After the first big day: reading the numbers correctly

The meeting that happens on the Saturday morning usually tries to do both at once, which is why it produces decisions people regret. It helps to name them separately.

The operational read asks what went wrong and what can still be fixed. It has a deadline, because the weekend is still trading and the answers expire on Monday. It is mostly about failure: things that errored, things customers could not find, things that sold out while demand was still arriving.

The commercial read asks whether it worked. Which channel earned its budget, what the promotion cost in margin, whether these customers come back. It has no deadline at all, and every one of its inputs is still incomplete on the Saturday. Attempting it now produces confident answers that will be wrong by January, and they will be repeated in meetings long after the data has been corrected.

Do the first one before breakfast. Book the second one for a date well after the returns window closes.

A decline is ambiguous — fewer people wanted the thing, or fewer people came, or they came at a different hour. A failure is not. Somebody tried to do something on your store and could not, and that is both unambiguous and often fixable in twenty minutes.

The list worth going through, in this order: error rates on the add-to-cart, checkout-start and payment paths compared with the same hours a week earlier; checkout completion split by payment method, because a single provider degrading looks like a general conversion dip until you split it; zero-result searches, which on a big day are a direct list of what customers expected you to be selling; sold-out products that were still receiving traffic, which tells you what to substitute in merchandising today; and any 404s that spiked, usually from a campaign link pointing somewhere that no longer exists.

Each of those has an action available before the weekend ends. That is the test for whether something belongs in the Saturday read: if there is no action, it is not urgent, and it will read more accurately in December anyway.

Friday's total includes orders that will be refunded, orders that will be cancelled because they cannot be fulfilled, and orders sitting on a fraud hold that will not all clear. On a discount-heavy day the return rate is usually higher than your annual average, not lower, and gifting pushes the eventual refunds further out than normal.

Before anyone circulates a figure, find out one mechanical detail about your own reporting: when a refund is processed, is it recorded against the date of the original order or the date of the refund? Both conventions exist across the tools a merchant uses, and they produce very different-looking Novembers — one where peak revenue quietly deflates over the following months, and one where December carries the cost of November's returns. Check yours, write it down next to the number, and be consistent about it year on year.

The other deduction people forget on the day is discount cost. Gross revenue up and gross margin pounds down is a plausible outcome of a heavy promotion, and it is not visible in any figure that has the word "sales" in it. If you can produce only one commercial number on the Saturday, make it margin after discount, not revenue.

Last year's Black Friday fell on a different date. Comparing to the same calendar date and comparing to the equivalent trading day give different answers, and both are legitimate — the problem is a chart that does not say which it used, which is most charts.

Weekday alignment usually reads better for trading, since shopping behaviour tracks the day of the week and the position relative to payday more than the date. Date alignment reads better for finance, which cares about the month. Pick one for the operational read, state it on the slide, and keep the other available for the people who need it.

The same discipline applies to campaign timing. If last year the sale opened on the Wednesday and this year on the Monday, a day-for-day comparison is measuring your campaign calendar rather than your performance. Compare the whole promotional window against the whole promotional window, and treat any single-day headline as a curiosity.

Conversion rate is a fraction whose denominator you have deliberately distorted. Peak traffic carries more paid, more first-time and more browsing-only visitors, and each of those moves the rate without telling you anything about the store. A conversion rate that fell while orders rose is the normal shape of a successful campaign, and every year somebody treats it as a defect.

The ratios that survive a traffic mix shift are the ones measured further down the funnel, where intent is already established. Checkout completion rate — of the sessions that reached checkout, how many finished — is the most useful single number on the Saturday, because it barely moves with who turned up and it moves sharply when something is broken. Add-to-cart to checkout-start is the next.

Split both by device before drawing any conclusion. A drop on mobile alone is a site problem you can act on; a drop across both usually means the traffic changed, which is information for the paid team rather than for engineering.

Declaring a channel winner. Peak is the period when attribution is least trustworthy — more paid traffic, more cross-device journeys, more shifting in the consented proportion of your audience — and the first report you open on the Saturday is precisely the one that will be revised. Nothing about the budget decision it would inform has to happen this weekend.

Running an A/B test through the weekend and believing the result. The traffic is not the traffic you will have in February: different intent, different discount exposure, different device split. A variant that wins in peak conditions has been measured on a population you will not see again for eleven months, and rolling it out permanently on that basis is a decision built on the least representative sample of the year.

And killing a product line, a category or a page on four days of data. Peak sells what was discounted and what was promoted; it is a poor referendum on anything else. Write the observation down, put it on the list for a proper read in the new year, and go and get some sleep — the Sunday and the Monday are still to trade, and being rested for them is worth more than the analysis.

• Shopify conversion rate optimisation
• Post-peak: what your data is telling you in December
• The Black Friday runbook