Stop spending the day
looking for documents.
Your PDFs, spreadsheets and Word files are read across all at once.
Ask a question, and the answer comes back with the documents it came from.
It does not fill in what the documents do not say.
01 What the problem is, and what changes
This is the time that goes away.
- Opening folder after folder for a file somebody made two years ago.
- Asking a colleague because it is faster than searching.
- Four people checking the same thing separately.
- Waiting because the one person who knows is out today.
Looking for something costs twice.
Once for the person searching, and again for the colleague they interrupt. Neither half shows up anywhere as work, which is why it is easy to leave alone for years.
Say you need last year's hot-spring feature in Kyoto.
Today
- Open the shared drive and guess at folder names.
- Try the file server, then the old project folder.
- Ask the producer who might remember.
- Find it, and check whether it is the final version.
With AI Desk
- Ask in the chat app you already use.
- The answer comes back with the file and the page it came from.
- If the documents do not say, it tells you that instead.
02 How it works
Three steps.
-
Register your documents
PDF, Excel and Word are handled as standard.
-
Ask from your chat app
The entry point is LINE, the messaging app already used across Japan.
-
Get an answer with its sources
Each answer names the documents it was read from.
So the answer can be checked.
- Every answer names its sources. You can open the file it came from.
- It refuses rather than guesses. When the documents do not contain the answer, it says so.
- Your data stays inside your boundary. Search is scoped per customer and per area of work.
03 What is included
Included in a standard rollout
Decide the work to target
We narrow down together which searching and checking to reduce.
Take your documents in
PDF, Excel and Word as standard. Scanned paper and images are an option, depending on the condition and volume of the material.
Search across formats
One question reaches several documents without you thinking about which format each one is.
Return the evidence
Answers name the documents they were read from. When something cannot be confirmed, that is what comes back.
Make it usable from LINE
The chat app your team already has becomes the entry point.
Set the data boundaries
Search is separated per customer and per area of work. Different visibility per department or group is available as an option.
Verify the behaviour
The AI generates 20–50 representative check questions. You only need to review and add the 5–10 that matter to your work.
Go live
Monitoring after launch, and the early adjustments.
The standard scope
- One organisation, one LINE channel, up to about three areas of work.
- Initial load of roughly 2,000 files or 10 GB is the standard estimate.
- Four to six weeks is the usual timeline.
04 How a rollout runs
The five stages
-
AI Business Assessment
We ask about the work you want to improve, and confirm the kinds of documents, where they live, who will use them, and who may see what. The goal is to be able to say in one sentence who asks what, from which documents, over LINE.
-
Data and environment
We prepare your environment, take in the documents, and set the boundaries.
-
Build
Search, answers, source display, group separation and notifications are configured.
-
Verification
The AI generates 20–50 representative check questions and they are run against your real documents. You review the 5–10 that matter. Together we confirm that answers carry sources, that it declines when it cannot tell, and that the data boundaries hold.
-
Launch and operation
Monitoring after launch, and the early adjustments.
05 Why we build it this way
Why this product exists
An AI that sounds convincing and is occasionally wrong is not useful for work. What matters at work is being able to check the answer, and being told plainly when there is nothing to go on.
So this is built the other way round. The answer is assembled from your documents, it carries the places it came from, and when the documents do not say, it says that rather than filling the gap.
06 Whether it suits you
This suits companies where
A good fit
- There are many PDFs, spreadsheets and Word files, and past material is looked up often.
- Several people repeat the same checks.
- "Who knows about that one?" depends on a particular person.
- Searching and checking is the first thing worth improving.
Probably not a fit
- You are looking for a tool under a few tens of thousands of yen a month.
- Operation entirely inside your own servers is a requirement.
- You are planning to replace your core company systems.
07 What it costs
Indicative pricing
What is described above is delivered as a standard rollout, at the following prices. Amounts are in Japanese yen.
- AI Business Assessment
- JPY 150,000–300,000
- Decides the target work and settles the scope
- Standard rollout
- JPY 1,200,000–1,800,000
- Indicative, for the standard configuration
- Monthly operation
- JPY 100,000–200,000
- Typically around JPY 150,000
- Varies with the volume of documents, the number of users, and any systems to connect to.
- A more complex configuration, including integration with external systems, is indicatively JPY 2,200,000–3,000,000.
- The exact figure is quoted after the AI Business Assessment has settled the scope.
08 Questions we are asked
Frequently asked
How is this different from a general AI assistant?
It is not a general assistant used as-is. It is built to answer from your documents, with the sources attached. When it cannot confirm something from them, it says so.
Will our data be mixed with another company's?
Data is separated per customer, and the scope is decided at rollout. The details are settled during the AI Business Assessment.
Do we have to move to a new tool?
No. Questions are asked from LINE, which teams in Japan already use daily. There is no new application to roll out.
What if the answer is wrong?
Every answer names the documents it was read from, so it can be checked. Where the documents do not support an answer, none is invented.
Can it read scanned paper?
PDF, Excel and Word are standard. Scanned material and images are handled as an option, depending on their condition and volume.
Do we have to tidy up all our documents first?
No. Material scattered across folders can be used as it is. Where to start is decided together during the AI Business Assessment.
How accurate is it?
It answers what the documents say, with the sources attached, and returns "not found" for what it cannot confirm. How much it can answer depends on your documents, so that is checked against your real material during the AI Business Assessment.
Are our documents sent to an external AI service?
An external AI service is used to generate the answers. Exactly what is sent, and whether it is excluded from training, is explained during the AI Business Assessment, under the terms in force at the time of contract.
Can it be used from something other than LINE?
LINE is the standard entry point. Support for other tools is something to discuss with us.
We are not sure our staff will use it.
The entry point is the chat app they already use. During the verification stage before launch, the people who will actually use it try it themselves.
How long does a rollout take?
Four to six weeks is the usual timeline. The actual schedule is given once the AI Business Assessment has settled the scope.
Tell us what your team keeps looking for.
Write to us in English with the kind of documents you hold and the work you would like to reduce. A member of GOLBA replies.