Integrations
Product-usage sources
Where the numbers behind an account's health score come from. This page is the connection side; the modelling and scoring are under External data & metrics.
Five source types
What OppFlow will refuse to connect to
A connector destination is something a workspace admin types in, so it is treated as untrusted by construction. Every REST and GraphQL dial is checked at the socket: private, loopback and cloud-metadata addresses are refused, and refused again on any redirect. SQL hosts get the same check before a connection is opened.
Credentials
Entered once on the connection, never re-typed per metric, and encrypted at the storage boundary. A connection cannot be deleted while metrics still depend on it; the error says how many.
How a refresh runs
An hourly sweep claims the metrics that are due and fans out one job per claim, rather than fetching inline. Writes are keyed by metric, account and date, so a redelivered job overwrites rather than duplicating. The query is not free to retry, though: a failure after your source has answered re-runs the whole job and re-queries it, which is worth knowing if your source bills per query.
Matching rows to accounts
Each metric names the field that identifies the account and what it is matched against — account domain, Salesforce ID, HubSpot company ID or account name. Usage belongs to the account, not the deal, and in a seeded demo workspace only genuine customers carry rows: a net-new logo has never installed the product, so a confident health score on one would be fiction.
OppFlow.ai docs · by Traversal · v0.1.32