Understanding technology signals at scale

We get asked how anVendor actually detects what a company uses. Here's the short version of how the pipeline works — without the parts that would make it easy to fool.

Sources

We pull from a wide set of publicly observable signals. Some are obvious — site headers, page assets, DNS records, public job listings. Others are less obvious — CDN footprints, third-party script attribution, security headers, public commit metadata. No single source is reliable on its own. Most signals are noisy in the same direction: they over-report.

Verification

Every candidate signal is cross-checked against at least two independent sources before it surfaces in a result. If a tool appears once in a job post but nowhere on the site or in the headers, the signal is held back. If it appears in three independent places, it ships with high confidence.

Confidence scoring

Each signal carries a probability. A confirmed page-asset hit on a CDN-served script is near-certain. A single mention in a careers page is a hint. We expose that confidence in the API and in the export — and we never promote a low-confidence signal into the headline categories.

The hard part

The hard part isn't detection. It's de-duplication. A company that's been acquired, rebranded, or merged often shows three overlapping stacks. A tool that was rebranded last quarter shows up under both old and new names. The pipeline has to know that Segment and Twilio Segment are the same thing — and that Heap and Mixpanel are not.

Why this matters

Most technographic data on the market is built on a single source — a crawl, a vendor list, a leaked dataset. anVendor is built on a verification model. It's slower to add new tools, but the signals customers act on are signals they can trust. That trade-off was the most important call we made in year one.