Identity Resolution Benchmark Results

In head-to-head testing against the top 15 identity resolution providers, DarkMath's vector-based architecture demonstrated significant advantages in accuracy, match rates, and false positive reduction. These results represent real-world performance on production-scale datasets, not synthetic benchmarks.

Head-to-Head Comparison:
Accuracy & Match Rates

Metric

DarkMath

Competitor

Improvement

F1 Score (vs. Company 1)
86.44%
72.14%
+19.82%
Match Rate (vs. Company 1)
78.96%
61.36%
+28.68%
F1 Score (vs. Company 2)
83.42%
68.86%
+21.14%
Match Rate (vs. Company 2)
72.22%
54.91%
+31.5%

Household Resolution: Reach vs. Precision

Metric

DarkMath

Competitor (company 3)

Total Households Resolved
10,838,389
6,992,837
Additional Reach
+3.8 million households (+55%)
Baseline
Maximum Household Size
37 (realistic)
994 (false positive mega-clusters)
Precision Assessment
Eliminated mega-clusters
Severe false positive problem

Duplicate Detection Performance

Metric

Result

Input Records (Competitor Distinct Count)
10,838,389
After Deterministic + Fuzzy Matching Processing
+3.8 million households (+55%)
Final Deduplicated Count
37 (realistic)
Total Duplicates Found
Eliminated mega-clusters
Competitor Detection
0% (baseline—duplicates undetected)

What These Numbers Mean

F1 Score: The Balance of Precision and Recall

F1 Score is the harmonic mean of precision (what percentage of predicted matches are correct) and recall (what percentage of true matches are found). DarkMath's 86.44% F1 means we find most true matches while avoiding false merges, a 19.82% improvement over competitors who sacrifice precision for recall or vice versa.

Match Rate: Finding Connections Others Miss

A 78.96% match rate means DarkMath successfully resolved 78.96% of record pairs that should be linked. Competitors at 61.36% are missing nearly 40% of valid connections, fragmenting customer views and leaving revenue on the table.

Household Size: The Mega-Cluster Problem

When a competitor creates a "household" with 994 members, something is deeply wrong. These mega-clusters occur when overly permissive matching rules merge unrelated records, apartment buildings, common names, or simply errors compounding. DarkMath's maximum household size of 37 reflects real-world limits, ensuring your analytics aren't corrupted by false aggregations.

Duplicate Detection: The Hidden Problem

Finding 4.3 million duplicates that competitors missed means those competitors' customers are unknowingly marketing to the same people multiple times, calculating incorrect lifetime values, and making decisions on corrupted data. A 26.91% reduction represents massive operational savings.

Methodology

Benchmarks were conducted on production-scale datasets with ground truth established through extensive manual validation and synthetic corruption testing. Tests compared identical input data processed through each vendor's resolution pipeline. Results anonymize competitor identities to focus on capability comparison rather than vendor criticism.

To validate these results on your own data, contact us for a proof-of-concept evaluation.