Documentation menu

Unexpected metric spikes or drops

A metric moved more than you expected between two days. Before treating it as a platform event, rule out the three common data-shape causes.

Before you start

  • Know the exact date range you are comparing.
  • Know which metric moved. CPI, ROAS, and custom-event CPA behave differently under the same conditions.

1. Cohort maturity pulled a number

The most recent 7 days of Revenue D7 are excluded by design, 14 for Revenue D14, and 30 for Revenue D30. If your comparison window crosses the boundary where yesterday's data is still maturing, today's ROAS number will include more complete cohorts than yesterday's did, which can look like a drop. See Revenue and cohorts for why. This is the single most common cause of surprising ROAS shifts.

2. A recent keyword add is in its learning phase

New keywords spend early at a CPT that reflects Apple's match-quality priors rather than the bidder's long-run target. For the first few days, their CPA and ROAS will sit outside the campaign's target. If a single ad group added dozens of new keywords this week, expect campaign-level metrics to drift until the new keywords either converge or get paused.

3. Data freshness lag

Apple Ads spend refreshes roughly every hour. MMP installs and revenue arrive with a longer lag, typically 24 hours to a few days depending on provider and event type. If you check a metric in the first few hours after midnight local time, the denominator may be fresh but the numerator is not yet complete, or vice versa. Give the window a full day before drawing conclusions. See Data freshness.

4. An Apple platform change

Apple periodically changes search placement, match-type behaviour, or auction dynamics. These shifts can move CPI or impression share across an entire portfolio in a single day. If a move affects most campaigns in the same direction on the same day, a platform change is likely — check Apple's announcements before assuming a campaign-level cause.

5. A real change in the auction

Competitor entry, seasonality, a holiday, or a price change on your app all move the auction. These are real movements in the market, not artefacts. Cross-reference with your own commercial calendar.

When to contact support

If the spike cannot be explained by cohort maturity, new keywords, freshness lag, platform behaviour, or calendar events, and it persists across at least two full days, export the affected campaign and date range and contact your Catchbase representative.

When to run an incrementality test

If a change in spend appears to drive a change in installs but you are not sure whether the relationship is causal, run an incrementality test. Attribution movements are not the same as causal movements.

Related