An error can be real, repeatable, and worth investigating without giving us enough evidence to say that it cost a merchant revenue. Monitoring products blur this distinction too easily. A technical signal appears, a purchase did not happen, and the interface quietly turns correlation into a commercial claim.

Dozenfold treats those as separate questions.

First: did something technical fail?

This is the capture layer. It includes JavaScript exceptions, failed requests, GraphQL and cart errors, broken resources, performance degradation, and privacy-safe friction signals. At this point we know what the browser observed. We do not yet know what it changed for the shopper or the business.

Some signals should remain technical evidence only. A request that received no browser response, a report-only security-policy event, or a non-critical image failure may help a developer debug the storefront, but none of them should automatically enter a revenue calculation.

Next: was a shopper journey exposed?

To evaluate shopper impact, the issue must be placed in a meaningful journey stage. We then need eligible anonymous sessions that reached that stage, encountered the issue before purchase, and can be compared with similar sessions that did not encounter it.

Until that comparison exists, the honest state is impact not established or building a comparison—not zero impact, and not a guessed loss.

Only then: is there measurable commercial association?

When both groups are large enough, Dozenfold compares purchase completion across comparable journey stages and reports the gap with a 95% interval. A revenue estimate also requires usable order-value and currency evidence.

The resulting number is an observed association inside a stated evidence window. It is not booked revenue loss, and it does not claim that the issue caused every missed order.

These are distinct evidence states:

  1. Technical signal only — captured evidence, but not eligible for a commercial claim.
  2. Impact not established — the shopper relationship is not yet classifiable.
  3. Building comparison — eligible, but the issue or control cohort is still too small.
  4. Directional — the gap is positive, but its 95% interval still includes zero.
  5. Measured / no conversion gap — the interval is entirely above zero, or affected sessions converted at least as often as clean ones.

Uncertain evidence stays labelled

A positive gap with a wide interval is still worth knowing about. Dozenfold shows it as a directional signal, with the same comparison and order values behind it, and keeps it distinct from a measured loss. Strata where affected shoppers converted better count against the gap instead of being dropped.

The methodology explains the comparison, the interval and the labels. Amounts describe the selected period; short observation windows are not multiplied into monthly or annual forecasts.

Performance evidence needs the same discipline

Slow scripts, heavy resources, LCP, INP, and CLS can explain a weak experience. They still should not become revenue claims by themselves. We first show the performance regression and the segments where it concentrates. Commercial impact appears only when the affected journeys and comparable outcomes support it.

That separation makes the product less dramatic, but more useful. Merchants can prioritize what is measurably important. Developers can still investigate technical failures early. Neither group has to reverse-engineer whether a confident-looking number was actually supported by the evidence.

For the calculation boundary, read the revenue impact methodology.