How Vettique Went From AI Inference to Seven Evidence Sources Per Report, Cut a False "Do Not Wire" Score by 71 Percent, and Learned to Say "No Data" Instead of "High Risk"
Vettique tells Amazon and Shopify sellers whether to wire a deposit to an overseas supplier. When Epiphany Dynamics arrived, the 0 to 100 risk score came from a language model and nothing else. We wired in seven external evidence sources, then rebuilt the scoring so a factory with 7,514 real US customs shipments could no longer be scored 95/100 Do Not Wire because a directory listing was missing.
False Do Not Wire calls cut 71 percent
-
01 US customs shipment records
Verified live against a major retailer: 568,506 shipments.
-
02 Alibaba supplier profile signals
Gold Supplier years, Trade Assurance, on-site verification.
-
03 Category pricing table
Seeded across 20 product categories, editable by the client.
-
04 Google business-listing verification
Confirms the supplier exists as a real, listed business.
-
05 WHOIS domain age
Flags freshly registered supplier domains.
-
06 ScamAdviser and Google Safe Browsing
Known-scam and unsafe-site checks on the supplier domain.
-
07 Web and LinkedIn presence
Custom search for a verifiable footprint beyond the storefront.
A scoring model with no evidence has two failure modes: it misses real risk, and it invents risk where there is only missing data. The second is quieter and, for a seller who walks away from a legitimate factory, just as expensive.
We rebuilt the scoring so data gaps live in their own labeled section and cannot drive the number. A verified manufacturer with 7,514 real shipments dropped from 95/100 AVOID to 40/100 CAUTION; another with 1,367 shipments went from 85 AVOID to 25 PROCEED.
A per-source status panel under each report shows the seller which of the 7 checks ran, which were blocked, and which found nothing, so "no data" is never shown as high risk.
We told him the ceiling
Ziad asked what hit rate to expect on real factories. The honest answer was about 80 percent, with the gap owned by anti-bot walls on one source. We named the proxy layer that would push it to 90 to 95 percent, what it costs per thousand requests, and that it belonged in a follow-on rather than the current phase. A vetting tool that overstates its coverage is worse than none.
Vettique
- Built
- Seven-source supplier risk engine and honest scoring
- Sector
- Data Integration
- Site
- vettique.com
Vettique launched in spring 2026 with a blunt tagline: "Vet your suppliers before you wire." Every year thousands of Amazon and Shopify sellers wire deposits to suppliers they have never met and never hear back. The product returns a 0 to 100 composite risk score, red, amber, or green flags, and a verdict of Proceed, Proceed with Caution, or Do Not Wire, across suppliers in China, India, Vietnam, Bangladesh, Turkey, and Mexico. Free plan: three checks a month. Pro: $19 a month, unlimited.
"Patrick is an excellent developer and a genuinely trustworthy partner. He delivered a complex data integration milestone involving 7 external sources, built a clean pipeline architecture, and was honest about technical limitations rather than overpromising. He communicates proactively, fixes issues swiftly, and I'm continuing to work with him on a second project. Highly recommended."
Ziad Ayad
Founder, Vettique
- Why not just improve the AI prompt?
- Because a prompt cannot know a supplier's customs history. Seven external sources put evidence behind the score.
- What does "no data" mean on a report?
- A check ran and found nothing, or was blocked. It is shown as a gap and does not raise the risk score.
- What is the coverage on real factories?
- About 80 percent in customs data; the ceiling and the path to 90 to 95 percent were both stated to the client.
Have a Product That Makes Decisions on Data You Are Not Sure Of?
Get an audit of what the data actually says.
Free 30-minute audit. No obligation, no pitch. Or take the 2-minute AI quiz.