AI comparison and transcription for order and inspection documents
Comparing documents? AI checks them first.
Purchase orders and confirmations, two specifications, inspection reports and Excel. AI checks them line by line and marks only what differs or needs a human check. Your team reviews the marked lines and presses Enter to move on.
Writes values from a PDF into the Excel form you already use. The form is not rebuilt.
You provide
One PDF and the target Excel file
You get
Your Excel with values filled in (layout and formulas intact), plus a status for each cell
For example
Inspection report → inspection checksheetPaper form → summary Excel
Cell
Inspection report
Excel
Status
C5
Sample 001 Ra 1.2
1.2
FilledRead with confidence
C6
Sample 002 Ra 1.4
1.4
FilledRead with confidence
C7
Sample 003 Ra (blank)
(blank)
ReviewNever guessed
There is also an integration that drafts freee invoices from COREC order data.
Results
How daily checking changes.
TodayWith Totsugo
TodayLay two documents side by side and follow every line by eye
With TotsugoAI checks every line first. People look only at the marked lines
TodayNothing records what was checked, so the second check starts over
With TotsugoEach decision keeps the source location. The second reviewer checks the evidence
TodayNotation differences like "¥" vs "yen" distract from the real differences
With TotsugoNotation differences count as a match. Only real differences and blanks remain
TodayRetype PDF values into the Excel form by hand
With TotsugoCheck the filled Excel and save it. The form stays as it was
Customer case
Nabel Co., Ltd.
A bellows manufacturer in Mie, Japan, that compared around 100 × 100 PDF pages a day by hand. AI now does the first check, and people review the differences.
People on order checks
3 → effectively 1
Total working time
About half
Order checks a year
~20,000
Transcription result
8 hours a day → about 10 minutes
Transcribing from COREC to freee, including review
Choose the two PDFs to compare (or a PDF and an Excel file for transcription) and start.
2
AI compares them
Pages are rotated upright, grouped by order number or similar, and compared field by field. No prompt needed; add rules in plain words only when you have them.
3
Check the marked lines
Filter to differences and review items, compare with the source, and press Enter to move on. Export a report or the Excel file.
Why you can rely on it
AI does not decide. People do.
Where AI stops and human judgment begins is the core of the design. We prioritize not missing anything over making the review list shorter.
Example: a fictional purchase order × order confirmation
Fields compared
33
Match
26
Review
5
One side only
2
Instead of rereading all 33 fields, start from the 7 that are not a match. The counts are not an accuracy rate; they show how much a person needs to review.
01
Similar is not the same
"Fixing bracket (auxiliary)" and "Auxiliary fixing bracket" stay in review when the meaning could differ. You can state in plain words what counts as equivalent.
02
What is not written is not filled in
A blank measured value stays blank and is flagged, even when neighboring values make a guess easy. "Not measured" is itself a fact worth recording.
03
Dates are checked by people by default
So that a ship date is never mistaken for a delivery date. You can let AI match dates per job.
04
Every decision points to its source
Each decision links to where in the source it came from, so a second reviewer or a new team member can follow the reasoning.
Fit
Where it fits, and where it does not.
A good fit
You compare two documents line by line every day
You handle hundreds of comparisons or transcriptions a month
Your documents include faxes, scans, handwriting, or inconsistent wording
Not a fit
You want results finalized with no human review
Your documents stay on paper and cannot be scanned
You only have a few documents a month (manual work is enough)
Unsure? Bring your real documents and we will work it out together.
Getting started
Start by trying your own documents.
STEP 1
Free consultation (30 min)
Tell us which documents you compare, the fields to check, and your monthly volume. Online.
STEP 2
Trial on 1–2 documents
We process your documents and report the results with screenshots and decisions.
STEP 3
Quotation
We propose a plan and price based on the results, your volume, and team size.
STEP 4
Start
We issue your account, and you review the first results with us.
Plan guide
Pricing is quoted individually based on your document types and volume.
Q.Could we build this in-house with ChatGPT or another general-purpose AI?
You can try. To rely on it every day, though, you need to read hundreds of pages, handle scans and handwriting accurately, and give people a screen that shows where the AI looked for each decision. Without that, everything gets rechecked and no work is saved. AI models are also updated or retired often, and each change means revalidating and retuning the results. Totsugo carries that maintenance for you. If you are weighing building it yourself, include the upkeep after launch in the comparison.
Q.Do we have to write a prompt for every comparison?
No. Document-to-document comparison reads and compares without any instructions. Add plain-language rules only when your team has a specific standard, such as "treat ¥ and yen as the same" or "keep quantity differences in review". You can also state how to group documents, such as "by order number" or "by part".
Q.What is the accuracy rate? Can we check before committing?
We do not quote a single accuracy figure, because it depends on resolution, scan quality, layout, and character types. Instead, share one or two of the documents you plan to compare, and we will report the actual results with screenshots and decisions. The counts shown on screen are not an accuracy rate; they show how much a person needs to review.
Q.What if the AI misses something or gets it wrong?
Every result shows exactly which part of the left document was compared with which part of the right. AI never finalizes anything on its own; a person always makes the final call and can correct a result on the spot before confirming with Enter.
Q.What types of documents are supported?
Ask us about orders, invoices, specifications, inspection reports, checklists, and other PDFs. For custom layouts, faxes, and handwriting, we check the supported fields and accuracy with your actual documents.
Q.Will it break our Excel forms?
No. Only the values are written into your original Excel file, so merged cells, formulas, and print settings stay intact. AI proposes values and a person confirms them cell by cell.
Q.Do we need to change our workflow?
No. Use the PDFs and Excel files you already have. No format changes or system integrations required.
Q.How is my uploaded data handled?
All traffic is encrypted via SSL/TLS, and stored data is encrypted at rest. Your documents are never shared with third parties without your consent. See our Privacy Policy for details.
Q.How long does setup take?
It depends on your documents and review requirements. After the trial on your documents and the quotation, we issue an account for your email address.
Who builds it
Takuya Tominaga
CEO, Leach Inc.
Engineer
I do not believe AI alone solves this, nor that having people approve everything does. The goal is not to replace human judgment, but to give people time to use it.
At Toshiba's Software Technology Center he worked on R&D close to manufacturing operations, from HDD firmware to data platforms and business applications. After a secondment to the CDO Office at Toshiba Tec headquarters, he founded Leach Inc. to focus on generative AI.
Graduated from National Kumamoto Technical College Advanced Course. Completed Master's program at NAIST (Nara Institute of Science and Technology).
Sep 2025 - Lecture on Generative AI at JSSST
Jan 2024 - All AWS Certifications achieved
Jul 2014 - IPSJ DICOMO2014 Best Presentation Award (1st of 293)
Leach Inc.
CEO Takuya Tominaga
Fudanotsuji Square 9F, 5-36-4 Shiba, Minato-ku, Tokyo 108-0014, Japan
