How it works
How OutboundFix knows what it knows.
A verdict is trustworthy because of how it was reached, not because of who asserts it. This is the decision lifecycle, step by step, including what OutboundFix does not do.
Decision lifecycle
From a domain to a recorded decision
You submit company domains. Each domain is evaluated on its own.
Your targeting policy states which dimensions are required, what disqualifies an account and what counts as a positive signal. It is versioned. Until you define one, OutboundFix's default targeting policy applies.
OutboundFix reads the account's own official website: the home, about, contact and privacy pages, plus a few clearly relevant company, product and customer pages, within a fixed request limit. Job boards, news sites and third-party databases are not read today.
An AI model extracts structured observations and is instructed to look for counter-evidence as well as support. A claim has to be backed by text that is actually on a fetched page. If the extraction fails validation in any way, it is discarded whole.
A deterministic runtime sets the observations against the policy and issues PASS, RESEARCH_MORE or KILL. The model never chooses the verdict.
The verdict is written as a Decision Record with its evidence, sources, reason codes and policy version. A record is never edited afterwards.
New evidence or a new policy version produces a new record that supersedes the old one. The old record stays in the history.
Why RESEARCH_MORE exists
Not every account resolves cleanly
A system that only ever outputs PASS or KILL is hiding its own uncertainty. OutboundFix does not force a verdict where the evidence does not support one. RESEARCH_MORE means exactly what it says: hold this account, get the specific missing evidence, and do not spend sales capacity on a guess in either direction.
Fail closed
When research fails, nothing is guessed
If a site cannot be read, the model is unavailable or its output does not pass validation, OutboundFix does not fill the gap. When nothing can be retrieved from the site, no verdict is written and the failure is reported. When the site was read but the evidence could not be extracted, the account is held at RESEARCH_MORE under the default policy, and under your own policy no verdict is written and the failure is reported. Either way it never becomes a PASS or a KILL.
Signal quality & provenance
Every claim traces back to a source
Each observation keeps the page it was read from, so a verdict can be checked against what produced it rather than taken on faith. Where an account’s own site says little, that limit is stated in the record instead of being papered over with an inferred score.
Integrations
Integrations bring evidence, never verdicts
An integration or a provider can add evidence and context to a decision. It cannot supply the verdict, override your policy or widen what OutboundFix may do. What an integration sends is kept as evidence or context with its source, and the verdict still comes from the deterministic runtime under your policy. Integrations with HubSpot and Clay are built and not yet active.