Enrichment Quality, Measured in Production
Real benchmarks from real bank transactions. Not synthetic tests — actual data processed by our API from paying customers across 13 countries.
Avg Confidence
75.5%
Across all merchants
52% score ≥0.95
Avg Latency
287ms
With cache (71.7% hit rate)
p50 uncached: 5.2s
Merchants Enriched
129
Unique merchants identified
20+ categories
API Calls Served
1,874
Total production requests
13 countries
Confidence Score Distribution
How confident is the AI across all 129 production enrichments? Higher is better.
81.4% of enrichments score ≥ 0.70 confidence
51.9% score ≥ 0.95
Real Input → Output Examples
These are actual bank transaction descriptions enriched by our API in production. Click to explore.
Raw Input
AMZN MKTP US*2K4X9Y1Z0
{
"merchant_name": "Amazon Marketplace",
"category": "ecommerce",
"domain": "amazon.com",
"confidence": 0.98,
"mcc_code": "5417",
"country": "US",
"is_subscription": false
}Category Coverage
20+ merchant categories identified
Geographic Coverage
13 countries in production data
Latency & Performance
Measured from production API responses. Cached requests return in milliseconds.
Average Response Time
287ms
Across all 1,874 requests (cached + uncached)
Cache Hit Rate
71.7%
1,343 of 1,874 requests served from cache
Uncached AI Enrichment
5.2s
p50 for cold enrichment with AI + web search
Uncached Latency Percentiles
p50
5.2s
p75
6.0s
p90
7.3s
p95
8.2s
p99
12.7s
Cold enrichments use Claude AI with web search for maximum accuracy. After the first call, results are cached for 7 days and served in <50ms.
Quality Score Breakdown
Each enrichment is scored 0–1 based on field completeness.
Merchant Name
Clean brand name identified
Domain
Website domain resolved
Confidence
AI self-reported certainty
Category
Business category assigned
MCC Code
ISO 18245 code mapped
Phone
Customer service number
Support email found
Support URL
Help page URL
Categories
Multiple categories assigned
Description
Business description generated
Average quality score across production: 0.733 — Entries scoring ≥0.70 are cached for 7 days. Lower scores expire in 24 hours and get re-enriched.
Methodology
All data on this page comes from our production MongoDB database. These are real transactions from real customers — not synthetic benchmarks or cherry-picked examples.
Enrichment pipeline: Each transaction hits our cache first. On a miss, we call Claude AI (Haiku 4.5) with optional web search to identify the merchant, then cache the result for 7 days. 97% of enrichments use web search for maximum accuracy.
Confidence scores are self-reported by the AI model based on how certain it is about the identification. Scores below 0.50 typically indicate obscure local businesses or heavily abbreviated transaction codes.
Latency: The 287ms average includes cached responses (~12ms) and uncached AI calls (~5.2s). As the cache grows, average latency drops. At 71.7% cache hit rate, most requests are already fast.
Try it with your own transactions
20 free API calls. No credit card required. See the enrichment quality for yourself.