The exact formulas behind every number on your dashboard

1. Waste prevented

A pre-empted overspend that would have happened in the absence of an ADmetric rule. We only count an event as "waste prevented" if all four conditions are true:

• An automation rule fired (logged in automations.events).
• The rule paused or reduced spend on an ad set, campaign or account.
• A back-test against the prior 28-day spend curve shows the would-be spend exceeded the new cap.
• The platform confirms the change took effect (acknowledged write in the audit log).

The dollar figure is (projected_spend − actual_spend) for the remainder of the day on which the rule fired, capped at the campaign's planned daily budget. We never extrapolate forward more than 24 hours.

2. Funds recovered

Money that was already committed to one platform and re-routed to another within the same flight, when the source channel's marginal CPA crossed a customer-defined ceiling. Counted at the smaller of (a) the amount actually moved, or (b) the delta between the source and destination CPA × units the destination produced.

3. Forecast confidence bands (P10 / P50 / P90)

The cumulative spend forecast is the median of 2,000 Monte Carlo runs of a mean-reverting random walk fitted to the last 28 days of daily spend, conditioned on the campaign's stated cap and on auction-side volatility derived from CPM variance. The shaded band is the 10th and 90th percentiles of those runs — not a standard error.

If the model has fewer than 14 days of history for a given account, the band widens by ±25% and we mark the chart "low confidence".

4. Anomaly detection

We flag a metric as anomalous when its rolling 6-hour mean drifts more than 2.5σ from a 14-day baseline of the same metric, same channel and same day-of-week. False-positive rate, measured against a 90-day labelled sample, is 3.1%.

High-severity anomalies require a human approver in Monitoring mode; in Active mode, they trigger the autopilot rule and notify the assigned owner within 60 seconds.

5. Payback period

Months until cumulative gross profit from a monthly customer cohort equals the fully-loaded CAC for that cohort. Gross profit uses the customer's submitted margin rate (defaults to 65% for DTC, 80% for SaaS). We do not net out retention costs — payback measures the acquisition channel, not the lifecycle.

6. Blended ROAS / MER

MER = total_revenue_in_window / total_marketing_spend_in_window. We use the customer's invoiced revenue when available; otherwise the GA4 / Shopify / Stripe stream they connected. We do not adjust for tax, refunds or returns unless the customer maps a refund stream.

7. Savings ledger entries

Every entry in the savings ledger is hash-chained to the previous one. The full chain can be verified offline against our published verifier (/audit in your dashboard). Once written, an entry cannot be edited or deleted — corrections appear as a new entry referencing the original.

8. What we deliberately do not measure

We do not estimate brand lift, modeled attribution or "view-through" conversions. Those numbers belong to your MMM partner. We measure what is verifiable from platform APIs and from your own revenue stream.