What it is
An AI confidence score is the probability an AI system attaches to one specific output: how sure it is that a detected signal is real, that a candidate matches a role, or that a company matches your ICP. It is usually expressed as a percentage or a simple high/low flag, and it comes from the model itself, not from a human judgement call. A confidence score of 95% on a funding-round detection means the underlying evidence was strong and unambiguous; 55% means the system found something worth flagging but is not sure it read it correctly.
The key word is single. A confidence score belongs to one detection, one match, one inference, not to an account or a pipeline as a whole. That is what separates it from lead scoring, which is a composite priority ranking built from several inputs, ICP fit, signal recency, touch history, response likelihood, one of which can be the confidence of the underlying detections that fed it.
A confidence score answers one question: how sure is the machine, not how important is the account.
Why it matters
Without a visible confidence marker, every AI output looks equally certain on the page, whether it is a near-certain fact or an educated guess. A consultant who cannot tell the difference either over-trusts a shaky detection and wastes a call, or under-trusts a strong one and lets a real opportunity sit unworked. Both failures cost placements.
Confidence is also one of the mechanisms that keeps a signal feed usable in the first place. A feed that surfaces every weak detection with the same weight as every strong one trains a consultant to stop trusting any of it, the failure mode known as signal fatigue. Attaching a confidence score to each detection is what lets a system separate near-certain events from plausible guesses before either reaches an inbox.
How boilr handles it
boilr attaches a confidence read to what it finds, not just a yes or no. A signal built on a strong public footprint, an active GitHub presence, a well-documented executive move, is flagged high-confidence. A signal inferred from thinner evidence, say a likely hiring need read from a product launch and a company's current team structure, is labelled as probabilistic in the task itself, so you know it is an educated inference rather than a confirmed fact before you act on it.
Confidence never replaces your judgement, and it is not the whole story either. It feeds into lead scoring as one input alongside ICP fit, signal recency and touch history, and you can raise the bar yourself, for example tightening the threshold on signal clusters so only the highest-confidence combinations reach your task inbox. As outcomes flow back into the Company Brain, the system learns which of its own high-confidence calls actually convert, and sharpens accordingly.
