Your master data holds one business partner name. A corporate information API such as Tokyo Shoko Research returns another. When the two disagree, the cause may be a spelling variant, a difference in legal entity type, a change of company name, or a mis-set code. Partner Name Matching AI determines the reason automatically across 11 categories, and leaves only the cases that genuinely need a decision to your reviewers.
Many tools can calculate similarity. The hard part comes after that: classifying why the names differ. That is the part Partner Name Matching AI takes on.
Upload the reconciliation Excel you work with today. Normalization of the names, the classification itself and the generation of the result Excel all run automatically. There is no new input format to learn.
The input format is detected automatically from the column names. Both the conventional reconciliation sheet and the newer MDM + corporate information API format are accepted.
On real data, 78.2% was decided by the system and reviewers looked at the remaining 21.8%. Every case that remains carries a reason code for why it needs review, so you know where to look from the start.
Differences between the names are shown color-coded, with the views of the deterministic rules, the machine learning model and the generative AI placed side by side. Similar past cases are attached as evidence. Approvals and corrections can be handled one after another from the keyboard.
For each decision we record the input, the rule version, the model version, the similar past cases referenced, the source of the decision (rule / ML / AI / human) and the approver. The same input on the same versions returning the same result is a design requirement, so any decision can be reproduced later.
Measured on 8,251 rows of real data.
These figures are measured against the labeled data set. Re-measurement on new data is carried out in the pilot at onboarding, and we report the results. We do not construct numbers that make the product look stronger than it is.
Anything decided upstream never reaches the next stage.
Constraints that prevent an incorrect auto-decision are built into the design.
We look at your actual reconciliation Excel and align on the column structure and the classification criteria
We run the decisions on your real data, measure the auto-decision rate and the precision, and report back
User registration, permission setup and handover of the operating procedures
If the numbers fall short of what you expected, you are free to stop there.
The service is provided for a monthly fee, which includes cloud usage, maintenance and technical support.
Options are available for a monthly accuracy report, model updates that reflect your review results, and scheduled export of the decision history.
The actual price depends on volume and on the number of users, so please get in touch.
This service is the software itself. Checking and approving the results is done by your own reviewers. If you would like the review work handled for you, please talk to us separately.
The system assumes you supply an Excel file that already contains the search results, so it never calls the API itself. The API contract remains yours to hold.
Adjustments within normal operation, such as adding terms to the dictionary, are possible. Anything that changes the classification definitions themselves is quoted case by case.
It can be corrected on the review screen, and the correction is kept in the history. With the model update option, those corrections are reflected in later decisions.
No. Everything is handled by exchanging Excel files. No changes to your core systems are required.
Give us one of your actual reconciliation Excel files and we will measure the auto-decision rate on your data and report back. Decide once you have seen, in numbers, how much of the current workload can be removed.
Request a measurementFor applications and enquiries, please write to info@tuweii.com.