Measurement

Numbers that mean something.

The useful ones

Acquisition cost against lifetime value, conversion at each funnel stage, and retention. Impressions and followers are inputs, not outcomes.

Find the stage with the biggest drop-off before adding more traffic.
Attribution is genuinely difficult; treat any single-touch model with suspicion.

Testing

A/B tests need enough sample for the difference to exceed noise, and stopping early when a result looks good invalidates them. Decide the sample size before starting.