Breadth vs depth in integrations
Zachary Kirby, co-founder of Vessel, on building the Vercel for integrations
The real moat in integrations is not the common API shape, it is the extraction stack underneath it. Vessel can add connectors quickly because most GTM systems expose usable public APIs, so the job becomes wiring auth, endpoints, webhooks, caching, and ETL. Plaid and HRIS connectors live in much messier systems, where scraping, file transfers, closed partnerships, permissions, and reliability work eat the engineering time before breadth can scale.
-
In Vessel's own stack, the jump from roughly 20 to nearly 100 integrations in two weeks came from a reusable framework where new connectors mostly mean adding authentication and mapping endpoints into an existing pipeline. That is a software factory problem, not a one off data access problem.
-
HRIS is harder because the market is both fragmented and closed. Finch describes about 6,000 U.S. employment systems, with many workflows still running through SFTP, CSVs, and gated partner APIs. Human Interest had built 40 to 50 connectors and still used a 100 plus person ops team to cover the rest.
-
Plaid sits on the other hard extreme. Its value came from making bank data programmable despite hostile sources, including screen scraping and bank specific workarounds. That is why Plaid became a large business, reaching an estimated $390M ARR in 2024, but it also means engineering effort goes into reliability and coverage, not just shipping dozens more connectors fast.
The market is heading toward a split architecture. Breadth first platforms will keep winning where data sources are open and standardized, while closed or heavily customized systems will reward specialists built for extraction depth and maintenance. As AI raises demand for product integrations everywhere, the winners will be the platforms that turn fragile connector work into repeatable infrastructure.