Core concepts: signals, decisions, and confidence

Four components produce every recommendation you see. Understanding what each one does makes the reasoning trail much easier to read.

Lens Intelligence

Normalises data from every connected platform onto one standard: one definition of spend, one definition of a conversion, one currency, one timezone. Without this step, cross-channel comparison is arithmetic on incompatible numbers.

Insight Graph

Joins entities across systems — campaigns, creatives, sessions, contacts, deals — so a closed-won deal can be traced back to the campaign that started it.

DecisionCore

Scores candidate actions on expected business impact, statistical confidence, and time-to-signal, then ranks them. This is what makes the queue an ordered list instead of a pile of alerts.

Action Engine

Turns a scored decision into a concrete instruction: which campaign, what change, how much, and what to expect. It also tracks the result after you approve, which feeds back into future confidence scores.

How to read a confidence score

• High (80–100): the pattern holds across the full lookback window and enough conversions support it.
• Medium (55–79): the pattern is real but the sample is thin or recent. Consider a smaller test.
• Low (below 55): shown for transparency only, and never placed at the top of the queue.