Concern-Level Scoring and Name-Matching Technology: The Engines Behind Screening Accuracy
Hello, this is Hamamoto from TIMEWELL. Today I want to take a deep look at two key technologies that underpin accurate counterparty screening: "concern-level scoring" and "name-matching technology."
"I can see the screening results, but I can't tell which cases to prioritize." "When names are slightly different, I can't judge whether it's the same person." "I'm worried we might be missing cases due to name variations."
These are concerns we hear from many companies. Screening is not an end in itself — the goal is to identify counterparties that warrant concern. This article explains the technologies that make screening accurate and meaningful in practice.
Chapter 1: The Importance of Concern-Level Scoring
"Match Found" Alone Is Not Enough
Traditional screening returned a binary result: "match found" or "no match." But that alone often does not provide enough to act on.
When "match found" comes back, you still need to determine whether this is a genuinely concerning target or simply someone with the same name as a different individual. Conversely, "no match" does not mean the risk of a missed detection is zero.
Scoring as the Solution
Rather than a simple hit/no-hit result, TRAFEED provides results in the form of a "concern-level score." The S-to-C ranking system makes it easy to judge which cases deserve priority attention.
Concern-level score rankings:
| Rank | Meaning | Recommended action |
|---|---|---|
| S | Highest concern | Detailed investigation mandatory. Consider suspending the transaction |
| A | High concern | Detailed investigation required. Gather additional information |
| B | Medium concern | Risk-based judgment. Investigate as the situation warrants |
| C | Low concern | Proceed through normal process. Continue periodic monitoring |
Table 1: Concern-level score rankings and recommended actions
How Scoring Works
Concern-level scores are calculated by evaluating multiple factors together.
Primary evaluation factors:
- Degree of match with sanctions lists (exact match, partial match, similarity)
- Name similarity (how closely the names resemble each other)
- Risk level of the country of operation
- Concern level associated with the industry/business type
- Presence of related information (affiliated organizations, prior transaction history, etc.)
- Multi-LLM consensus results
Each factor is weighted, and the weights are combined into an overall score. Rather than relying on any single element, the composite evaluation enables more accurate judgment.
Replace siloed classification work with AI.
METI's FY2024 data shows 52% of foreign exchange law violations stem from classification errors. Download the TRAFEED product catalog covering features and rollout.
Chapter 2: The Difficulty of Name Matching
The Reality of the Variation Problem
The most technically challenging aspect of counterparty screening is name matching. Even the same person or organization can appear in many different forms depending on how the name is written.
Romanization variations: The Arabic male name "محمد" can be romanized as "Muhammad," "Mohammed," "Mohamed," "Mohamad," "Muhammed," and more — upward of ten variations. Which romanization appears on a given sanctions list varies by list.
Kanji and kana variations: Even within domestic Japanese business, name variations are common: "Kabushiki Kaisha ○○" versus "(Kabushiki) ○○," or "Tokyo ○○ Kabushiki Kaisha" versus "Tokyo ○○."
Name changes over time: Rebranding, mergers, and spin-offs mean that a company may have a different name today than it did in the past. Sanctions lists often record the name as it was at the time of designation, so searching by the current name may return no results.
The Limits of Traditional Approaches
Exact match search: Only flags when strings are identical character for character. Any slight variation means a miss.
Partial match search: Flags when part of the name matches. Produces too many false positives to be operationally useful.
Fuzzy search: Uses edit distance or similar measures to flag high-similarity strings. Setting the right threshold is difficult, and accuracy has inherent limits.
Chapter 3: TRAFEED's Name-Matching Technology
Applying Linguistic Knowledge
TRAFEED performs name matching using linguistic knowledge. The system has learned the transliteration rules that govern how Arabic, Chinese, Korean, and other names are converted into Roman letters, and automatically generates the range of possible variations.
For the name "Muhammad," the system generates alternative spellings like "Mohammed" and "Mohamed" and cross-references each one against sanctions lists. This prevents missed detections caused by spelling differences.
Joint Validation with Okayama University
This name-matching technology was developed through collaborative research with Okayama University. Combining academic expertise in language processing with the practical demands of export control operations produced a high-precision matching algorithm.
The system achieves particularly high detection accuracy for Chinese, Korean, and Arabic-origin names — the name groups that Japanese companies most commonly encounter in cross-border business.
Using Contextual Information
Matching considers not only the name itself but contextual information such as address, nationality, date of birth, and affiliated organizations.
"Wang Wei" is an extremely common name in China. Matching on the name alone would generate enormous numbers of false positives. But factoring in address and organizational affiliation makes it possible to narrow down to the genuinely concerning targets.
Inferring Organizational Relationships
The system can also infer that "ABC Holdings" and "ABC Corp" are related entities and evaluate concern levels for both. When a parent company is subject to sanctions, transactions with its subsidiaries may also warrant scrutiny.
Chapter 4: How to Use These Features in Practice
Setting Triage Criteria
Concern-level scores can be used for triage — prioritizing which cases deserve attention. Within limited time and resources, they provide a clear basis for deciding where to focus.
Example triage rules:
- S-rank and A-rank: Always conduct detailed investigation
- B-rank: Conduct detailed investigation when transaction value exceeds a set threshold
- C-rank: Quick check only. Continue monitoring through periodic reviews
Designing Approval Workflows
Adjusting the approval workflow based on score level is also effective.
- C-rank: Practitioner-level approval
- B-rank: Section manager approval
- A-rank: Department head approval
- S-rank: Executive approval
Adding higher-level review to higher-risk cases ensures appropriate governance.
Integration with Continuous Monitoring
TRAFEED performs ongoing monitoring of all registered counterparties. When sanctions lists are updated, an automatic re-evaluation is triggered, and alerts are issued if scores change.
A counterparty that was previously C-rank can become S-rank if newly added to a sanctions list. Continuous monitoring ensures these changes are not missed.
Chapter 5: Accuracy Improvement Track Record
Reducing Missed Detections
Companies that have implemented TRAFEED frequently report that "cases that might have been missed under the old approach are now being detected."
Feedback from an implementing company (manufacturer): "In the three months after implementation, we identified 15 cases that our previous approach likely would have missed. Detailed investigation confirmed that none were actually sanctioned parties, but the reassurance of knowing our miss risk has been reduced is significant."
Suppressing False Positives
Increasing detection sensitivity typically increases false positives — a fundamental tradeoff. TRAFEED mitigates this by leveraging contextual information alongside name matching.
By evaluating names together with supporting information, the system can eliminate "names are similar but clearly different individuals" cases as false positives. Narrowing the cases that practitioners need to review to a manageable number reduces the operational burden.
Ongoing Accuracy Improvement
TRAFEED's accuracy improves continuously through user feedback. When false positives or missed detections are reported, those cases inform algorithm refinements.
Chapter 6: Operational Considerations
Scores Are Not Absolute
Concern-level scores are a tool to support judgment — not an absolute standard. Low-scoring cases with genuine concerns exist, as do high-scoring cases that turn out to be entirely unproblematic.
The final determination must always be made by a human who takes responsibility for it. Treat scores as one input, and combine them with other available information for a well-rounded assessment.
Threshold Calibration
It is possible to calibrate the score thresholds to reflect your organization's risk tolerance and the nature of the products you handle. Companies dealing with many controlled items can set stricter thresholds; companies with lower-risk product lines can apply somewhat looser criteria. Operating parameters can be tailored to your actual situation.
Regular Accuracy Verification
We recommend periodically verifying scoring accuracy. Analyze cases where "S-rank was returned but turned out to be unproblematic" and cases where "C-rank was assigned but a problem emerged later," and use those insights to improve operations.
Conclusion: Accuracy Can Be Improved Through Technology
Name variation issues and result prioritization have been longstanding challenges in screening work. Manual approaches have inherent limits — the risks of missed detections and inefficiency cannot be fully eliminated.
TRAFEED addresses these challenges through a combination of linguistic knowledge, AI technology, and continuous improvement. High detection accuracy alongside appropriate false-positive suppression; prioritization that is genuinely actionable in practice — these capabilities support export control practitioners in their day-to-day work.
If you have concerns about your screening accuracy, we invite you to try TRAFEED. A free trial will let you experience the results directly.
References [1] Okayama University, "Research on Multilingual Name-Matching Algorithms," 2025 [2] ACAMS, "Best Practices in Name Screening," 2025
If you are reviewing export-control operations or classification workflows, download the TRAFEED product catalog (PDF) or contact us.
Related Articles
- 2024 Guide: What Is Export Classification? From Basics to Practical Application
- Critical Minerals: Next-Generation Mining Through Vertical Integration and Technological Innovation
- Defense Innovation Frontier: DIU and Applied Intuition on Closing the Technology Gap and the Future of National Defense Strategy