short term, staff are trained to index image codes and transcribe readings in English and Thai. In the long term, two AI tools take the searches nobody can do quickly by hand: similar images, and names that sound alike (Soundex). Both already exist as services. Examiners still judge what the law prohibits.
Mapped the end-to-end process with the agency’s teams across filing, examination and registration, found the root causes, and designed the single-channel workflow, the action plan and the staffing recommendations.
Holds the legal requirements and the examiners’ judgement, and decides what is implemented, and when.
What came of it
One online channel that checks every file at entry, one examining team, and two AI tools instead of an AI system. The design targets registration in no more than 7 days, against 12 months on average today, excluding the 60-day public notice.
Alongside the workflow, the design moves staff from intake and registration into examination; gives walk-in applicants computers, scanners and outsourced help to file online; and trains and certifies design agencies to advise and file for applicants. It also asks the agency to consider retiring the paper form, and letting local post offices, banks and municipalities confirm an applicant’s identity.
This is a design, presented to the agency’s management. Implementation and any change in registration time are not established; the 7-day figure is a design target, not a result. The agency is not identified.
How we label evidenceIs this your situation?
Signs
- Every stage has a backlog, and the proposed answer is to automate all of them.
- Work keeps returning to the start for corrections.
- Corrections come back as scans or PDFs, so each one means reading the whole file again.
What it needs from you
- The people who own the rules and the people who run each desk, designing it together.
Start here
Bring the result you need, and the thing you have been told cannot change.
One hour with a founder. You leave with a first view of where the answer may sit, and which engagement, if any, fits.
Teerasak WongpiyaCo-Founder and Managing Director
1 hour, in Thai or English, online or at your office.