Meta Ads · Level 4
Measurement & Delivery
Connect Meta measurement, attribution, learning stability, budget capacity and audience structure so delivery decisions are based on usable signals rather than platform folklore.
Measurement starts with the event the system is learning from.
Northstar has reliable purchase events from Pixel and Conversions API. Add-to-cart and view-content events are useful diagnostics, but they should not displace purchase just because they occur more often.
Captures browser-side events when available.
Can strengthen event delivery and resilience when implemented correctly.
The commercial outcome the campaign is meant to acquire.
Measurement quality is both technical and strategic: the event must arrive reliably and represent the outcome you actually value.
Decision check
Purchase tracking is reliable and volume is healthy. What should remain the primary optimization event?
Attribution is a measurement lens, not an objective truth machine.
Meta-reported conversions help explain platform contribution, but business decisions should also consider analytics, backend revenue, incrementality and the buying cycle.
Useful for optimization and platform-level reporting.
Helps validate whether reported performance matches actual commercial outcomes.
Use multiple sources when the decision is material and attribution is uncertain.
A discrepancy is a diagnosis prompt, not automatic proof that one system is “wrong.”
Decision check
Meta reports strong ROAS but backend revenue does not move with spend. What is the better response?
Give the delivery system room to learn without turning stability into superstition.
Frequent structural changes can make performance harder to interpret. But “do nothing because learning” is not a strategy either; serious tracking, budget or creative problems still need intervention.
Avoid changing several controls every few hours.
Broken tracking, bad offers or unsustainable spend should not be protected by fear of resets.
Clean change history makes performance shifts easier to explain.
Stability is useful when the setup is sound. It is not an excuse to preserve a bad setup.
Decision check
A campaign is spending normally but the purchase event stopped firing yesterday. What should you prioritize?
Budget increases should follow capacity, not excitement.
Northstar has a £28 target CPA. A campaign at £22 CPA with stable volume may have room to scale, while a campaign already at £36 CPA needs diagnosis before more budget.
More budget can make sense when performance is strong and there is evidence of delivery headroom.
More spend can simply buy more expensive conversions.
A great CPA on two purchases is not the same as a stable acquisition system.
Scaling is a hypothesis about additional spend, not a reward for a pretty historical ROAS.
Decision check
Prospecting is at £22 CPA on meaningful volume and repeatedly budget constrained. What is the sensible next step?
Separate audiences only when the separation creates useful control or learning.
Broad prospecting, customer exclusions and retargeting can serve different jobs. Splitting similar prospecting audiences into many small ad sets can reduce signal density without creating a meaningful experiment.
Useful when creative and conversion signals can guide delivery.
Separate only when you genuinely need to compare strategies.
May deserve separate control when volume and message justify it.
Audience structure should answer a question or protect a business rule.
Decision check
Three prospecting ad sets have heavy overlap and no distinct hypothesis. What is the cleanest direction?
Put the measurement and delivery principles together.
Choose the action that protects business-signal quality, interpretability and sustainable delivery.
Separate scaling pressure from creative or audience deterioration.
Investigate attribution before assuming incremental growth.
Check audience saturation and message exhaustion.
The best Meta optimizer does not react fastest. They isolate the reason performance changed.
Decision check