Understand revenue impact
Dozenfold estimates how many purchases an issue costs per day, and what those purchases are worth. It compares sessions that hit the issue with comparable sessions that did not, and shows how certain that comparison is. The estimate helps you decide which fix comes first. It is not booked revenue loss, and a useful technical finding does not need a price tag.
A week of evidence behind every window
A single day rarely holds enough shoppers for a fair comparison. The evidence window is the selected window or the last seven days, whichever is longer. Pick the last 24 hours and the estimate still compares seven days of sessions; pick 30 days and it compares 30.
Because the evidence window can be longer than the selected one, amounts are shown per day: the total for the evidence window divided by the days on which the issue actually occurred. An issue that appeared on 3 of the last 7 days is divided by 3, not 7, so a new issue is not diluted by the days before it existed.
Compare like journeys
Every session that reaches a journey stage (product page, cart, checkout steps) is an entry at that stage. Entries are compared within the same stratum: the same journey stage, device, environment and sampling rate, inside the evidence window.
- Affected entries hit the issue within 30 minutes of entering the stage.
- Clean entries come from sessions with no occurrence of this issue at all.
- Sessions that had the issue but not in that window are left out of both groups.
The calculation requires a verified link between the browser SDK and the Shopify Pixel session, verified event timing and a known, uniform sampling rate. Entries without them are excluded and stay visible as coverage: the issue detail shows how many affected sessions were compared.
Give both groups the same follow-up
A purchase counts if it happens within 30 minutes of entering the stage, for affected and clean entries alike. An error that happened after the purchase still counts as exposure. Requiring the error to come first would quietly favour the affected group, because it would only keep shoppers who stayed long enough to see it. Counting both sides the same way can only make the gap smaller, never larger.
Recent entries wait until their 30-minute window is complete. Session evidence is recomputed about once a minute, so the newest activity can be a few minutes behind.
Pool the gap and show its certainty
Within each stratum, Dozenfold compares the clean and affected purchase rates. It then pools those differences into one conversion gap (a Mantel–Haenszel risk difference). Strata where affected sessions converted better pull the gap down; they are not ignored. Each pooled gap comes with a 95% interval.
| Label | Meaning |
|---|---|
| Measured | The whole 95% interval is above zero. Affected sessions convert less. |
| Directional | The gap is positive, but the interval still includes zero. |
| No conversion gap | Affected sessions converted at least as often as clean sessions. |
| Collecting | Fewer than 30 affected or 100 clean entries can be compared so far. |
Put a value on it
Missed purchases are the gap times the compared affected entries. Their value uses the clean average order value: complete order values from clean sessions, in the store’s most common order currency. At least 20 such orders are required; otherwise the gap stays visible without an amount.
For example, 480 compared affected entries purchased 10% of the time, while comparable clean entries purchased 34% of the time. The pooled gap is 24.0 points with a 95% interval of 21.2–26.8 points, so the loss is measured. That is about 115.2 missed purchases. At a clean average order value of $159.90, the estimate is $18,420 for a 30-day evidence window. The issue occurred on all 30 days, so the issue shows $614/day.
Amounts describe observed sessions. They are not expanded to traffic that was not captured.
When the store sends no order values, a measured or directional purchase gap reads as missed orders per day instead, for example ≈ 10 orders/day.
When purchases are too few: next-stage progression
Many stores see too few purchases in a week for a purchase comparison. Each issue therefore also carries a second comparison: whether a session reached a later journey stage (product to cart, cart to checkout, and so on) within the same 30-minute follow-up. It uses the same strata, pooling and 95% interval. Progression is far more frequent than purchase, so it reaches a clear answer with far fewer sessions.
Progression is shown only while purchases are too few to compare. It reads as shoppers lost per day, for example ≈ 15 fewer continue/day, and never carries money: it is not a revenue estimate.
How issues are ranked
The issue list and the Overview attention queue order issues by the strength of their evidence, not by the largest amount:
- Measured purchase loss, by missed orders per day at the lower end of the 95% interval.
- Measured progression loss, the same way.
- Directional signals, by affected sessions. A noisy amount never lifts an issue up the list.
- Everything else, by activity.
Using the lower end of the interval means a small, uncertain sample cannot outrank a smaller but well-measured loss.
No monthly or annual forecast
Dozenfold does not multiply a short window into a monthly or annual loss; the per-day figure is an average over days the issue was observed, not a forecast. Changing from 7 to 30 days changes the compared sessions, so the estimate can rise or fall. Separate issues can affect the same sessions, so their amounts must not be added into a store total.
Know the limits
- Association is not causation. This is a comparison, not a controlled experiment: traffic mix, campaigns, releases and shopper intent can still contribute to a gap.
- A directional result is a reason to keep watching, not a confirmed loss.
- No conversion gap does not mean an issue is harmless; it can still hurt experience or other metrics.
- Occurrences, affected sessions, regressions and experience findings stay useful when money is unavailable.
Read Revenue impact is unavailable for the states shown while evidence builds up.