An e-commerce outage rarely starts with a customer’s phone call. It starts quietly: payments fail on Friday after working hours, the feed breaks after a price change, tracking dies after a plugin update. The trace of each one waits in the daily figures the next morning; the trouble is that usually nobody looks there. This guide is a catalogue of failures and their marks in the figures, rules for alert thresholds, a morning data check and the automation that does this check for you.
Catalogue
E-commerce failures and their marks in the data
Every important failure leaves a characteristic trace in the figures, usually long before the first customer complaint:
| Failure | First trace in the data | Without monitoring, it comes to light… |
|---|---|---|
| Payments down | jump in the share of unpaid orders above the norm | after customer calls, often after a whole weekend |
| Broken product feed | product impressions and clicks in campaigns drop towards zero | with the invoice for a month without Shopping sales |
| Broken tracking | GA4 coverage drops while shop orders stay stable | when someone notices “strange reports” weeks later |
| Bestseller out of stock | category conversion drops, product clicks without sales | after budget burnt on advertising a phantom product |
| Price import error | products at €0 or a spike/collapse in average order value | after a wave of orders you do not want to fulfil |
| Checkout after an update | conversion drops on one device type (usually mobile) | never directly; it looks like “weaker traffic” |
Shop pulse
What an outage looks like on the daily pulse
The simplest monitor is paid orders day by day, set against the norm for their day of the week. A payment failure that took down half a Saturday in a pet shop (illustrative data) looks like this:
Key details: a Saturday is compared with Saturdays (against the whole week, that day would still look decent), and on the day of the outage traffic and baskets were normal; the last step failed. That is why the pulse is built on paid orders, with the share of unpaid orders kept alongside as a sensor: when a day ends with several tens of per cent unpaid against a norm of a few, it is almost certainly a technical issue, not the customers.
Thresholds
Alert thresholds: when a deviation is an alarm
Working definition
An anomaly is a day’s deviation from the metric’s pattern for its day of the week, rather than the difference from yesterday. Saturday has a different norm from Monday; the alarm compares a Saturday with Saturdays.
A two-tier approach prevents a flood of false alarms: a hard alert is a day clearly outside the typical range for its weekday, while a soft deviation is confirmed by a second day instead of triggering an alert after the first. A separate class covers binary signals, which alert immediately at the next refresh: products priced at €0, zero impressions for the whole feed, zero purchase events despite non-zero sales. The third rule: every alert needs an owner and a first action (“spike in unpaid orders → check the payment provider’s dashboard”), otherwise nobody reads them after two weeks.
Routine
Morning data check in five points
Before automation takes over the job (and alongside it), a simple check does the work:
Share of unpaid orders
Yesterday compared with the norm for its weekday. A spike above the norm is the first payment warning sign; checking it takes less time than making a coffee.
GA4 coverage
Orders in analytics versus paid orders in the shop. A drop with stable sales means a measurement failure; the guide to the GA4 audit explains how to diagnose it.
Product impressions and clicks
Is the feed alive? A sudden fall in impressions in product campaigns almost always means disapprovals or expirations in Merchant Center.
Bestseller stock levels
Campaign products falling below the days-of-stock threshold drop out of ads before they start burning budget (mechanics: the guide to stock levels).
Yesterday’s spend vs plan
A campaign that spent double or nothing at all is also an anomaly; limits and budgets tend to “come unstuck” after account changes.
Scope of the topic
Monitoring catches sudden deviations; it does not explain trends. When figures slide for weeks without a single event, that is a job for change decomposition, described in the guide “Sales dropped: how to find the cause”. Monitoring says “something has just broken”; decomposition says “what, and by how much”.
DataOrganizer
How DataOrganizer does the check for you
DataOrganizer works on daily aggregates refreshed once a day: it compares each closed day with the metric’s seasonal pattern, automatic anomaly detection reports deviations, and data quality control flags gaps and inconsistencies between sources. There is no minute-by-minute view, and that is not the point: the point is that no broken day goes unnoticed. The morning check then comes down to a single question:
Three deviations from the pattern, two need action:
Methodological note: figures in the examples are illustrative. Norms and thresholds belong to each shop (different for seasonal categories, different for stable ones), and loss estimates for the failure window are calculated from the norm for that time of the week, not from the annual average.
Key takeaways
The cost of a failure grows with every day nobody knows about it. A catalogue of fingerprints, weekday norms, a five-point check and automatic anomalies: this set turns “we found out two weeks later, from the invoice” into “the alert was waiting in the morning after the refresh”; a problem caught the next day costs a fraction of one discovered at the end of the month.
FAQ
Frequently asked questions about shop data monitoring
Which failures show up in the data?
All the important ones: payments (spike in unpaid orders), feed (fall in product impressions), tracking (drop in GA4 coverage with stable sales), stock (category conversion down), prices (products at €0, odd basket values), checkout after an update (conversion down on one device type). Each leaves its fingerprint in daily figures, usually long before anyone calculates the losses.
How to set thresholds without drowning in alerts?
A norm per weekday (Saturday compared with Saturdays). Hard alert: a day clearly outside its norm’s range; a soft deviation is confirmed by a second day. Binary signals (€0, empty feed) alert immediately at the next refresh. Every alert has an owner and a first action.
How does an anomaly differ from a sales decline?
An anomaly is sudden (single days) and usually technical; a decline is a trend over weeks with a business cause. Monitoring catches the first, change decomposition breaks down the second. Confusing the two modes ends with looking for a strategy problem where a plugin has failed.
What should be checked daily?
Five figures: unpaid %, GA4 coverage, product impressions in campaigns, bestseller stock, yesterday’s spend vs plan. This check catches the vast majority of technical failures before they grow into a week-long problem.
How does DataOrganizer detect anomalies?
On daily aggregates refreshed once a day (no hourly view): each closed day compared with the metrics’ seasonal pattern, data quality and consistency control, and a chat summary of deviations that separates likely failures from explainable peaks (e.g. a newsletter send).
Cheat sheet for the whole guide
Shop monitoring in three layers
1. Know the failure fingerprints. Payments, feed, tracking, stock, prices, checkout: each failure leaves a characteristic trace in the figures and has its own first corrective action.
2. Alert against the pattern, not against yesterday. Use norms per day of the week, two conditions at once (out of range + persisting), and trigger binary signals immediately. An alert without a recipient and a first action dies in the inbox.
3. Automate the round. Five numbers checked every morning catch most problems; DataOrganizer does this round on its own, comparing each closed day with the pattern and reporting anomalies at the morning refresh, before they grow into a week’s worth of losses.
DataOrganizer · anomalies report themselves
Connect your shop, analytics and ad platforms: daily refreshes, seasonal metric patterns and automatic deviation detection. Demo account available instantly; you only pay for connected sources.